[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/h/ - Hentai


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: higger.png (1.84 MB, 1248x2019)
1.84 MB PNG
Previous Thread: >>>>8865040

>LOCAL UI
forge neo (ANIMA SUPPORT): https://github.com/Haoming02/sd-webui-forge-classic/tree/neo
reForge: https://github.com/Panchovix/stable-diffusion-webui-reForge | Stable commit: 6964ceb
Comfy: https://github.com/comfyanonymous/ComfyUI | https://comfyanonymous.github.io/ComfyUI_examples

>NOOBAI-XL
https://civitai.com/models/833294/noobai-xl-nai-xl
https://huggingface.co/Laxhar

>NOOBAI SHITMIXES
102d custom: https://civitai.com/models/1201815?modelVersionId=1491533
291h: https://civitai.com/models/1301670/291h

>RESOURCES
4chanX Catbox/NAI prompt userscript: https://rentry.org/hdgcb
Easy booru tag export userscript: https://github.com/Takenoko3333/Danbooru-Tags-Sort-Exporter
Tags: https://danbooru.donmai.us/wiki_pages/tag_groups | https://danbooru.donmai.us/related_tag | https://tagexplorer.github.io/#/
Inpaint: https://files.catbox.moe/fbzsxb.jpg | https://huggingface.co/Wenaka/NoobAI_XL_Inpainting_ControlNet_Full
Upscalers: https://openmodeldb.info
Booru: https://aibooru.online
ControlNet: https://rentry.org/dummycontrolnet (OLD) | https://civitai.com/models/136070 (look at the green links in the description)

>TRAINING
Guide: https://rentry.org/yahbgtr (WIP)
Anon's scripts: https://mega.nz/folder/VxYFhAYb#FQZn8iz_SxWV3x1BBaJGbw
Trainers: https://github.com/derrian-distro/LoRA_Easy_Training_Scripts | https://github.com/bmaltais/kohya_ss | https://github.com/Nerogar/OneTrainer

>NEWS
https://huggingface.co/circlestone-labs/Anima

OP Template/Logo: https://rentry.org/hggop/edit | https://files.catbox.moe/om5a99.png
>>
File: 61370221.jpg (176 KB, 832x1216)
176 KB JPG
>>
>>8870997
>>8871002
>>8871010
Beta I agree with. SGM Uniform does a lot better. ER SDE is alright thobeit. Keeps the slop at bay.
>>
https://files.catbox.moe/ywv3gg.png
>>
otoko no doko?
>>
glad she got her bottom surgery n.n
>>
his balls looks weird now
>>
File: 00002-4030598583.png (2.07 MB, 1248x1824)
2.07 MB PNG
fucking extra fat wrinkle fixed
>>8871145
no
>>
she (formerly he) is suffocating him with that pillow
>>
>>8871161
that's his (my) fetish
>>
was pony the most influential local model of all time?
>first model to break free from the nai leak
>first local model to tackle SDXL, a task previously thought to be impossible due to censorship
>introduced score-based tags which are still used in anima today
>pioneered the idea of combining multiple datasets beyond just danbooru, a technique also used by noobai
>capable of both anime and realism
>>
how many times have you used this bait by now
>>
how could you forget its proficiency for such concepts as GPO and CES?
>>
>>8871168
no pony bad and we actually never ever used it
>>
my best memory of pony is deleting everything related to it as soon as illu 0.1 dropped. truly the end of dark times
>>
>>8871178
based and same
>>
File: ComfyUI_00337_.png (1.11 MB, 1024x1024)
1.11 MB PNG
>>
my best memory of pony is never having to use it since I am not poor and can afford a NAI sub
>>
>>8871168
>introduced score-based tags which are still used in anima today
and this is a bad thing
>>
>>8871159
catbox?
>>
>>8871223
https://files.catbox.moe/8ph0z5.png
caught me right before sleep
>>
>>8871136
/ldg/ highlights, very neat
>>
File: 1777787421123168.jpg (2.16 MB, 3785x3418)
2.16 MB JPG
>>8871234
Sorry forgot the highlits....../ldg/ highlights, very neat
>>
>>8871235
xlop
next
>>
>>8871235
damn these are fucking dogshit
>>
>lonyu_yog fries anima gens
how the mighty have fallen...
>>
>shitty meme snakeoil doesn't work
oh noes!
>>
>he doesn't know
stay uninformed, chuddie
>>
is ai image gen just getting worse? why does it feel like things are so stagnant the past couple of years? sometimes it feels like we are actually regressing
is it just that all the money is going towards video gen i guess??
>>
>>8871247
is z-image & flux klein not enough for you? not to mention anima. be grateful you're still getting fed.
>is it just that all the money is going towards video gen i guess??
yep. and little to none of it is going open weight :)
>>
>>8871247
business level budgets are needed to train at scale, payment processors don't like it, therefore there's no money in it, therefore businesses are not incentivized to do it

a simple train of logic
>>
>>8871250
>payment processors don't like it
explain how Kling and Vidu are clearly operating a successful business model
>>
>>8871247
its not really getting worse, its just not making leaps in progress. for me, it's the inability to really create custom face shapes/outfits. even the simplest character ideas can be made recognizable with slight tweaks to the face/hair. tags just aren't enough. we need a system like koikatsu character creator and an improvement to loras so that you can save and load things easily. i want to be able to mix and match eye and hair styles
>>
>>8871251
those are censored dumbass
>>
why the FUCK do i need to constantly remind people that this is /h/
>>
this
post stinky poop
>>
File: 61573890.jpg (176 KB, 832x1216)
176 KB JPG
>>
>>8871252
>we need a system like koikatsu character creator and an improvement to loras so that you can save and load things easily
this could be done with what we already have but it'd require a major overhaul to the tagging system. it'd additionally require VLMs that're able to accurately identify these things in the dataset.
>>
>>8871247
Retards spew shitposts like this meanwhile Pony v7 literally just released only a couple months ago
>>
File: ComfyUI_25812_.jpg (1.97 MB, 2048x3072)
1.97 MB JPG
xentai
>>
i am failing at loracoping... i tried to train an anima character lora with different outfits but the activation tag i made isn't working. this is my first time training a character
-if using a custom outfit activation tag, do i need to train the TE?
-if using a custom outfit activation tag, should i remove all outfit associated associated tags (white_shirt, boots, etc)?
i am already making sure to not drop out or shuffle the activation tags.
>>
>>8871260
unless you are taking over a tag that already exists and the training data is vaguely similar to it, you need to train the TE. you should generally always tag everything. if you skip some tags things will get baked in and associated with other tags which can create unpredictable results.
>>
Us in the know love that lil nigga touchdown russell like you wouldn't believe.
LET RUSS COOK.
>>
why not just vibe transfer the outfi-oh...he is still using local lol......
>>
more like lolcal
>>
basing anima off of cosmos was a mistake. that license is a nightmare.
>>
oh no not the license
>>
The licens- artists for leg fetish? i like large calves
>>
>>8871259
tamamo is my wife so by extension that has to be my penis. kinda small desu
>>
>>8871271
that was MY pov however
>>
>>8871271
she's going to tickle a kidney stone with her finger
>>
>>8871272
your pov would be from the chair in the corner
>>
>>8871274
Why is xhe a cuck?
>>
>>8871259
gonna be honest senpai, that right there is an illustrious 0.1 lookin dick
>>
>>8871271
>tamamo is my wife
Proof?
>>
>>8871277
>not using the penis lora
ngmi
>>
>>8871259
catbox? i like how smelly her groin line things or whatever they are called look
>>
>>8871277
no shit, retard. there is no e621 in dataset
>>
>>8871280
>catbox
She doesn't know..
>>
are furries coping with chroma or are they still stuck on sdxl?
>>
>>8871278
>Proof?
she said so when i played fate/extra
>>
>>8871286
rin already fucked her before you had the chance doe
>>
sirs where is the henti
https://litter.catbox.moe/fja18x.png
>>
>>8870136
https://mega.nz/folder/j7hH3LzK#6vrCekDb48zk6zeSGIjYHA
your slop, sir

again, this is another overcomplicated gacha design it only got the basics of. you also should have tagged "cross in eye" because it didn't really get it. it also didn't get the gold thing on the right shoulder, the gold rope and ID card on the right shoulder, the correct expression for the hairband, and the thigh bag design. i could probably have baked it twice as long as maybe it would work better, idk? i doubt it. to work properly it needs more data, lots of close-ups, lots of tagging.
>>
File: 20260503062209-4217090256.png (1.94 MB, 1280x1280)
1.94 MB PNG
>>
>>8871289
not like she was a virgin before that
>>
>>8871290
did she inject glowstick liquid into it or something
>>
File: ComfyUI_25829_.jpg (1.75 MB, 2048x3072)
1.75 MB JPG
>>
>>8871294
those are mutated churls, you uncultured swine
>>
>>8871284
.......................................................l-litterbox?
>>
how has ai still not caught up with the level of bizarre bestiality of the average modded fallout/skyrim
>>
chat, how do we stop anima from looking this ugly, realistically.
>>
>>8871303
more RL
>>
>>8871303
Skyrim had since 2011 to grow a library of insane fetish mods
>>
>>8871291
>@asura \(asurauser\) in negatives
based
>>
>>8871301
what kind of bizarre bestiality would one see in an average modded fallout/skyrim?
>>
WHEN CAN WE GET E621 ON ANIMA?
>>
>>8871307
any creature apart from mudcrabs
>>
>>8871308
Are you a furry or something?
>>
>>8871310
i don't like sterile boring illu v0.1 henti
>>
>>8871310
i just like genning nsfw, danbooru is mostly for 1girl standing
>>
>>8871310
it already has deviantart and other slop in the dataset. more knowledge cant hurt
>>
>>8868378
see >>8868398 >>8871136
>>
>>8871309
if anima had e621 in it's dataset it would catch up
>>
>>8871136
>Guide: (WIP)
>This guide is meant to replace https://rentry.org/59xed3, an outdated piece of crap
>last updated 7 months ago
yet another outdated piece of crap stuck in the OP
>>
/2hgg2/ would have a new and updated OP without outdated sdxl trash links
>>
https://litter.catbox.moe/um3sfs.jpg
>>
>>8871324
this kills the lolipedo
>>
>>8871324
box me up senpai
>>
>>8871327
personally i just like women so it's fine
>>
new /hdg/ is up
>>8871325
>>
>>8871330
>#780.1416
what?
>>
>>8871330
>such effort on op image for an indian cesspool thread
this is just sad
>>
>>8871333
probably some oldfag schizo trying to win back posters because of muh tradition and thread history
>>
>>8871334
better ban more loliposters
that'll revive the thread
>>
>>8871330
let it die, chud
>>
>>8871335
pedo melties did tank /hdg/s quality quite a lot but I think IAschizo is most at fault
>>
>>
>>
>>8871341
the alcohol (the good stuff)
>>8871342
the chaser (wash it down with some slop)
>>
>>8871342
What's your upscaling technique?
>>
>>8871345
high res lora
>>
doesnt anima support NL prompting? why not just include in the prompt that the image should look good
>>
File: file.png (80 KB, 942x367)
80 KB PNG
>>8871345
highres lora
this 2x model
https://github.com/Phhofm/models/releases/tag/2xPublic_realplksr_dysample_layernorm_real
downscale lanczos to 3.5mpix
(sometimes optional) img2img er sde, 7 or less steps out of 20 (eg: 7 steps .35 denoise)
then that opencv filter that got posted a ways back at these settings
i think it was this node, manually installed opencv to my comfy venv https://github.com/geroldmeisinger/opencv-comfyui
then saver node applied optimized png
>>
>>8871348
and all the opencv filter does is minutely denoise the image to save bytes on the png
>>
>>8871348
neat, thanks
btw id maybe change lanczos to cubic, lanczos causes pretty bad ringing when downscaling
>>
>>8871334
its the same thread, im making the next hgg but adding the number 781
>>
Yikes, who let the comfytroon break containment
>>
>>8871362
Chadmix is hdg/hgg thread culture, chuddie.
>>
>>8871364
nah, he's edg culture.
he was the schizo spamming their thread recently.
>>
/hgg/ culture is deadnoodles, loras and playful banter about seasonal anime
>>
>>8871366
Prove it.
>>
noticing a lot of "SEA people" in this thread
>>
>>8871372
>SEA people
We don't use that word here, brother.
>>
>>8871371
>>>/e/3071098
>>
>>8871375
Doesn't look like a proof that he's the spammer.
>>
>>8871379
yeah pure coincidence he gave a bunch of buzz to a civitslopper directly before that said civitslopper got scraped and had their content spammed
surely it was also pure coincidence that he was also active on hdg during its most unusable point when it was getting spammed.
just like it's pure coincidence that both threads were spammed in the same exact way; by slop scraped from civit and sloptreon scrapes from kemono.
>>
>>8871379
i am mr. chadmix, i gave pink poster 10K yellow buzz to try anima
>>
https://civitai.red/models/2595035
e621 bros, we're so barack
>>
Chadmix is the only cumfytroon that respects our eyes enough to share readable metadata so he is incapable of wrong doing. Too bad!
>>
>>8871383
are you okay?
>>
>>8871383
Imagine if the person behind this post actually went out and contributed to society
>>
how come there is still no script in the op that upscales downscales datasets to fit into batch divisible buckets?
>>
>>8871387
so long as I have access to corkboard, yarn, pushpins and a functional printer.
in all seriousness though it's either really bad luck for someone known for being a notorious faglord
or
the guy is just a notorious faglord and the only bad luck was that he wasn't expecting the pinkfag to namedrop him.
>>
>>8871389
just ask your favourite llm to slop one up for you
>>
>>8871383
sus
https://files.catbox.moe/obqlbh.jpg
>>
>>8871391
can't wait for it to import ffmpeg and every python library known to man
>>
>>8871394
ask it not to do that then dumbass
>>
>>8871394
pillow is all you need
>>
File: 1777816736557828.png (2.63 MB, 1024x1536)
2.63 MB PNG
/hgg/ opinion?
>>>/g/108744242
>>
File: TeaTime2.png (2.11 MB, 1216x832)
2.11 MB PNG
>>
File: 1748038786437713.png (1002 KB, 1024x1024)
1002 KB PNG
i'm still new to this, but it's a lot of fun.
still haven't figured out how to upscale without it looking bad
>>
>>8871404
i'm going to shit down your throat
>>
>>8871404
The completely fucked Teto wink at the bottom really is a cherry on top of this absolute dogshit guide.
>>
>>8871404
I also love spreading misinformation on the Internet
>>
best vpred checkpoint?
>>
>>8871404
waow are those uncensored nipples?
>>
>>8871383
It could, in fact, be a coincidence. Or actually him gifting points to pink retard could be the trigger to schizo's meltdown.
Seems to me that you're the spammer trying to frame another poster.
>>
>>8871411
lil blud still stuck in 2025 ::skull::
>>
wait I can give buzz to make people spam here? who is signing up?
>>
>>8871414
not downloading anima until it actually releases, waste of effort finding mixes
>>
Tier list of revolutionary local image models:
S TIER:
>Pony Diffusion - Ushered in a new era, the first real model to break local free from the NAI leak. Managed to unify anime, furry, pony, and realism in a single model. For a moment, all was at peace.
>NovelAI leak - Kickstarted local, introduced all of the major concepts that local still relies on today such as bucketing. In a sense, NovelAI founded local.
A Tier:
>Anima - The first promising local model in over 2 years. Better prompt comprehension and clarity, but outdated dataset and limited parameters keep it from reaching S-tier.
>Illustrious XL - Rekindled the flame for SDXL and brought in a large wave of lora bakers
B Tier:
>NoobAI XL v0.75 - An expansion to illustrious with a bit more data
C Tier:
>DreamShaper 8 - Proved how anime and realism could be blended together
>Waifu Diffusion 1.2 - Introduced nsfw to local models but was quickly surpassed by the NAI leak
D Tier:
>Chroma - Furry slop that failed to learn a single character despite $200,000 worth of training
>NoobAI XL Vpred - A desperate attempt to chase NovelAI. In the end it was completely fried.
F Tier:
>Flux - nothing but a full year of coping
>NewbieAI/NetaYume - Bloated failbake
>>
This post brought to you by Novel AI
>>
>>8871418
if we did a tier list of /hgg/ posters where would you fall?
>>
>>8871418
Sir you forgot mugen >:(
>>
>>8871418
You forgot
>SSS GOD PLATINUM DELUXE TIER
>Pony Diffusion V7
>>
>>8871422
bro is literally leaking trade secrets in general chat :skull:
>>
>>8871422
>v7
>he's not in the super ultra goated whale tier hyper sekrit v8 beta
shiggydiggy
you'll never get anywhere like that jabroney
>>
clowns
>>
clownussy
>>
File: ComfyUI_temp_ugdio_00257_.jpg (579 KB, 1688x2168)
579 KB JPG
>>
>>8871428
w gen
>>
>he doesn't even hide his shitposting
you gonna spam this thread with patreon watermarked slop too?
>>
let's see paul allen's slop
>>
File: 1769816717586318.png (2.7 MB, 1728x1344)
2.7 MB PNG
BBClowns
>>
feces
finally
>>
>still seething at superior black chads
when will little ytboi learn
>>
not a fan of literal shit, i'm sorry
>>
>>
>>8871291
much appreciated sir
and yeah, it's just overdesigned gacha sloppa so it's understandable
I don't need it perfect
>>
>>8871435
there were a non zero number of times I saw a bbc thumbnail and though it was scat kek
>>
>>8871291
I would use cross-shaped_pupils instead. Very low image count for a concept, and this one is close enough.
>>
someone bake the princess peach lora already ffs
>>
i sleep on gacha until ppk comes to gfl2
>>
>>8871453
heh peepee
>>
File: 1759607694134843.png (1.98 MB, 1024x1600)
1.98 MB PNG
>>
>>8871458
can you make the proportions even more hilarious
>>
>>8871458
hello sir.. maek their penises touch pls. make ite man penny shiver plz
>>
>>8871458
that's borderline /aco/
>>
>>8871461
it is /aco/
they post it here on purpose
>>
>>8871458
ignore those losers. catbox plz
>>
>>8871458
just cut to the chase and start spamming already. you already killed any interest in this thread, may as well go the full mile.
>>
>>8871428
>>8871434
>>8871440
>>8871458
this is the nsfw version of astronauts riding dogs on mars and the kind of post you'd expect from /ldg/ users
>>
>>8871465
it's not groidy, please relax
>>
Why do people here lose their shit the moment AIslop doesn't affirm their whiteness
>>
>>8871465
did the size comparison hit a nerve?
>>
>>8871465
hilarious meltie
>>
if noodlekek is a slut for bbc so am i
>>
>>8871466
>astronauts riding dogs on mars
sounds like some reddit tier gens ngl
>>
>>8871474
you're smart
>>
I’ll spam up this miserable aikek pajeet thread later.
>>
>>8871478
promise?
>>
>>8871477
sorry bud can't be bothered to read the whole thread and decipher between useless shitposts and other
>>
why are anti-ai cucks always so high cortisol?
>>
>>8871482
Because they spend their time looking for shit like >>8871483 on cucktreon so they can spam it here
>>
>>8871483
i want to worship xer...
>>
>>8871483
This is insanely good, dare i ask:
More like this???
>>
File: ComfyUI_38710_.png (1.39 MB, 1024x1024)
1.39 MB PNG
>>8871453
M16 for me
but Endfield girls are fine in the meantime
>>
>>8871483
Okay now, why was this deleted?
>>
>>8871493
Given how quickly it happened it must have been either loli, furry or he deleted it himself.
>>
https://aibooru.online/posts?tags=drawfag
How come no one in this thread ever gets his works uploaded here, except for firedotinc?
>>
>>8871458
who else is this besides feipin zangshi?
>>
File: 1772113531029074.png (2.72 MB, 1248x1824)
2.72 MB PNG
>>
>>8871447
They aren't cross-shaped pupils, though. They are highlights shaped like a rounded cross.
>>
>>8871290
>>8871434
box plz
>>
>>8871521
cute fat slut
>>
>>8871526
why is bro yassified
>>
>>8871526
blud got adetailer face.
>>
>>8871527
it's ayakon
>>
>>8871526
>the tail grab
>the looks of ecstasy
>that wonderfully deep penetration
>the amount of motion where the image is practically animated
>the subtle but tasteful use of color
>the metaphorical merging of their legs showing that these two people have become one during love making
It's really hard to make true art with AI but you have found a way. Bravo, sir. My hat goes off to you.
>>
>>8871521
Why is the penis so small
>>
>>8871536
It's my pov
>>
>>8871536
look at his skin color
>>
>>8871536
it's white
>>
>>8871521
can you add more ring light reflections please
>>
fake ayakon
>>
>>8871507
He uploaded one of my worst image, and nothing else after.

>>8871522
Yeah, but retraining a tag that's "close enough" is easier than training one that doesn't yet work at all.
>>
>>8871536
it's average
>>
>>8871543
uh, no? it got it kind of right already even without being tagged
>>
>>8871507
because no one else recognizable even posts here anymore, the majority of images posted are shitposts so why would i bother uploading those
>>
>>8871547
>i
>>
there are too many layers of trolling for me
>>
this general is like an onion
>>
>>8871550
We stink?
>>
>>8871551
if the shoe fits
>>
>>8871551
and make people cry, yes
>>
you reckon anyone has closed this general with a tear rolling down their cheek before
>>
me
>>
>>8871546
Yet you still told him to add the tag. You're hoping it'll make things better, and my guess is it won't due to its low image count.
>>
you reckon anyone has closed this general with a tear rolling down their leg before
>>
I piss and shit and cry every time someone doesn't give me a (you)
>>
>>8871554
ye I was peeling onions
>>
why doesn't /hgg/ have its own highlights
>>
>>8871556
no. it's well-represented in the dataset, which as i said, is why it kind of works a bit already. it's much more reliable than that fuckass thigh bag design which it got right precisely zero times. it's just that an eye tag was not put in the dataset so it didn't train on it.
>>
I miss highlights
>>
>>8871527
consequences of 99% unbalanced 1girl standing training data
>>
said no one ever
>>
>>8871561
fair enough
>>
>>8871526
why deleted?
>>
>>8871526
the real ayakon would've posted the current FOTM gacha slut btw
>>
>>8871563
more like consequences of running Ddetailer with the same prompt on both faces
>>
>>8871566
fur on face is over the line
>>
>>8871566
pulchra is banned on /h/
>>
Furry/guro/lolikon/shota porn outside of /b/
>>
>>8871507
Hey, my slop pics got there at least!!
>>
Is there much fur on that face, maybe some sideburns
>>
>>8871526
Yeah this is certainly a shitpost, no way someone can be this bad after 3 years.
>>
>>8871576
you so sure about that?
https://www.pixiv.net/en/artworks/144088016
>>
I actually kinda like that
>>
>>8871577
This is nowhere near as awful.
>>
why is anima so sloppy
>>
>>8871577
why is he not tagging his art as AI?
>>
>>8871580
>the thumb just disappears after the knuckle(no, it is not just "in her mouth," there is no depth to the insertion, her tongue does not depress, it's just missing)
>absolute MOUNTAINS for shoulders with a sunken neck
>garbled mismatched earrings
>melted eyes on something where they're a primary focus
>some of the worst upscaling imaginable
it's pretty awful, anon.
>>
>>8871583
pixiv don't give a shit unless you spam images and it means you don't get filtered
>>
>>8871584
I know. Didn't say it's good but it's kinda regular levels of slop. You kinda have to look at the pic for more than 3 seconds to really see all the shit while with Pulchra gen you can immediately see how fucking bad it is.
>>
it's just ai generated images
>>
Sars! (๑˃̵ᴗ˂̵)ﻭ
>>
new kid on the block
https://huggingface.co/well9472/Nanosaur-1.2B-Preview
danbooru + e621
>>
>>8871591
Anima?
>>
>1.2b
lol
lmao
so it's sd1.5+
>>
>>8871591
naisaur...
>>
>>8871591
whitesaur...
>>
jav lora doko
>>
>>8871600
make the dataset and ill bake it
>>
>>8871583
It's AI assisted
>>
>>8871606
adding a watermark with photoshop does not make it assisted
>>
how do I get as dedicated a fanbase as ayakon has here
>>
>>8871608
post bobbleheads
>>
>>8871608
add a name to your gen and make your gens consistently mediocre to bad
>>
>>8871610
but why does anyone read the filename
>>
>>8871607
yeah the only person ive seen who does ai assisted art is firedotinc, he basically redraws all the line work before inpainting multiple times
>>
>>8871611
yes
how do you think 'ayakon' stuck
>>
lots of /hdg/ culture in /hgg/ today
>>
>>8871614
whats the difference
>>
>>8871616
shitposts are /hdg/ and genuine tech discussion about loras is /hgg/
>>
the good things are by me and the bad things are by him ahhh post
>>
yes
>>
name a more influential poster than oekakischizo
>>
>>8871619
post #10 tourist spotted
>>
File: ComfyUI_19136_.jpg (1.65 MB, 2944x1408)
1.65 MB JPG
>>
>>8871624
no fre*king way in HECK that this is ia
>>
>>8871624
highlights coded gen
>>
>>8871621
iaschizo caused the split of /hdg/ and ruined /e/
miqoanon and cathag caused the great pony meltdown
noodlefag ushered in a new era with /hgg/
jordach grifted 10000$ from the posters here
>>
>>8871631
>jordach
kek i forgot about the cascade arc. model still cooking on a 2x480ti rig
>>
>>8871621
highlights chad
>>
File: ComfyUI_19100_.jpg (1.48 MB, 2432x1408)
1.48 MB JPG
>>
>>8871631
>>8871633
Why doesn't Ia schizo at least use something like civitai at least?
He dedicates so much time and passion to aislop and never genned anything.
>>
>why isn't the guy that obsessively hates AI and has made destroying thee AI threads on 4chan his life's mission genning anything
>>
ia schizo doesnt hate ai doe
him ruining threads by his actions doesnt mean he has malicious intent. literally everything points to him being just a low iq spic with a very poor taste
>>
>>>/trash/82614653
>>>/trash/82614664
noodlecuck's contributions to AI art today!
>>
>>8871640
churl masks doko
>>
What would deadnooodles think of what noodlekek has become?
>>
>>8871640
why did noodlefriend abandon us...
>>
>>8871640
honestly wasted on the rest of that dogshit thread
>>
File: 00019-1695515612_result.png (1.32 MB, 2432x1664)
1.32 MB PNG
>>
why is it so blurry
>>
>>8871645
what's the point of avatarfagging with offtopic shit
>>
>>8871640
i get why he wants to gen netorare but blacks really are just poor taste at this point. just make it some hairy muscular guy or even a fat bastard.
>>
>>8871645
>upscaling with a model that cant handle blur
ngmi
>>
>>8871649
>i get why he wants to gen netorare
0 testosterone from making porn and playing braindead videogames all day?
>>
File: 00017-780369530.png (3.94 MB, 2432x1664)
3.94 MB PNG
>>8871641
there

>>8871649
that's oddly specific

>>8871650
it was a basic bitch upscale, yes
>>
>>8871652
wai is it so wai?
>>
>>8871652
are genshin girls really such sluts that getting dped by monsters in some lonely forest is only a mild annoyance to them?
>>
>>8871651
some of the angles and expressions look pretty good. also it's not like you get a testosterone boost from genning pov cowgirl all day kek
>>
>>8871654
that is why the my go in the forest
>>
>>8871654
on the particular cause of Eula, it is
>>
>>8871654
stop trying to rationalize cuck porn, there's never rhyme or reason to it
>>
>>8871652
finally, the real /hgg/ culture
>>
cuck schizo is seething again
>>
[headcanon] is flipping out guys omg
>>
File: 00230-3937435278.jpg (1019 KB, 2688x1536)
1019 KB JPG
Let's member when (You) hated how vanilla my gens used to be
>>
>>8871666
I don't mind your churled or blacked or rent a gf ntr posts
>>
File: 00026-3462728431.png (3.61 MB, 1536x2560)
3.61 MB PNG
>>
>>8871666
Hated by some shitposter? Anyway, nothing bad about this composition, the style is mid though.
>>
File: 00252.jpg (2.29 MB, 3072x5376)
2.29 MB JPG
>>
>there is nothing in between vanilla and bestiality ntr
what causes this
>>
File: 00163-1198748599.jpg (589 KB, 1664x2432)
589 KB JPG
>>
File: 00135-680542362.jpg (770 KB, 2432x1664)
770 KB JPG
>>
File: bocchiMastFinal.png (2.94 MB, 1536x2560)
2.94 MB PNG
>>
>>8871671
>>8871673
>>8871674
>sex with nothing to censor
Should have started pixiv.
>>
File: 00157-2026257402.jpg (645 KB, 2560x1536)
645 KB JPG
>>
File: 00214-2656891727.png (3.92 MB, 1664x2304)
3.92 MB PNG
Have a lot more of course but I am trying to make a point here
>>
>>8871678
now some blacked to balance it out again
>>
>>8871672
same with loli shit here, it's either toddler shit or "petite"
>>
>>8871678
2000 vanilla gens don't wash away the sin of making a single blacked gen, sorry, it's over for you.
>>
File: 00459-684587117.png (1.04 MB, 832x1216)
1.04 MB PNG
>>8871683
I hate you and I hate you all
>>
>>8871682
loli is against the board rules tho and it's better to be strict with "petite" so nobody slides down the slippery slope
>>
its crazy how vanilla is somehow more gay than a blacked otoko no ko
it feels like such an over compensation for closeted homosexuality
>>
>fag woke up
time to leave
>>
>>8871685
>loli is against the board rules
good thing that toddcon isn't, shitposter-kun
>>
what do i gen
tell me
>>
>>8871698
otoko no ko x hilichurl
>>
>>8871698
beach nanpa with haughty gacha sluts
>>
>>>/b/948726750 KEK look at her blushing, dreaming of majestic creature, too bad /b/ can't execute a good idea for shit
>>8871698 ^ this but better
>>
>>8871645
>>8871684
>sfw
>loli
>>>/g/adt
>>
>>8871708
those are just reposts
that person hasnt posted here in a long time
>>
>>8871709
noodlekek is reposting his shit all the time
>>
>>8871667
why it's deleted
>>
>>8871710
his thread. back to /hdg/ if you don't like it
>>
is image aspect ratio important when it comes to training? I noticed some images feel stretched on some square images rather than having the a certain width of a torso, making it look thicker.
>>
>>8871711
Dude forgot his penis at home. Can't be /h/ without penis.
>>
>>8871714
>Can't be /h/ without penis.
You can do 1girl and if it has a drop of semen in the far left corner, it's /h/.
>>
>prompt for two separate characters and unique features for both
>anima fails and mixes features between them
I thought a LLM encoder was supposed to avoid this?
>>
2girls, kiss, yuri, tongue out, saliva trail, hand in another's hair, blush, body blush, steam, cum on drawer
>>
>>8871591
>from scratch
DOA, don't even need to download to know, lmfao
>>
>>8871720
i think the lack of images itself paints a grim picture for whatever that guy is cooking
>>
>>8871591
>no preview images
>>
>>8871718
Until we get a novel AI v4.5 leak I wouldn't expect local to ever be able to do this without numerous extensions and massive amounts of inpainting
>>
>>8871723
I gen with NAI quite often, it mixes traits from separate prompt boxes all the time.
>>
>>8871718
utilize NLP its not hard
>>
>>8871718
>I thought a LLM encoder was supposed to avoid this?
Are you specifying which character has what feature though? with just character, feature, feature, feature, character, feature, feature, feature it would need to either: be trained like that, which it wasn't so it will mess up, or learn to automatically associate features with characters, which would take really long to train, to test the bleeding you can do one of those prompts like: a red glass sphere on the top right, a blue metal cube on the top left, a magenta marble cat statue on the bottom right, a orange die on the bottom left
>>
>>8871714
It was a quick gen and i fucked it up. Might reposted it later once I fix it
>>
>>8871698
Uncensored handholding
>>
>342 / 35 /
the fall of /hgg/
>>
lots to discuss as ALL of our loras bake
>>
>>8871729
lora discussion general, tourist
>>
>>8871730
this/ is the true lora general >>8871404
where is your infographics?
>>
>>8871732
where are your factual loras? ldg has more loras per anon than you. this week we updated the chris chan lora, we created the family guy lora and updated the basedjak lora for both Anima and Z image
>>
>>8871734
FF7 remake lora when
>>
>>8871734
where jav lora
>>
>>8871678
add overexposure to negs blud
>>
>>8871737
you are two years later bro
>>
>>8871737
This never works...
>>
>>8871734
I can't tell if you're joking
>>
File: 00024-1489171954.png (1.43 MB, 1088x1536)
1.43 MB PNG
did someone mention something about a new lora
because I just had a lora finish training
>>
File: 00026-2782122412.png (1.36 MB, 1088x1536)
1.36 MB PNG
it's a good lora. I promise. better than anything any other shitty general thread could produce. it's homegrown, organic, very healthy, super good for you.
it cures ed, no doubt. prevents testicular and prostate cancer. increases virility. gives big energy.
>>
>>
>>
>>
if its not jav lora i dont care
>>
>>8871758
Is it the style, the character, or the swirly eyes concept
>>
>>8871698
genderswap churl
>>
>>8871521
>2xPublic_realplksr_dysample_layernorm_real
Sir, the source of this snakeoil please?
>>
>>8871769
it was posted in this VERY thread
do some sleuthing, chuddie
>>
>>8871600
>>8871601
Hasumi Kurea, my good sirs. JAV has been in an eternal tailspin since the leggy goddess paid off her debt to the Yakuza and now exclusively works gentleman bars.
>>
File: catbox_baydri.png (1.35 MB, 1536x1088)
1.35 MB PNG
>>8871760
>>8871761
>>8871762
these are SHIT!
HORRIBLE!
BAD!
MAKE DICK GO SOFT!
learn from the REAL master!
download the newest greatest latest best lora today instead of using this garbage!
>>8871764
style!
just use trigger tag "@cshow" !
it's a pun! get it? cshow? shishou? because is master! will teach and improve all things!
pixeldrain com/u/zGwENorm
(just insert the dot between pixeldrain and com)
here is a catboxed gen with example metadata! it's so easy even child can do it!
>>
:skull:
>>
>>8871770
>chadmix approved
Chad don't miss.
>>
yo new shitpost just dropped
>>
File: 00034-1924198869.png (1.38 MB, 1088x1536)
1.38 MB PNG
lora is versatile! can do no wrong!
so powerful my power went out during bright sunny day with clear skies on first attempt of training!
(seriously, it actually did)
one can barely hope to contain such greatness!
but it has been done!
>>
ugh
>>
>>8871713
yeah
>>
File: 00042-624763090.png (1.37 MB, 1088x1536)
1.37 MB PNG
even smaller bodies prove it no problems!
https://files.catbox.moe/vjg87m.png
>>
pony-v6-extract looking good ngl
>>
what causes this
>>
who is melting and why
>>
File: 00046-3052319193.png (1.19 MB, 1536x1088)
1.19 MB PNG
okay but if I may be serious for half a moment, this shit is so cursed that, in addition to my power mysteriously blinking out for about 5 seconds when it was at around 300 steps, even pixeldrains login servers are completely shitfucked. I had to create a new account to upload it there, and that session lasted just long enough to upload it and grab a link and it immediately locked me out afterwords(with no ability to log back in).
>>8871782
you just wouldn't get it, man.
>>
>>8871784
i don't get it what's the point of it
>>
display xis mental illness yet once again
>>
File: 00051-2419800302.png (1.68 MB, 1088x1536)
1.68 MB PNG
>>8871786
I found a cursed dataset I was stupid enough to assemble a year ago.
and this thread is bottom of the barrel trash tier in how bad it's been. So what better time to do this than when it can't possibly make things any worse?
>>
see?
>>
>>8871788
okay so you're just a fucking retard got it
>>
man this shit hole really is /hdg2/
>>
>21 hours
>382 / 45 /
the fall..
>>
File: 00055-831857958.png (1.39 MB, 1088x1536)
1.39 MB PNG
>>8871790
let's see what you've done to be productive today.
>>
>>8871794
That's actually a rise, tourist. A rise of shitposting but a rise nonetheless.
>>
>>8871795
already did
>>
>>8871590
pretty nice pose honestly. box?
>>
>>8871600
i will also bake if you make a decent dataset for it
>>
nobody will make a dataset for that, that's why we're asking, because we're lazy fucks.
>>
>>8871795
slutbox?
>>
>>8871747
jack up the power
>>
how do i find a successful artist mix to use for my AI patreon? what should i be focusing on
>>
>>8871808
getting a job
>>
>>8871803
Why is it that a weeb oriented degeneracy general cannot produce a dataset or lora for jav but there was a BLACKED lora in /hdg/?
>>
I just couldn't get the controlnet models to work with anima, even the ones who supposedly were made for it.
>>
File: 00062-142102846.png (1.43 MB, 1088x1536)
1.43 MB PNG
>>8871804
https://files.catbox.moe/7xpzqy.png
lora is here >>8871772
>>
>>8871813
openpose is fucking shit always was
do any of the other ones work
>>
>>8871813
because forge neo needs an update for them to work.
(I don't believe it has received said update)
>>
>finally update Neo Pancho
>Shift
tf is this?
>>
>>8871817
it's a feature of models that use qwen's TE or one similar to it such as anima. it has something to do with how much time the model spends drawing details as opposed to composition, as i understand it, but i haven't tested it at all
>>
basically, it's a real setting and not some snakeoil
>>
good anima gens.... anyone?
>>
>>8871821
on my hdd
>>
>>8871822
you mean hgg?
>>
>>8871823
nyo
>>
last thread had some cute gens tho
>>
>>8871819
>>8871820
Huh. So cranking it up gives more detail? I ran a pre shift seed with the default 3 and didn't notice much difference. There were but not much. I cranked it up to 5 out of curiosity and anything above 3 kept on flipping a foot to the wrong direction. Granted, it was one of those angles that can confuse ai but it was consistent after 3. Below 3 it got it fine.
>>
lolikon doko
>>
>>8871814
what is this lora supposed to be? an artist style lora? cshow gets no hits
>>
>>8871829
looks like @ai-generated
>>
File: 00065-1474599180.png (1.4 MB, 1088x1536)
1.4 MB PNG
>>8871829
it's a style.
and ancient chinese secret.
don't worry about it.
>>8871832
I mean, you're not exactly wrong.
>>
>>8871834
so it's a style lora trained on gens of an artist mix. just lead with that bro
>>
File: ComfyUI_25923_.png (1.78 MB, 1536x1024)
1.78 MB PNG
>>
>>8871835
anon, it's ayakon.
>>
>>8871836
change the artstyle
>>
File: file.png (55 KB, 785x420)
55 KB PNG
>>8871819
>>8871826
It basically just drags the scheduler's curve left or right. Has nothing to do with the model. Different schedulers will have different effects when shifted, obviously.
>>
>>8871840
i was lied to
sad!
>>
>>8871815
Poorly. And it becomes obvious that it tampers with the style.

>anima-lllite-any-test-like-1-step1000
It tries. Low-middle success.
>anima-lllite-any-test-like-1-step2000
Doesn't even try at all. Fleshfiends.
>anima-lllite-any-test-like-3-step00000550
Doesn't even try at all.

>anima-lllite-scribble-1
Utter dogshit.
>>
>>8871843
be that as it may i have examined your prompt and determined you are a fucking retard so i shan't be talking with you anymore
>>
>>8871845
I literally get even worse results if I try to use natural language.
>>
>>8871846
>unable to get a specific 1girl pose with natural language
simply a skill issue i think
>>
>>8871848
Run tests with controlnet and show me.
>>
>>8871849
Show me pose you want her to be in and I'll do it without a controlnet.
>>
>localkeks still coping with controlnet
it really is true, local never left 2023
>>
>>8871850
What I want is not to roll the damn dice.
>>
>>8871851
didn't nai promise controlnets for v3 like 3 years ago and then just never did it
>>
>>8871853
Keep coping then, chuddie.
>>
so many of the baked in artstyles on anima are utter garbage. when will we get a model that doesnt require loras
>>
>>8871856
this so much this
if a gen isnt using a lora its usally ugly
>>
how hard is training a controlnet
>>
controlnets should've never been used on anything beyond sd 1.5. let it go.
>>
probably just retarded shitposters but i've already found decent artist mixes on anima. most of the posters here that aren't just retards spamming with shitposts train loras so naturally they'd have them in their gens
>>
unfortunately my go-to artists for mixing either
a. overpower every other artist
or
b. are a far cry from the actual artstyle
>>
This thread is unusuable
>>
>>8871854
yes but then they realized it was outdated tech and didn't bother with it any further. controlnets are cope in today's era, you can just ask grok for any pose and it natively understands
>>
>>8871864
>you can just ask grok for any pose and it natively understands
i highlight doubt that works at a certain level beyond simple poses
>>
you don't need to use artists in your prompts just add "masterpiece"
>>
>>8871858
>>8871860
>>8871861
>>8871862
What the fuck are these nonsense posts? Is this a bot? Brain fatigue from reading this kind of post searching for something interesting behind hundreds of shitposts, this sucks.
>>
>>8871867
Again what is this?
>>
im not using my brain until we add the counter from all the hdg threads to hgg its the same thing
>>
lil bro is lost
>>
lil bro is going from doing nothing to doing nothing
:skull:
>>
>>8871868
hello, newcutie
>>
>>8871872
>>8871870

Dude is not normal. 441 messages in less than a day is either a raid, a bot raid, or an anon with typing Tourette's who can't keep to himself what he's thinking at any given moment. This thread is unnavigable.
>>
Guys help, I'm unable to glean information from context clues! What does controlnet mean? Is it a schizo?
>>
actually im a regular and you got baited, thanks for the you
>>
File: catbox_toxkid.png (744 KB, 1024x720)
744 KB PNG
>>8871867
damn
>>
File: 00004-4169825499-edit.png (2.55 MB, 1248x1824)
2.55 MB PNG
>>8871878
kek
>>
>>8871880
(wet, liquid, water:5)
>>
>>8871880
nice idea but I am not vibing the style
>>
>>8871856
Was it really necessary to make this comment, you absolute moron? What kind of response were you expecting? You could answer that yourself if you just sat still and thought for a second, faggot
>>
(good thread, high quality thread, score_9_thread, photorealistic thread:3)
>>
>>8871881
you forgot shiny skin and oiled
>>
>>8871885
>>8871881
This is noise, both posts could've perfectly well never been posted and it would've been better
>>
>>8871886
>>8871883
Why is ChatGPT in Uppity mode?
>>
>>8871887
I'm not chatgpt—you know
>>
>>8871886
This is noise, this post could've perfectly well never been posted and it would've been better
>>
otoko no ko
>>
>>8871888
>—
Uh.. sisters?!
>>
This thread isn't a central hub for image gen—it's a cesspit of bottom feeders, endlessly nipping at the toes of anyone who might enter.
>>
>>8871889
This is noise, this post could've perfectly well never been posted and it would've been better
>>
this thread isn't just the hottest place for img gen discussion — it's a community
>>
>>8871892
is this where you segway into the /ldg/ part of the routine?
>>
This isn't noise—it's a sign of mental illness
>>
are your steps aligned?
>>
>>8871895
Waiting for /ldg/ new bake and highlights to share it here :3
>>
>>8871898
>>
xir the steps have fallen out of order
redeem the steps xir
>>
File: 00020-1305766657.png (2.31 MB, 1432x1840)
2.31 MB PNG
>>
>>8871901
nikkerbox?
>>
File: 00022-877151777.png (2.27 MB, 1152x2016)
2.27 MB PNG
too much spam and not enough images so we must redeem
>>
>>8871901
>>8871903
Demanding some boxes to keep the ai agents at bay
>>
>>8871903
i like the anime aesthetic with tastefull hint of realism
>>
>>8871905
boy do i have a website for you
>>
>>8871906
they peaked in 2018-2020 and it's all been downhill since
>>
File: 1777857583825622.jpg (1.93 MB, 3859x3458)
1.93 MB JPG
>>8871136
/ldg/ collage, neat
>>
are you ready for pony v8?
>>
>>8871903
I like the realism aesthetic with tastefull hint of anime
>>
>>8871910
It just arrived, look up
>>
File: ComfyUI_25952_.png (1.63 MB, 1536x1024)
1.63 MB PNG
tuggin'
>>
>>8871913
can you crop out the women
>>
>>8871909
Las manos mas asquerosas de un hombre con defecto mental.
>>
>>8871913
>>8871836
Women never loved me, I can't conceptualize myself being with 2 women smiling at me while naked, never happened in my life, I couldn't even imagine how I could have arrived at a situation like that.
Sorry but can't immerse myself in those gens, 0/10
>>
>>8871917
>>8871913
can you img2img them to men laughing at me
>>
por favor te lo ruego publica más generaciones de caca negra
>>
>>8871913
can you make the females smaller
>>
now that's what I call relatable
>>
can you make this thread better?
>>
>>8871913
Usecase for the naked male in the gen?
>>
File: 3dgifmaker53239.gif (420 KB, 300x300)
420 KB GIF
>>
File: ComfyUI_25953_.png (1.64 MB, 1536x1024)
1.64 MB PNG
>>8871917
does this help
>>
causings are strong today
>>
no decline in sight n.n
>>
kys darknessfaggot
>>
File: 00143-877151777.png (2.04 MB, 1248x1824)
2.04 MB PNG
>>8871901
what style(s) is this?
>>8871904
year 2025, @feipin_zhanshi, @ie_\(raarami\), @hintobento, @mikoyan, @saigalisk
>>
>>8871890
I did one the other day but I don't think it came out as cute as I wanted
>>
hes gone i like to keep the memory alive
>>
tears of laughter from the spinning cock for anyone asking if i've ever cried from the thread btw
>>
>>8871925
A little bit, in my limited dating experience, it's basically them being uncomfortable 10 minutes into the convo, looking away, hugging themselves and glancing around searching for knowing looks from passersby with a "please someone rescue me" expression.
>>
>>8871935
you should date men instead
>>
>>8871924
>jeetspin.com
>>
>>8871935
All this because beig shy and virgin is worse than being Hitler or a serial killer.
>>8871936
Thats why i'm here.
>>
monkaS
>>
>>8871932
proofs
>>
>>8871938
>>8871935
>>8871917
>>>/r9k/
>>
>>8871942
oh fuck off, i just jacked off and spurted a little cum splotch on those posts so those are okay to post on /h/
>>
>>8871941
check the 'cord
>>
really anything is allowed here unless it's /h/ but 1% too furry
>>
>>8871946
just dont do anything beyond ears and a tail and you'll be okay
>>
>>8871946
keep seething ayakonda
>>
doro is allowed btw
>>
tamamo tailjob...
>>
>>8871950
large fox tails sound kind of shit for tailjobs
>>
>>8871951
im a fan of the excessive fluff
>>
>>8871952
why are you a furry
>>
>>8871953
i dont like anthro
>>
File: sm.mp4 (3.17 MB, 1440x1920)
3.17 MB
3.17 MB MP4
>>
>>8871492
what style is this? trained on eosfield screenshots?
>>
>>8871953
nta but personally there's a big difference between "western" furry and jap furshit. i don't mind characters like the fluffy catgirl from dungeon meshi or that one bitch from that trigger anime but when you look at what furries post and it's an aco drawing with the head replaced by an animal it's disgusting
>>
idk about you but i'm not kissing a snout
>>
>>8871957
I agree but that is why I think pulchra should be allowed as an eastern borderline character
>>
>>8871957
>nta but personally there's a big difference between "western" furry and jap furshit
it's called kemono
>>
that sure is a lot of words to just say you're a furfaggot
>>
lil bro believes his words hold any weight
>>
furry is inherently /aco/, the same way like pedophilia is anime
>>
>>8871959
i don't follow gacha shit so i had to look ehr up but yeah that char is right on the borderline for me
>>
i'm tired of looking at local gens
everywhere i go (/2huAI/ is the most recent one i've visited) it's all local slop and people arguing about the technicalities of training and prompting ad nauseam for entire threads as if any of it is gonna make a difference but no it all looks like the most stereotypical local SLOP
>>
i'm tired of reading your posts
>>
i'm tired
>>
i'm
>>
all of me...
>>
no idea why more people arent using novel ai, it just works i never have to reroll and you can actually prompt more than one character at a time
moving onto this from local was like night and say its honestly insane how far ahead of feels
>>
>>8871975
>why more people arent using novel ai
i'm poor
>>
>company with a shit ton of revenue and access to huge clusters can train better models than hobbyists
color me surprised
>>
Inpainting LLLite controlnet lora
https://civitai.red/models/2596579/anima-lllite-inpainting
>>
>ayakon lora got posted
>suddenly nai schizo is freaking out spamming the thread
hmmmmmmm
>>
>>8871977
You don't need any money to train good models, just technical know how and some time.
>>
>>8871983
>You don't need any money to train good models
forgot my grandpa owns an 8xH100 cluster whoops
>>
>You don't need compute to train good models
how does someone become this retarded?
>>
sirs does my ps2 have enough computer to generate the hentai
>>
yes
>>
anima was trained on a ps1 and a couple rubber bands
>>
Here come the trolls to pretend they know anything just because they post in hgg. Post a lora that you've trained. I know you won't because you never have.

Meanwhile I have more buzz than you can ever imagine.
>>
reminder that jordach simulated an universe on his coal powered rig
>>
SDXL bros...
>>
reminder that cascade is pumping along on the steam-powered intel 486 rig
>>
I actually made gpt 5.4 on my rasberry pi, it's just about being efficient
>>
>>8871986
you could probably do image gen on an android at this point
>>
i gave an artist a quicke and got a drawing for free
looks better than anima
>>
>>8871994
yeah just open chrome and go to novelai.net
>>
i'm using brave tho
>>
>>8871996
i can already access forge or cumrag from my phone tho
>>
552 / 62
1104 / 124
2208 / 248
2260 / 300

Is this a good post per image ratio?
>>
>>8872001
meanwhile on /hdg/- 625 / 235
>>
File: file.png (133 KB, 1520x1406)
133 KB PNG
>>8871977
>he still thinks it's about money and compute
this is why local will stagnate forever
>>
>>8872004
>inclusion of furry datasets
NAI does it though? lol
>>
>>8872004
the jordach graph
>>
File: file.png (23 KB, 386x108)
23 KB PNG
>>8872005
guess what nai doesn't do
>>
>>8872005
yes, and since v4 is known for it's aesthetic, lol
>>
>>8872005
yeah and nai is really ugly now
>>
>nai nai nai nai nai
i'm not subbing kurumuz fuck off
>>
i'm subbed and really happy and my gens are pristine
>>
nothing wrong with a little knotting in the dataset
>>
You're actually trolling right now. I tried subscribing today and discovered NAI is has payment processor issues and nobody can subscribe with any debit card or even PayPal. That company is gonna be toast if they don't fix this, what a shame.
>>
>is has
changed your shitpost mid-sentence?
>>
who wants to see my otoko no ko bugs bunny hentai?
>>
>>8872020
No, seemed like it didn't matter if I left it so I just left it there, but what I'm saying is true, I want to subscribe to NAI but I don't know how to make the payment go through.
>>
>>8872022
me me
>>
File: file.png (45 KB, 916x567)
45 KB PNG
>>8872023
works for m--
huh?
>>
File: image-33.png (361 KB, 499x876)
361 KB PNG
>>8872023
Feel you
>>
>>8872019
>>8872023
>>8872025
>>8872027
lots of payment-related support tickets on the discord, enough that they stopped using the discord thread-ticket system entirely and are redirecting everyone to email support (which is what they should've always done because using discord threads as a support ticket system is incredibly fucking retarded but many such cases exist)
apparently there's nothing they can do about it either because it's entirely on the payment processor's side, maybe one of the literal jeets at chargebee tripped over a cable
>>
>go back to sdxl for the first time in like two weeks
>forgot how much faster it was to gen on
damn
>>
>>8871909
>made it in
:D
>>
ain't no way I'm going back to using adetailer
>>
lol
>>
File: 00041-973222140.png (2.01 MB, 1152x1920)
2.01 MB PNG
People still use adetailer in the year of our lord 2026?
>>
cucks
>>
>>8872041
bwo you're still using the default pony anime style
>>
>>8872048
so?
>>
OWNED
>>
ponyGOD i kneel
>>
>>8872041
why is your roll so fat
>>
anime was real all along :skull:
https://litter.catbox.moe/h6fmj48t699oz9ut.png
>>
anime israel?!
>>
I wish I looked like anime
>>
i am ironanime
>>
File: 00002-2530326438.png (2 MB, 1152x1920)
2 MB PNG
>>8872053
Hyper Roll MkII. Dr. Light's feature creep, you see.

https://www.youtube.com/shorts/HBtH3jg0BCU
>>
pony and wai base style is peak ai gen aesthetic, deal with it, chuddies
>>
>>8872059
trvke, they look nice and smooth unlike the ugly oversharpened vomit that is anima or the fried dogshit of noob vpred
>>
>>8872054
>every drawing has a real life reference
holy fucking artoid skillet cope
imagine projecting your skill issue this hard
>>
>>8872062
it's just a female reposting feminist slop about period blood and abortion
>>
if she breathes...
>>
you can kill it
>>
>>8872066
why does she need to breathe?
>>
>>8872064
>not just an artoid but an art hoe
disgusting
>>
>>8872070
>artoid
obsessed, she has nothing to do with it
>>
Yup, all of those ghost, dragon, alien, and Jesus pictures were drawn from real life reference
>>
churl sisters, our response?
>>
real living bobbleheads were used for reference
>>
>>8872074
even worse then
she's just a regular femoid
>>
>f-word out of nowhere
>>
>>8872079
Had to cart them in because they kept tipping over like weebles
>>
>>8872079
https://en.wikipedia.org/wiki/Hydrocephalus
>>
>>8872085
AYO WTF
>>
>>8871956
yeah, some civit lora
>>
>>
>>8872097
I just noticed the fucked up tail, dang it
>>
>>8872098
Don't worry bro, this thread has 69 other shitty images, no one cares:3
>>
>>8872109
Based participation award anon.
>>
Nevermind though, I forgot noodlekek posted some art earlier, about 65 then.
>>
>>8872109
This is why we can't have middling things
>>
>waaahhh waahhhh boo hoooo images muh images muh pictures!!
Fuck off /hdg/ tourists. Nobody gives a shit. The purpose of this thread is to discuss tech, not jerk your little chud dick. If you want that, go to /trash/ where you belong.
>>
>discuss tech
>>
>>8872126
*smiles and nods agreeingly* mhm.
>>
>>8872126
>The purpose of this thread is to discuss tech
It never was, here is dick central.
there are tech boards to discuss that >>>/g/ldg/ (tdrusell post there and some other interesting people)
>>
i wish the shitpost would at the very least be funny
>>
>>8872126
There's nothing preventing random people from spamming, trolling and dumping random slop. Gatekeeping an unmoderated open forum like this is basically asking for trolling. Nobody's posting gens here in good faith because hdg (this thread) is a shithole with more purity tests than a liberal trans Discord while being shat up and trolled every day.
>>
>>8872138
Good thing we have moderated open forum on another chan! It also turned out to never require said moderation simply because shitposters know beforehand that they will be dealt with unlike here:3
>>
noooo we are hgg it's very different
>>
>>8872140
>because shitposters know beforehand that they will be dealt with
Do they even know about that place to begin with?
>>
>>8872145
Not a big secret. If you've been here for a while, you've probably come across direct links a couple of times, not just mere mentions. And shitposters are 100% thread regulars, I would know!
>>
>621 / 69
/hdt/ - Hentai Diffusion Thread when?
>>
>>8872146
It got me off-guard because I remember the text-spammer asking "another /hdg/? at one point but, again, having direct moderation over there is all the deterement you need to stop them. Be advised tho, that place is going to have site maintenance soon.
>>
>>8871136
I propose doing a mass migration to >>>/g/adt/. At least the moderation there is good, and there’s no need to have two hentai generals here.
/adt/ is currently on life support, and I think an influx of anons from here would really help. It’s also a thread that was originally intended for discussing tech and loras
>>
>>8872154
I’d be happy to post in /adt/. I don’t like hentai, and most of the time I browse 4chan at work or on public transport, so it’s annoying having to be careful reading here because it’s an NSFW thread.
>>
>>8872154
Not a bad idea and you can always catbox/litterbox /h/ stuff.
>>
thanks for your valuable input, /ldg/ creature
>>
>>8872154
Already there, but I'm a sex pest so I don't have much to post there in terms of images.
>>
>>8872161
/ldg/ anon here, this is great! more like this?
>>
>>8872154
True. I haven't genned porn in months anyway as it mostly just gives me the ick. The whole fun of these threads is discussing tech so honestly I think I speak for most people here when I saw moving over to /g/ is probably the best option to help revive these threads.
>>
>>8872154
I miss pre thread #48 /adt/
>>
>>8872162
>>8872158
He is shitposting ldg too, he is a /edg/ creature
>>
bored out of my mind making the same shit over and over as every model is trained on the same garbage

what else is there to do but shitpost? seriously
>>
>making the same shit over and over
creativity issue
>>
File: 20260504090351-3708914218.png (1.66 MB, 1024x1344)
1.66 MB PNG
>>8872170
personally i have like 7 doujin ideas but it would take a long time to actually make them so nah
>>
>>8872172
if you actually made images you would realize there is only so much you can do with these models
after around 200 tb over the past couple years there is nothing left outside autistic controlnet and inpainting bullshit which no one likes doing anyway
>>
>>8872168
No, I’m from /ldg/.
>>8872162
Not me.
>>8872158
Ironically, I found out about your general because an anon reposted a message mentioning /ldg/ and its low number of anime gens about a month ago or so. Otherwise, I wouldn’t have even known you existed.
>>
>>8872174
i made this though >>8871913
>>
if more fetishes were allowed besides vanilla and bbc you'd see alot more creativity
>>
>>8872177
that's bannable
>>
>>8872170
You do the same but better. And there's no ceiling here. Execution >>>>>>>>>> everything else.
>>
>>8872161
MORE. LIKE. THIS!!!!!!
>>
Why is the creature working so early in the morning wtf
>>
>>8872170
i challenge you to gen a girl spanking boy, over the knee, proper anatomy (no long torso on the boy or fucked up legs for the girl), no controlnet. in my experience even anima can't do it. and that's a relatively simple prompt, i tried some more elaborate stuff and afaict it'd only work with controlnets, maybe
>>
>>8872190
>to gen a girl spanking boy
not going to do your weird pedo request
>>
>>8872192
enjoy boredom then
>>
>>8872192
enjoy doing nothing then
>>
lil bro's wee wee got flaccid because anon denied his request
>>
>>8872195
it was flaccid from the start widepeeposad
>>
why is shitacon so repulsive
>>
this thread has a very unique culture
>>
just stop genning coprophagia?
>>
im a big fan of genning [buzzword] and also [buzzword]
>>
kyojiri loli
>>
mhm. *smiles and nods agreeingly*
>>
if this site wasn't fucking dead we would have /ai/ by now
>>
>>8872203
Or the other way around. The site wouldn't be dead if we had our own board.
>>
just sent another suggestion for /ai/ straight into their trash box
>>
Any decent paid online services to generate hentai?

My buddy had some system up for a while but he's reworking other things right now. I don't have the power to be generating images but I do have the money to paypig in the mean time.
>>
why is inapting still so terrible?
forge:
>send image to inpaint with parameters
>none of the actual resolution/sampler/scheduler parameters send
cumstain:
>lol
>>
>>8872212
kri
>>
File: 9708862.jpg (381 KB, 1622x2142)
381 KB JPG
>>8872210
>Any decent paid online services to generate hentai?
none whatsoever
>>
>>8872212
feels ok for me on cumfy with inpapint crop, and I was used to the one from a1111
>>
>>8872215
im finna gonna kms
>>
>>8872212
by lol you mean it just works

>>8872210
novelAI is the only real SAAS for porn, but it's 25$ or nothing cause the other tiers are terrible
otherwise there's sites like civitai, seart, etc that give you access to the same stuff you have on local, though without some specialized tools
>>
I got a single 2070 super, no shot im generating anything I think
>>
>>8872219
I used to gen on a 2060, you will be fine
>>
>>8872220
how long were gen times?
>>
>>8872221
i don't remember but it was a little slow, yeah
patience is the key
>>
>me genning on a 970 on the original LDM model with no xformers or optimizations
kek
>>
>>8872216
comfyui inpainting is crap because you need to drag and undrag a bunch of shit every time
>>
>>8872226
like what? I just mask areas and change prompt
>>
>>8872219
Anima + turbo lora works fine on 1660ti. 70 seconds per image
>>
>>8872203
I don't want an ai board. an ai board is just diluting the whole fucking pot with anything and everything. Probably half or more of all the activity is basically hyper degenerate territory that require board rules like /trash/, /d/ or /b/. And if you have that then good luck having whatever little niche thread you want to make since someone can just come in, drop 150 AI generated scat images and nothing happens because mods are do nothing fagets and "well it's AI generated it belongs here :)"
And with shit like that it's also really helpful to remember that the only reason the /h/ thread became the bastion of discourse it had been was because the rules are because it's a 300 image cap thread, the board is slow enough that text post bump limit is largely meaningless and the board rules forces content to not be full degenerate civit slop.
I think at this point the only real thread I'd even want that would only be able to exist on an AI board is an actual lora repo where the only purpose of it is to post a preview grid and ddl links. The thread would technically be off-topic to /h/, it wouldn't survive on /g/ due to low activity and there isn't anywhere else it could go.
>>
>>8872219
im genning on a 3070 ti. aside from optimizations you may lose out on i think you'll be just fine
>>
>>8872227
switching from empty latent to load image-vae encode, adjusting denoising somewhere on the grid miles away,
>>
>>8872231
make a separate workflow just to inpaint lil bro
>>
>>8872231
>not making an inpainting workflow once and then just reusing it
>>
>>8872231
I keep both workflows in the same file, and just hit "bypass" on the resulting save image node when that branch is not being used.
>>
I don't inpaint, I don't even look at my gen before I post it here
>>
give me an anima inpainting workflow, okay?
>>
>>8872236
we can tell
>>
>>8872237
anima doesnt have inpainting controlnet/baked in, only with very low denoise and on easy areas, or lanpaint (very slow)
>>
>>8872240
>anima doesnt have inpainting controlnet
https://civitai.red/models/2596579/anima-lllite-inpainting
>>
>>8872241
damn
>>
>>8872241
>22 MB
lmao
>>
>>8872241
If it's as good as the other CNs he made, it's not even worth the 20MB download size
>>
>>8872241
those results... aren't great
>>
>>8872244
lmfao just because I was getting excited
>>
>>8872241
https://civitai.red/images/129550734
amazing
this blends worse than just using 1.0 denoise without the CN
>>
just inpaint anima gens with noob?
>>
>>8872241
when will actual good styles be possible with anima? all i ever see is the same i saw with flux which was default anime style. nobody actually likes that shit lol
>>
>>8872249
Was it really necessary to make this comment, you absolute moron? What kind of response were you expecting? You could answer that yourself if you just sat still and thought for a second, faggot
>>
>>8872249
I'm guessing your idea of "good" is very narrow and specific, you should provide an example.Regardless the answer is, it already is capable.
>>
>>8872248
are you stupid
>>
>>8872252
what's the difference? it's just inpainting.
>>
>>8872250
im just shitposting, dont crash out too hard there blud
>>
>>8872254
nta but anima is just much better at inpainting and the vae is way more forgiving
>>
im gonna make a shitty gacha game with anima and get rich
>>
>>8872256
it's better at inpainting even compared to noob w/ an inpainting controlnet?
>>
>>8872259
no, I think these people are inpainting blank background or other garbage like that to be able to use it without controlnet, you never know with these retards
>>
>>8872259
always found the noob inpainting controlnet to be really shit so yeah. no color shifting or inpaint artifacts. there is a bit of a learning curve tho
>>
>>8872261
says the one who can't inpaint without controlnet
>>
File: sdxl-gpu-graph-3.jpg (199 KB, 900x1421)
199 KB JPG
>>8872219
I'll post it again, expected sdxl times for some gpus and more, just add more seconds for naisma
https://chimolog.co/bto-gpu-stable-diffusion-specs/
>>
>>8872263
you can't either, else controlnets wouldn't be a thing, but go on with your bait/stupidity, I will keep giving (you)s out of generosity
>>
Can someone please explain this inpaint CN meme, because I don't get it. You people are seriously loading a whole ass CN and setting denoise to 1 just to fix eyes and clothing details?
>>
>>8872266
some "people" here are that stupid, yes
>>
>>8872266
It acts as a bridge to feed the model context as to what it's filling in.
>>
>>8872264
I might give it a shot. Not feeling up to learning how to set it up TODAY but its good to know that I can. Paypig shit is just a bit limited but $25/mo for right now is sounding fine for a quick GOONIN
>>
>>8872268
>It acts as a bridge to feed the model context as to what it's filling in.
NIgger just write the fucking tags
>>
>>8872270
>model doesn't align seams correctly
Okay, what now?
>>
>>8872266
inpainting cns are great if you want to add or change something seamlessly without having to worry about unreleated things getting mangled too much in the process. like changing a character's outfit. you can do the same without an inpaint model/cn but it's just more tedious. too bad nobody trains inpaint models anymore
>>
>>8872271
you use soft inpainting and a proper set of setting for it
>>
>>8872273
So, what you are describing is:
a. if need be, roughly color in what you want the specific part of the image to look like
b. crop that part of the image and perform a low denoise
c. composite back onto the original
yes?
>>
>>8872266
What do you expect the people that aren't using Novel AI to do? You have to compromise a little bit when using technology which is half a decade old.
>>
>>8872271
Don't use retarded denoise values and increase the padding a little.
>>8872272
That's the one use case I ever loaded those CNs for, but it's not something I can't live without.
>>
>>8872275
Are you some sort of autistic individual that refuses to use custom nodes in comfy or are you simply retarded?
>>
>>8872278
Do show the custom nodes you refer to that make your retarded process more streamlined.
>>
>comfy
oh here we go lol
refu/g/ees please just fuck off forever
>>
>>8872264
>Ayaka XL
why do they always choose the dumbest models for this shit
>>
>>8872281
Ayaka the character, they were using wais
>>
>>8872215
box?
>>
>>8872277
>but it's not something I can't live without.
yeah but it's such a huge time saver for stuff like that. you don't understand how powerful i'd be if we had an anima inpaint model. i could go around making peoples waifus kiss big black men and it'd only take 2 minutes per image instead of 15
>>
>>8872277
Using retarded denoise values is the main use case. Another is outpainting, which by definition has to be 1.0 denoise and a huge mask.

>>8872272
The other option is to draw in some blobs, get the colors close to what they're supposed to be in the end result. Then you can use a lower denoise and do not risk changing other stuff. Whether you find it tedious or appreciate having more control, it's good to have both options available.
>>
>>8872279
>inpainting only the masked area is more retarded than using the whole picture as reference + CN to fix details
You mongoloids deserve to be spammed to death
>>
>>8872285
I was talking about fixing details in an upscaled gen, which is 99% of the inpainting tasks people have to do.
For changing major stuff or outpanting, a CN is definitely a big help.
>>
>>8872286
Why would I allow the model to make up bullshit in a masked area when the CN tells it exactly what that area (and the contextual seams/lines/etc) are?
>>
>>8872275
The technical name for "soft inpainting" is Differential Diffusion. It makes the edges blend better by reducing denoise based on how you feather the inpaint mask. All of it is automated.
>>
>>8872287
>i was talking about fixing details in an upscaled gen
that's little boy inpainting. we are talking about big boy inpainting.
>>
>>8872288
>make up bullshit in a masked area
See >>8872270
>>
why do the people here NOT want the model to know the surrounding information to what its inpainting?
>>
>>8872292
Because I have eyes and the faculty to write a series of danbooru tags that tell the model the surrounding information to what it is inpainting.
>>
File: 1755789235374446.png (2.59 MB, 1344x1728)
2.59 MB PNG
Don't mind me, just posting some henti.
>>
why don't people train inpaint models anymore?
>>
>>8872294
bbcbox?
>>
>>8872294
such a monkey paw that blacked gens are always the most interesting looking and doing the most impressive poses but is blacked shit so its fucking trash
so annoying
>>
>>8872293
You're putting a blindfold around the models proverbial eyes and telling it to draw a hand, for example, in a set area. Surely at a high denoise it'll be able to know how to connect it relative to the arm and clothing present in the rest of the image.
>>
>>8871316
I'm guessing whoever ended up baking reverted the inpaint link on purpose
>>
I am going to crash catbox servers again
>https://files.catbox.moe/1h6okz.png
>>
>>8872295
the only one worth doing that on is anima and that it's too early for that
>>
>>8872294
if i had an inpaint model i could remove the black man and cloth her up no issue
>>
>>8872298
How is that impressive
>>
>>8872283
same prompt, just changed the model, keep in mind it only works on 1% of the seeds though, Im sure it can be made better with modifications/ prompt scheduling but I aint bothering with it
https://files.catbox.moe/s91i0b.png
>>
>>8872294
if i had an inpaint model i could remove the man, add cum and make her a bobblehead no issue
>>
>>8872301
>naiXLVpred102d_custom
get with the times old man
>>
>>8872304
prompt that exact pose and post the result

ill even be nice and give you an entire hour to do so
>>
>>8872307
I have post nothing but SDXL gens for the past 3 weeks or so
>https://files.catbox.moe/mdit6s.png
>>
>>8872308
nta but i have a feeling anima could easily do that pretty consistently
>>
>>8872294
how do you prompt that sex position where the guy is turned away?
>>
>>8872312
roughly 55 minutes left
>>
>>8872308
Nigger I am the one who posted it
>>8872297
https://litter.catbox.moe/pf3nl47ysidrtbo3.png
>>8872313
In this case, luck. But you can do some NL work to force the guy to show you his balls and ass.
>>
>https://files.catbox.moe/ljdwp7.png
>>
>>8872314
i'm training right now do it yourself
>>8872315
kek this. the most annoying thing is when it insists on drawing the dude's anus but it's easy to mitigate. ty for box
>>
>https://files.catbox.moe/n6y1rb.png
All of those are canon btw
>>
>>8872315
keep moving those goalposts, 51 minutes left.
>>
>>
https://files.catbox.moe/5w4tps.png
>>
>>8872321
Suzuka is supposed to be flatter
>>
>>8872322
you are right, my bad

>https://files.catbox.moe/zj4bxk.png
>>
why is he spamming this shit again what set the creature off this time
>>
>>8872301
>>8872310
>>8872316
>>8872318
>>8872321
>>8872323
So apes are fine but you are forced to catbox the wholesome doggos and majestic horses?..
>>
>>8872325
his mom left for work and his uncle came to visit yet again
>>
>>8872325
noodlekek is just pathetic like that
>>
>>8872325
I am allowing your eyes to see my carefully crafted horsed gens
>https://files.catbox.moe/x8kzxf.png

>>8872326
cuck website after all
>>
rather see a dog dick than a black dick, am i the only one?
>>
>>8872330
No, there are many other whites here
>>
>>8872330
Depends, is it pov?
>>
This one is the best as far as dogs involved: n6y1rb.png
Saved all the horse ones, gonna cherish them forever.
>>
>>8872332
only if its female pov
>>
Still I think if the receiving party is also a horse (female) I wouldn't mind it either.
>>
>>8872335
say sake right now before it's too late
>>
>>8872336
rice wise?
anyway he probably meant uma not actual horse pussy
>>
>>8872335
We need futa umamusumes impregnating mares right fucking nyow
>>
>>8872337
No I didn't mean umas...
>>
>>8872337
>anyway he probably meant uma not actual horse pussy
I really hope so
>https://files.catbox.moe/90j64x.png
>>
hard to beat mare pussy, shit is so cash
>>
>>8872342
No I didn't mean umas how many times do I have to repeat? But thanks for another horse gen.
>>
>https://files.catbox.moe/wh5m7a.png
>>
seems more gay to fetishize horse cock than horse pussy tbdesu
>>
>>8872342
what style? also catbox actually working for more than an image or two is impressive
>>
>>8872346
why are you focusing on the dick in the porn image lil bro?
>>
If you do the horse on horse thing, there has to be lots of juices and the motion lines should depict significant force. And NO ponies. REAL horses. Like I said: grand, imposing, regal, majestic.
>>
>>
File: 00113-4244874736.jpg (712 KB, 1664x2432)
712 KB JPG
Well that's it for now, show is over idiots, go home or I'll force you to do it posting futa on female or cute boys next

>>8872347
>what style?
the same as the other horse gens
>>
>>8872353
>the same as the other horse gens
oh clicked on that one and assumed none of them had meta. thanks
>>
>>8872350
>no boy
>not even a drop of cum
>>>/e/
>>
>>8872269
>Paypig shit is just a bit limited
in what way?
you don't need any of the bandaid shit local needs
>>
>>8872266
With good 1 denoise inpainting you can fix huge chunks of image like hands, backgrounds, faces, or add new parts to the image without having to sketch anything. You can edit art and other gens without knowing prompt and it will blend well. You can even mask most of the image and leave some small part (if you really liked the expression but want a different rest of the image for example).
It's insanely convenient, but localfags will deny how good it is until they get it, like they always do. I literally don't bother inpainting with anima when I have nai inpaint.
>>
>>8872357
Toys are forbidden on /e/, you'll have to deal with these 1girl gens here!
>>
>>8872360
How do you have nai inpaint if I also have nai inpaint and I'm supposedly the only one in this thread with NAI???
>>
File: 1775939533830622.png (2.7 MB, 1344x1728)
2.7 MB PNG
>>
>>8872363
I haven't posted images here for more than 2 years
>>
noodlekek has to be trans
no one else acts like this
>>
>>8872366
Oh, the asanagi thing, OK. Hello there.
>>
File: rikka_angry_089.png (944 KB, 896x1152)
944 KB PNG
>>
File: rika_angry_093.png (987 KB, 896x1152)
987 KB PNG
>>
File: rika_angry_cen_095.jpg (221 KB, 896x1152)
221 KB JPG
>>
>>8872368
Don't know who that is, I only post on pixiv
>>
>>8872372
oh thank god
>>
>>8872357
toys are /h/, retardbro
>>
Almost started a pixiv the other day, but couldn't come up with a nice name for account so it didn't happen.
>>
artist mixing
yes or no??
>>
>>8872377
cum mixing from two boys inside of me yes
>>
File: ComfyUI_25982_.jpg (360 KB, 1024x1536)
360 KB JPG
mmmm fox sex
>>
>>8872379
guve me your metadata
>>
So noodlekek is fine with being a faggot and a cuck, but I ask him politely to have 2 horses having sex and suddenly he is too insecure for that and runs away???
>>
>>8872381
lmao what is your play here cocksucker
>>
>>8872365
more like this?
>>
>>8872381
its noodleMAAM
>>
>>8872365
hello sir. can i ask you a question?
>>
>810 / 81
>>
File: 1774769598481199.png (2.63 MB, 1824x1248)
2.63 MB PNG
>>
>>8872392
guve me your metadata too
>>
>>8872393
No, Slave Knight Gael
>>
>>8872393
It's right there. Do you guys really not use the browser extension and stealth metadata? Using the extension to see if the image has metadata is the first thing I do when I see a gen I like.
>>
>>8872400
its my way of showing appreciation
>>
>>8872400
nta, same. But it's rare enough that I'm not surprised people don't think to try it unless it has a NAI filename
>>
>>8872392
Do you really like the shiny ai-generated style of anima-rl?:(
Not like anima requires any help being nudged towards ai-generated, but anima-rl really seals the deal.
>>
OK horse sex is completely broken on noob, I get it why noodlekek refused. Impossible to make it look good.
>>
Should I even try with anima?..
>>
>>8872407
Call me blind but I don't see how that's similar to stuff like WAI or AOM. If anything, it gets the style much closer to what I remember of the artist and how NAI did him back when I was subbed pre-Anima. Not a fan of ring highlights but oh well, they aren't a tag so I can't get rid of them easily.
>>
>>8872407
nta but anima-rl helps a ton with anatomy and composition. and a bit of shiny slop here and there is good too
>>
>>8872411
I mean the blatant AI gradients? Look at her boobs. Surprisingly the male is actually drawn in a proper style LOL.
>>
>>8872413
Actually before I spoke I should have checked how yunsang draws... but yes, no AI gradients, his works look 50 times better.
>>
File: 1754928010792841.jpg (692 KB, 2009x1331)
692 KB JPG
>>8872413
I mean that's pretty much how he draws, and his style is kind of the role model for the "modern gacha art" style I'm aiming for among all the other artists I've tried.
>>
>>8872413
>>8872415
you're annoying
>>
>>8872416
It's not even close though? Anima is 10% there. I'm surprised people still don't recognize generic AI gradients, they are so fucking awful. No artists draws like that besides those that use AI in stealth.
>>
>>8872417
Gonna cry?
>>
File: 1765875958824185.jpg (1.15 MB, 4000x3683)
1.15 MB JPG
>>8872418
IDK anon, to me it looks mostly as it should. If anything, the shading of the skin looks not detailed enough, but the gist is there.
>>
>>8872416
In fact looking at your gen, anima switches between AI gradients (booba) and some digital painting type shit with photoshop brush strokes (her arm and clothes and the 1boy as a whole), but non of those are present on the artist image you uploaded.
>>8872421
Yes there is actual skin texture, and there is 0 indication of it in your gens.
>>
>>8872416
>>8872421
And don't get me started on the hair highlights of the artist and... this >>8872392 (horrible)
>>
>>8872421
desu that gen looks nothing like Yunsang. still looks good tho.
>>
>>8872392
Broken eyes too, which I noticed anima-rl is a LARGE contributor to, no matter the style.
>>
>>8872422
>>8872423
>>8872425
Non-RL gens look too flat to me. This artist was straight ass until preview3 too, like basically not even remotely recognizable.
>>
>>8872427
>Non-RL gens look too flat to me
Maybe because you are used to dogshit like WAI
>>
File: 1746957236285622.png (1.11 MB, 832x1216)
1.11 MB PNG
>>
hyper is so fucking based no idea why more people dont do it
>>
>>8872428
That artist doesn't look flat at all though. Which means getting away from flatness is the same as getting closer to the artist's style.
>>
>>8872431
Or in your case getting closer to WAI because the shading looks (AI generated:1.67) and nothing like the artist.
>>
>>8872429
insert a straw into his urethra and make her drink out of it
>>
File: 1770434965899850.png (2.52 MB, 1248x1824)
2.52 MB PNG
>>8872432
Ok, here's an actual WAI gen of the same artist. I really don't see how the previous Anima gens are closer to this than to the actual artist.
>>
File: 1770776219324234.png (2.33 MB, 1280x1728)
2.33 MB PNG
>>
>>8872436
Wai do you have WAI on your PC?
>>
>>8872438
I grabbed an old gen from half a year ago.
>>
>>8872437
Penises look weird as fuck when they are thickest in the center. Idk why some artists draw them like that.
>>
>>8872437
her age and physical age?
>>
>>8872436
Both are about equally bad I would say. Anima-rl is winning slightly over WAI, but only because it switches to that digital brush texture sometimes so not the WHOLE picture looks dogshit.
>>
>>8872442
>her age
1000+ years
>physical age
Chinese in her 30s
>>
>>8872433
https://litter.catbox.moe/7437v1umoh8782ai.png
>>
>>8872445
that's his testicle tube. she sucked the nuts outs of the sack.
>>
>>8872429
she's underage you sick fuck
>>
>>8872444
11
11
>>
>>8872451
you'll also get 11 years on rikers island for having csam on your computer
>>
Bro, I'm putting feral in pos and anthro in negs (negpip) and these horses still have these stupid human-like torogao faces while having sex. Noob sux.
>>
>>8872451
67
67
>>
File: 1769283683950335.png (2.52 MB, 1824x1248)
2.52 MB PNG
>>8872443
Alright, here's the same everything minus the RL LoRA. Unless I really am blind, the shading is basically the same but it has worse anatomy, less detail and didn't follow the intended composition at all.
>>
Like the female horse looks back at the male horse slily while being fucked. Yes, VERY convincing...
>>
File: 1764924066955940.png (2.61 MB, 1824x1248)
2.61 MB PNG
>>8872457
And a slightly cherrypicked (second reroll lol) example that did get the composition right.
>>
>>8872457
Gradients notwithstanding, it's much better than RL, look at the eyes, they almost look fine. Shitty hair highlights are toned down to acceptable level. Still just as much not even close to the artist but you'll just have to bake an actual proper artist lora for that anyway.
>>
>>8872460
>you'll just have to bake an actual proper artist lora for that anyway
No way I'm wasting so much of my time and effort on that which the model should be doing OOB.
>>
(and yeah, it's still not the style I would ever gen with, rl or not, mainly because it's more average ai style than the actual artist, but you do you)
>>
the noodle has spoken. now take your trash and go
>>
https://litter.catbox.moe/6j8zbo4p5xwvpdwg.png
>>
>>8872467
Noodlekek hasn't spoken shit, he's been quiet in his cuck corner ever since he failed to generate me some nice stallion-on-mare action.
>>
yunsang lora when
>>
I put high quality image in the prompt and it solved all my problems
>>
>>8872473
10 days ago
https://civitai.red/models/2569402/
>>
>>8872472
nigga just search a national geographic video on YT
>>
>>8872476
>gook
>blacked shit
every single time
>>
>>8872476
Ok now post a real lora baked by a white person.
>>
>>8872477
Listen up you little shit, when you are genning your stupid anime 1boys and anime 1girls I'm not telling to go watch pornhub instead????
>>
>>8872479
white people make extracts from full finetunes
>>
>>8872481
I'm pretty sure chadmixfaggot is brown
>>
>>8872476
Those are some horrendous preview images.
>>
>>8872483
it's civitai
>>
>>8872477
ngl little brody but he pwned you real bad here >>8872480
>>
File: spade.png (929 KB, 693x847)
929 KB PNG
>>8872476
why are civitjeets like this
>>
>>8872480
why the fuck would you like to see something like that on ANIME style you troglodyte
>>
>>8872486
it's korean culture
>>
it seems the "SEA people" are waking up
>>
>>8872486
>why are civitjeets like this
sir that's /hgg/ culture
>>
File: 1767418389962266.png (1.3 MB, 1216x832)
1.3 MB PNG
>>8872476
>Hindy features completely unprompted
>bunch of utterly nonsensical features out of nowhere
Yeah nah, I'm going back to no LoRA.
>>
how long till the next catGODDESS post?
>>
>>8872493
the skin texture is so much better, however
>>
File: 1770919029456167.png (2.88 MB, 1824x1248)
2.88 MB PNG
>>8872493
And no, Hires. fix doesn't make it much better.
>>
does this nigger have to post every single failgen he got from some random jeetvitai lora
>>
it's the spammer
>>
better than (You) posting no gens
>>
https://litter.catbox.moe/7v0xfm.png don't like the size difference
>>
>it's the [headcanon]
>>
https://litter.catbox.moe/oy658s.png
This one's okay I guess. I'll settle for this for now.
>>
>>8872503
>>8872505
pretty good thanks for catobx
>>
what keeps (you) coming back to this i can't believe it's not feces general
>>
>>8872506
Sorry there's no meta but it's just noodlekek's style from his previous boxes.
>>
>>8872503
>>8872505
what the FUCK
>>
>>8872507
shitposting
>>
>>8872510
its nature chud
>>
>>8872510
If you really want to know how this was generated so much you can just ask you know??
Here are the tags after some 30-minutes e621 STUDYING:

(@xerbatt:0.8), (@quasarcake:0.0), (@sho \(sho lwlw\):0.9),
2others, (fur,:1.1) horse, large female, feral on feral, from side, equine penis, equine pussy, equine genitalia, penetration, hooves, male/female, male penetrating female, meadow, uncensored, motion lines, sex, (female penetrated, genital fluids,:1.3) (size difference, anthro,:-1) feral,
[:masterpiece, newest:0.3],
>>
I didn't ASK for anything you fucking fuckhead
>>
>>8872516
>after some 30-minutes e621 STUDYING
but how?? i thought anime didn't have the e621!!
>>
>>8872519
noodleboxes were noob
>>
>>8872520
then whys xhe @iing
>>
>>8872521
Oh shit I forgot to switch those back after testing on anima (mix works on anima btw). Well it still worked I guess.
>>
>>8872522
yeah no worries.
>>
File: file.png (1.14 MB, 1812x1046)
1.14 MB PNG
>>8872228
>1660ti
>70 seconds per image
what
>>
i use arch btw
>>
>>8872525
my thoughts and prayers towards you bro
>>
>>8872525
So is this genned on the CPU or the integrated card?
>>
I use windows 11 including sending optional data to microsoft for analytics
>>
>>8872528
it uses rocm docker image, so it's rocm device (integrated gpu presumably?)
>>8872527
i have a home lab with 3090+3060 in it bro
>>
>>8872533
>i have a home lab with 3090+3060 in it
send me your thoughts and prayers then, bro
>>
do you also have a pair of those cute programing socks too?
>>
>>8872533
Google AI is telling me rocm doesn't support integrated graphics cards so that's probably running on CPU cores.
>>
>>8872540
https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/index.html#expanded-platform-support
rocm docs say some integrated graphics are supported
>>
>>8872300
no they just don't care enough to use a template
>>
that DOES IT! I'VE HAD IT WITH THE SHITPOSTING! Makes me SO mad that now i want to train a lora on kirushi/kirsi engine! FUCK!!!!
>>
the faggot is the most disgusting shitposter this place ever had
>>
>>8872545
the other changes made it into this OP
>>
File: file.png (75 KB, 877x426)
75 KB PNG
>>8872544
oh well
>>
>>8872553
which ones?
>>
I have THREE nai subs
>>
id pay 50 a month for novelai if i could
its that good
>>
>>8872481
nobody has done this since noob was the norm
>>
>>8872558
Can you give one to noodlekek? It's fucking sad seeing him still use noob...
>>
File: ComfyUI_25962_.png (1.38 MB, 1536x1024)
1.38 MB PNG
tuggin'
>>
I'm just glad that vidgen ended up in the trash where it belongs, just like I said
it's fucking awful and local won't see anything like cloud vidgen for a long time, especially for hentai
>>
>>8872568
nice tits bro
>>
>>8872572
>for a long time
Or forever, considering "you will own nothing" is rapidly becoming a thing and making PC parts for consumers is cringe and out of fashion.
>>
>>8872575
Yep. Basically eveyone has been kicking themselves for not keeping everything they make proprietary and only available through a subscription.
>>
>>8872556
oh yeah nvm
>>
hello? where's the hentai?
>>
sir this is a thread for technical discussion and lora baking/sharing
>>
File: 00156-877151777.png (2.14 MB, 1248x1824)
2.14 MB PNG
here's your hentai bro
>>
>>8872568
the most prominent features of this picture are all men, this isn't /h/
>>
This artist broiler does some cute stuff when he isn't busy drawing guro!
>>
bestiality isn't /h/
>>
>>8872588
agreed but you could have said the same with a picture
>>
>>8872591
Real art is off-topic thooo
>>
>>8872592
who cares
>>
>>8872592
I meant a gen using the artist
>>
>>8872594
Gladly, but (You) bake a lora first.
>>
>>8872575
Nah, not forever for sure but it could be something like 10 years to see anything close to sora on prosumer local hardware. And sora isn't even available anymore, kek.
>>
>>8872595
Hmm yeah, 200 booru pics is usually more than enough for anima or noob. But his style seems to be all over the place, and it didn't train well.
>>
its boring pedo shit :skull:
>>
jav lora first
>>
>>8872301
any survivors?
>>
>>8872365
lilniggabox?
>>
why should i post my beautiful images in this thread if there's no highlights for it to land in
>>
File: mikotry5.png (1.24 MB, 1024x1024)
1.24 MB PNG
>be big lady enjoyer
>favourite big lady artist puts dicks on 85% of his work and fucking ruins it
>https://danbooru.donmai.us/posts?tags=mikoyan+&z=5
>doesn't take commissions either so can't commission normal big ladies
>Fuck it, lets try AI
>download one
>https://civitai.com/models/241844/galena-cat-galena-citron-anime-treasure-illustrious-and-noobai
>Do some reading
>build dataset
>train LoRA
>Pic rel is the best i can get it

How do you guys get such good gens? No matter how strong/weak i make the LoRA strength/clip and no matter what sampler i use the eyes and mouth are almost always blurry, abysmal dogshit
>>
>>8872412
RDBT, or whatever it's called, cfg distilled at a low weight improves all anatomy and doesn't introduce any slop issues in my experience. Just use it with regular cfg and steps.
>>
>>8872608
>RDBT
what now?
>>
>>8872241
I tried using this, images have this faint reddish color on where you put the masks but connects kinda well, does anyone have a normal non CN version of anima inpaint that works with high denoise?
>>
>>8872607
i like mikoyan but his art style is very varied
id need like 3 loras for him
>>
>>8872607
i can't tell if this is a shitpost or not
>>
>use anima character lora
>style lora goes to shit
>prompts stop working properly
>remove character lora
>style looks accurate
>prompts work great
holy shit loras are outdated copium that need to be replaced asap. also anima is way too small and dumb for natural language, sdxl is still smarter in many ways.
>>
>>8872613
>too small and dumb for natural language
omega skill issue
>>
>>8872613
based 64dim lora
>>
>>8872613
>sdxl is still smarter in many ways
keep using WAI sar
>>
clip x clip incest > clip & qwen
>>
File: 144112103_p8_master1200.jpg (672 KB, 822x1200)
672 KB JPG
>>8872613
truth. this pose is impossible on anima, it's actually quite terrible at tags too. i guess it's nice if you like text on signs
>>
nai doesn't have these issues
>>
>>8872613
You act like the problem is the model and probably not the overcooked character lora.
>>
>this melted piece of slop isn't possible on anima
good, it shouldn't be.
>>
>>8872613
>character loras
Local is disgusting...
>>
imagine not being able to just type the characters name and get a perfect rendering of your character
>>
>>8872620
because novelai was actually intelligently designed from the ground up unlike anima which throws in a bunch of joycaption 'natural language' sloppa on top of a model designed for generating street signs
>>
File: mikotry19.png (1.13 MB, 1024x1024)
1.13 MB PNG
>>8872612
Its not a shitpost, im just a massive newfag. I have no fucking clue what im doing.
>>
>>8872620
of course not. nai can't do style in the first place.
>>
>>8872627
fair enough
>>
i miss e621 so damn much bros...
i can't get that hint of primal sexiness on anima...
>>
>>8872630
train a lora on it retard
>>
just... write
>>
>>8872625
why does local still cope with the most awful of base model choices?
>>
>>8872609
>https://civitai.com/models/2364703?modelVersionId=2848821
Keep it under 0.4 and just use your regular steps/cfg. I use ER SDE/DPM++ 2M SDE and SGM if it means anything. Make sure it's not the dmd2 you download.
>>
>>8872604
yes, sadly all of them are still up
>>
lolcal
>>
>>8872607
1. install comfyui (portable version)
install comfyui manager for extensions
2. download the official anima workflow
3. follow the instructions in the workflow to get it working (you have to download 3 files, put them in the folders the workflow tells you to)
4. use @mikoyan or (@mikoyan: <weight) -- for example, (@mikoyan: 1.0) to add him and other artists you might like to your prompt. 1.0 is the default weight. escape ( and ) like \( and \), also escape \. and \: in tags such as artist names or series like nier\:automata
5. you can optionally download anima loras (RL lora, turbo lora, others maybe)

you can find a guide on how to install comfyui on >>>/g/adt, i think the official workflow for anima is this: https://www.comfy.org/workflows/image_anima_preview-cc053bb55df6/ and you should download the anima files from here: https://huggingface.co/circlestone-labs/Anima/tree/main/split_files (preview 3 is the latest version). you can get anima loras on civitai.red, you have to make an account and search for them (RL lora and turbo lora are from the same person who's making the model so they're higher quality)
>>
>>8872642
oops forgot to make comfyui manager install step 2 but you get the idea, i can't count
>>
>1. install comfyui (portable version)
LMAO
>>
1. grab a pen
>>
>>8872642
also read up the hugging face page on anima (https://huggingface.co/circlestone-labs/Anima), it tells you how to prompt it
>>
2. push pen in eye socket
>>
instructions unclear it ended up my urethra
>>
>>8872642
>you can find a guide on how to install comfyui on >>>/g/adt
it was good until this part, then it became too obvious
>>
>>8872642
Thanks anon, I'll give it a go
>>
https://litter.catbox.moe/t9ueuc.jpg
>>
rip.. he got him
>>
comfy just got billions in venture capital money and you are still doubting him
>>
Which trainer and config are we using for Anima loras chat?
>>
>>8872654
kohya sd scripts and a powershell file baby
>>
request anchor
>>
i'm going to post request anchor at the start of every new thread then samefag completing some requests until hgg is renamed back to hdg
>>
what a cuck
>>
finally an honest idea to boost engagement around here, i'll also link the anchor to /g/
>>
File: tmp0o8pcztk.png (2.7 MB, 1632x1632)
2.7 MB PNG
>>8872658
Remake this but better.
>>
what's good negative prompt for anima?
>>
File: 00005-473367591.png (1.32 MB, 1248x1824)
1.32 MB PNG
>>
>>8872668
bad image
>>
>>8872668
noodle, bobblehead, oekaki, loli, child,
>>
mindbroken general
>>
what have you been using to train your pony v7 loras
>>
my GPO and my CES
>>
is there a way to inpaint with a fixed-resolution context around the masked area? when inpainting a hand or background character on a 2k+ generation, it fails because anima doesn't support above 1536x1536 total pixels. i'd like to just use the bounding box of the mask and expand it to 1024x context and then just reapply it
>>
File: 00009-3358081337.png (1.32 MB, 1344x1728)
1.32 MB PNG
thanks for the box earlier anon. neat pose
>>
>>8872679
i wish the man took up more of the image or if you could see his taint
>>
>>8872671
these go to positive though
>>
/hgg/ shitposter thoughts on this? https://civitai.red/models/2495369/kirazuri-anima
>>
>>8872642
Thanks didn't know of >>>/g/adt/ is a really good thread
>>
File: anima_00536.jpg (1.84 MB, 1536x2688)
1.84 MB JPG
>>
File: 00002-1331925772.png (1.44 MB, 1344x1728)
1.44 MB PNG
>>8872680
be the change you want to see. inpaint it
>>
>>8872685
requesting ryuko getting fucked by that ugly fat father
>>
>>8872607
Mikoyan works fine without a lora, i tried a mix with it a few months ago and noticed that it tends to overpower other artist tags. ( ͡~ ͜ʖ ͡°)
>>
>>8872682
a new finetune and no one cares
what a decline
>>
>>8872682
>A full finetune of the Anima preview3-base
Based oil baron.
>>
>>8872687
some artists are really strong like mikoyan and cutesexyrobutts, i guess it depends on style mostly?
>>
>Minaba Hideo doesn't work on Anima
>>
>>8872690
and where is your image, fucking retard??
>>
File: 00009-1331925772.png (1.96 MB, 1344x1728)
1.96 MB PNG
>>8872686
ugly? don't do my man barazo like that
>>
File: 00006-1331925772.png (1.48 MB, 1344x1728)
1.48 MB PNG
>>
>>8872694
I don't know what causes it besides the loras used in the shitmixes; artists like laserflip, thirty 8ght, fishine, black beat, riz, blushyspicy and others that i am missing right now do that too, regardless of art count
>>
>>8872697
no u
>>
Man noob mogs anima so hard when it comes to style
>>
>Man [fully trained model] mogs [preview of a model] so hard when it comes to style
>>
>implying it would do better if it was fully trained
anon...
>>
not this shit again
>>
how come noob only took 2 months to fully train, but anima is taking 4 months of only previews??
>>
File: 1753261316096873.png (158 KB, 680x330)
158 KB PNG
>[inherent flaw in the model] will be fixed once he trains it on more images
>>
uhhh kirazuri-anima gens?
>>
this but unironically
>>
this but nai
>>
nais :DD
>>
nais >:(
>>
I wish you could just mix noob and anima together and get a model that does style as good as noob and overall prompting as good as anima
>>
File: 1771864248268566.png (1.19 MB, 1152x1408)
1.19 MB PNG
>>
>>8872733
it's called wai anima, chuddie
best of both worlds
>>
>>8872735
>all of the shit prompting from noob
>all of the shit inconsistent artstyle from anima
>plus baby oil
>>
it's called localgrease
>>
This is a touchdown russell general where we LET RUSS COOK. Brown hands on keyboards and touch screens must do the needful and leave.
>>
pony ahh shit
>>
this?
this isn't /ldg/, lil bro
>>
this?
so much this
>>
Can you faggots stop being gay and ironic for a second
>>
ok fine
...second's over!
>>
haha
gottem
>>
File: 00086-3925437374.png (3.78 MB, 1792x2304)
3.78 MB PNG
>>
File: 1763862440547036.png (1.33 MB, 1152x1408)
1.33 MB PNG
>>
chad ai artist vs virgin jeet slopper
>>
both are brown
>>
good morning all. butiful day over here in india.
>>
>>8872756
Good morning bloody benchod (ง ͠° ͟ل͜ ͡°)ง
>>
why is every image here a shitpost
>>
>>8872764
constant shitposting and bullying loliposters killed /hdg/ and it's derivatives
>>
>>8872765
yet another shitpost
>>
File: 00010-2805185524.png (1.17 MB, 1248x1824)
1.17 MB PNG
>>8872750
nice urethra
>>
>>8872766
but there's no image
>>
>>8872769
his words genned an image in my mind
>>
>>8872770
that should be your next gen
>>
>>8872771
im a virgin loser i'll never create my next generation
>>
was anima a flop?
>fried harder than vpred
>styles are only half-accurate
>poor lora trainability
>many characters not learned
>dumber than sdxl as basic nsfw poses
>qwen te too small for natural language
>>
post kirazuri-anima gens
>>
hmm nyo
>>
>>8872775
it's dogshit
>>
wai v17 is better
>>
losers
>>
Can we collectively make a curated set of rules for /hgg/ and when there's an infraction, mark the post and inform a 4channel janitor of it so there is consequence for our posting habits?
>>
>>8872774
>>styles are only half-accurate
Thiccwithaq works fine. That's all I need.
>>
many say that
>>
File: Acrobat.png (1.53 MB, 832x1216)
1.53 MB PNG
>>8872764
cuz entropy is a bitch anon
>>
>>8872733
It's called chenkin noob
>>
>be chinese
>name your model chinky’ noob
hehe, based
>>
kek
>>
geg
>>
gegebox?
>>
File: 00013-4126178853.png (2.4 MB, 1344x1728)
2.4 MB PNG
thoughts on foreskin?
>>
>>8872774
anima is the exact opposite of everything you just said so unsure if troll or enormous skill issue
>>
is chat jew?
>>
>>8872795
i like foreskin (i am straight) but not this image
>>
>i like foreskin
>i am straight
and I use arch btw
>>
>>8872796
just disingenuous
>>
foreskin is the enemy of the jew
>>
why? how?
>>
>>8872795
can you make the heads bobble?
>>
>>8872795
should be default without having to prompt for it
>>
>>8872804
truth nuke
>>
>>8872804
this plus smegma
>>
>>8872795
This is the lora I was waiting for. Upload it please thanks.
>>
>>8872694
>>8872707
Back on illu it used to depends on how far the style was from the "danbooru average". I'm not sure how it works now that the model has DA and ye-pop included, and now that it's so much better at separating styles without leakage. But the same ones seem to pop out as being overly strong.
>>
I have intact foreskin and it almost fully retracts when hard. I think this overhanging foreskin thing is only a thing for small dick lil' bros.
>>
>>8872812
you can just prompt for it

>>8872804
should be the default for JP artists, not like they do circumcision over there so it's really bizarre
>>
One guy spams /e/ for WEEKS non-stop, with 10-20 second intervals, doesn't even try to hide his spam, and has never been banned. Instead, you get banned for REPORTING him there.
Why can't they fucking delete that board already? It's literally worse than useless now. There's less moderation than on /trash/.
>>
go back melty
>>
>>8872816
Just post the /e/ here like in the olden days.
>>
>>
>>8872814
I think it's a case of artists without foreskin fucking up basic anatomy.
>>
>>8872816
the random words in all caps make me imagine you as a large black woman clapping
>>
File: ComfyUI_27908_.png (1.74 MB, 1120x1520)
1.74 MB PNG
>>
>>8872815
>you can just prompt for it
Artist, not the foreskin..............
>>
>>8872819
this place isn't any different. if the spammer decides to shit up hgg instead of edg, nothing's gonna stop him
>>
>>8872826
except ia schizo isn't here, so there's no reason to shit the place up
>>
>>8872827
I'm not entirely sure if he's actually spamming due to ia schizo or not. It might be an excuse because another day he started spamming because anons "weren't friendly" or some bullshit like that
also it's not like he's the only dude who spammed diffusion threads. We had spammers before doing it for months. Although they were a bit more subtle
>>
>>8872826
It is currently different in that he is not spamming here. I assumed your goal was to enjoy an AI thread, not fight against the schizo.
>>
>>8872828
perhaps you and the other retarded /edg/ fags should start to consider the possibility it's more than one person
>>
>>8872828
It's also not /e/'s first rodeo. I think this guys is the usual spammer. Sometimes he's able to afford his meds and his depression lets up a little bit, then he goes on a posting spree. There's not that many people sad enough to spend their days spamming 4chan.
>>
>>8872828
>another day he started spamming because anons "weren't friendly"
>Me when I make shit up
>>
File: ComfyUI_27911_.png (1.35 MB, 1120x1520)
1.35 MB PNG
>>
>>8872835
so he's the hero in this spooky campfire story?
>>
/edg/ drama queens bringing their stupid shit to other generals again
>>
why do all you fully grown adult men act like this?
>>
File: 9Popoe3SZEaa5uETiaAC6.jpg (151 KB, 1152x2016)
151 KB JPG
>>8872835
That's not even me lol. You really are mentally ill and can't accept the fact it's not all one person
>>
>>8872840
shitposters went to sleep so now we're stuck with general gossip
>>8872841
he's not mentally ill, just a little dim
>>
>>8872837
to be fair he moved to /e/ from our place. It's the same pattern from last year, wildcard trash pics with randomized resolutions and randomized filenames matching all common UI outputs.

>>8872831
very subtle
>>
>>8872843
>to be fair he moved to /e/ from our place
>our place
go back. ia schizo is in /hdg/ so you can whine there
>>
what does the ia in ia schizo mean?
>>
>>8872841
Wait, ElectroWorld posts here?? Holy… i’ve been a tier3 sub for 6 months now. If possible, I’d like to make a humble request. Could you do some nuthyper/sackflation content? Basically it’s like regular hyper except only for the balls, while the actual shaft remains within realistic proportions. thanks, love your stuff!
>>
>>8872845
vocaloid ia. Dude only posts that character.
>>
>>8872846
kek
>>
File: ComfyUI_27919_.png (1.42 MB, 1120x1520)
1.42 MB PNG
>>
just post one kirazuri-anima gen
>>
PLEASE!
>>
just one is enough
>>
do it yourself faggy, i'm too busy genning kirazuri kino
>>
can someone be the man who stepped up
>>
come on dude
>>
About to try training a real person lora on Anima. Since you can make photoreal cosplay of anime girls, I want to see if the reverse would also work and turn a real person into flat color, or chibi. Any thoughts on tagging to minimize "style" bias? Was about to go with keep_tokens: 2, "photo of a young girl, ye-pop". Trying to avoid "photorealistic" since it's a booru tag meant for art not 3Dp.
>>
>>8872850
doesn't seem to be a booru tag
>>
File: 1762887798701863.png (1.96 MB, 1152x1408)
1.96 MB PNG
>>
https://civitai.red/models/2495369/kirazuri-anima
you NEED to try Kirazuri Anima and post gens of it
>>
>>8872861
only if you do it first
>>
>>8872863
be a man and take the initiative.
>>
File: ComfyUI_27938_.png (1.48 MB, 1120x1520)
1.48 MB PNG
>>
>>8872864
i'm being a gentleman and letting the woman go first
>>
>>8872858
>About to try training a real person lora on Anima
Hasumi Kurea JAV lora?
>>
>>8872861
I guess the main draw here is being able to use FOTM gacha sluts without a lora. I'm not ready to give up low imagecount styles though, and as usual it's an aesthetic tune smearing AI grease all over them.
>>
>>8872867
ever seen a gentleman in porn? no, because that's not what women want.
>>
>>8872869
yes bro this is clearly the same as wai anima
>>
>>8872868
Do people actually like JAVs irl? I thought that was just a last resort for cucked japanese not having access to our stuff. Even before the censorship shit, all the girls look kinda samey and without any personality.
>>
>>8872877
yes, they do. if you can't understand the appeal, stfu
>>
>>8872878
i say this about [my fetish]
>>
>>8872877
jav is the only acceptable form of 3dpd
>>
>>8872877
if jav was uncensored i probably would barely consume western irl porn at all
>>
>>8872877
if jav was uncensored and they got rid of those repulsive fucking moans i probably would barely consume western irl porn at all
>>
>>8872879
don't we all
or most of us I guess, don't exactly learn tolerance and to each his own amongst autists and schizos
>>
skill issue
>>
>>8872877
>without any personality
>JAVs
Nigga what? Watch a JAV with subs, skip the porn, and tell me it's at least not comparable to some daytime housewife soap opera. It adds to the appeal of why some cute chink became the ojisans cocksleeve. Western porn doesn't come close so it's always reliant on 'wink wink nudge nudge we're in a porn' acting.
>>
im just asking for one kirazuri-anima gen
>>
>>8872895
Sorry, anon. Not into gacha so not sure what slut you've been going on about.
>>
>>8872877
Bro what? JAVs are amazing even despite censorship. Especially oral, it's so much better than western shit.
>>
Just generated an amazing image using Kirazuri (Anima)
>>
based and same
>>
may i see them
>>
Hmm.. nyo.
>>
It's FINE. Nobody generated ANYTHING. Calm DOWN.
>>
my name is generator rex. i am generating wonderful images using Kirazuri (Anima)
>>
File: ComfyUI_38800_.png (1.23 MB, 832x1216)
1.23 MB PNG
>>8872896
probably >>8872861
From a few quick XYs I'm pleasantly surprised it doesn't fuck up existing styles much at all, guess it really is a finetune. Also Endfield girls work. I am getting consistently worse hands though, in compositions that even Pony would have handled fine.
>>
>>8872907
>it doesn't fuck up existing styles much at all
What? @freng doesn'r work for me at all
>>
>>8872907
>>8872908
Animayumebros.. has our position at the top of anime finetunes been compromised?
>>
>>8872908
yeah it fucks with style accuracy a lot
its kiraover
>>
>>8872908
looks like generic anime slopper to me
>>
>>8872908
Never heard of him. But you're right, I seem to have picked artists that didn't change much.
>>
>>8872907
Perlica works on base anima though
>>
>>8872917
Yes, but a lot worse in clothing. Even worse for Chen, and Fluorite doesn't work at all.
>>
HOW DO I EXTRACT A LORA FROM AN ANIMA CHECKPOINT
>>
schizo, schizo on the wall, who's the sloppiest of them all?
>>
>>8872814
some people just don't have that much, or have a lot, it's not that weird. the weirdest thing i hear is foreigners thinking everyone in the us is circumcised when it is actually very rare
>>
>>8872882
true
>>
>>8872922
you can do that in swarmui. there is also a comfy node but i haven't tried it yet
kirazuri v2 rank 64 lora: https://mega.nz/file/WuoXyKiZ#5wgYkxGv4vHLN9cQYid9wDIQWnaZvxF7BY4fNlTl-UI
kirazuri v2 rank 128 lora: https://mega.nz/file/3nRG0B6S#eaG0qRBKqI-nhTjdUZr737JMkxsP1lUnwtnAUu2Svhs
>>
THIS
WILL
HURT.

BE GONE.
>>
>>8872922
it's part of kohya_ss
>>
>>8872930
Because of porn, where foreskin is extremely rare for some reason.
>>
>>8872936
from what i've seen, i think it's also rare for it to be covering the head of the penis, when you're erect it's not terribly visible
>>
File: ComfyUI_27933_.png (1.53 MB, 1120x1520)
1.53 MB PNG
>>
>>8872930
>when it is actually very rare
it's not rare at all though, over 50% are still getting circumcised
>>
cut your foreskin if you don't want to have ball cancer
>>
good goyim
>>
>>8872941
lol
>>
>>8872942
Those seem completely unrelated
>>
>>8872946
quite.
>>
>>
during vpred days we called these fried
>>
>>8872958
my reaction to your post >>8872957
>>
better not post any gen in this thread
>>
>>8872922
C-B=A
>>
>>8872964
what tool
>>
File: ComfyUI_27945_.png (1.41 MB, 1120x1520)
1.41 MB PNG
>>
>>8872965
Vibecode something yourself. All it needs to do is spit out the difference between a bunch of tensors.
>>
>>8872607
If you want a good lora you have to do the needful and test settings with validation:
https://i.imgur.com/IrYNhX0.png
>>
File: 1755545025249660.png (1.25 MB, 832x1216)
1.25 MB PNG
I love foreskin (I don't have one)
>>
>>8872687
box pls
>>
>>8872976
>I don't have one
i'm so sorry
>>
>>8872976
imagine if she pulls too hard and it tears
>>
>>8872976
breastcurtainbox
>>
>>8872981
https://files.catbox.moe/3abkmr.png
>>
File: ComfyUI_27950_.png (1.71 MB, 1120x1520)
1.71 MB PNG
>>
>>8872958
oh shit a sunset image is highly saturated call the cops
>>
>>8872986
If this is not a shitpost then delete this and post catbox
>>
>>8872976
sorry for your loss
>>
>>8872988
that is a shitpost
>>
not a shitpost , I really need some help finding the style for this image in the box lol .

https://files.catbox.moe/93fk1d.jfif
>>
>>8872982
>foreskin_anima.safetensors
kek
>>
>>8872988
assume every post is a shitpost
>>
>>8872992
can't believe jfif is a real image format
>>
>>8872993
what no e6 data does to a mf
>>
>>8872976
I will hate my parents entire life, they even didn't ask me
>>
>parents gave me a more aesthetic penis and did it at a point in time that I wouldn't know or care about the pain while also being the fastest/easiest recovery time possible
based
>>
>>8872834
Okay, I won't be able to sleep tonight
>>
>>8873006
theres been a lot of botched circumcisions
shit is disgusting
>>
>>8873006
so they gave you SRS as a baby? crazy
>>
>>8873006
if at any point I wanted to mutilate my cock and lose a bunch on sensitivity because (they) wanted me to, I could
>>
>>8873006
>more aesthetic
They can just pull a foreskin down and it will look the same way
>>
I knew my first gf was a virgin because she was surprised about it not being circumcised because she only looked at porn (we live in a country with basically 0% rate of it being done)
>>
>>8872978
https://files.catbox.moe/j34eta.png
>>
>https://files.catbox.moe/dfihf7.png
>>
>https://files.catbox.moe/31igvp.png
>>
I think noodlekek may really be out of ideas at this point.
>>
being out of ideas never stopped me 1girl cowgirling
>>
I still have all the yuri/futa gens
>>
>>8873021
>yuri
Put cum in their cunt and post it.
>>
>>8873022
nah, that's stupid
>>
>>8873013
thanks
>>
File: .jpg (538 KB, 1328x1776)
538 KB JPG
>>
yuri gens are fine the male is just the one taking the photo
>>
>>8873023
yeah, stupid hot
>>
>https://files.catbox.moe/yknr1x.png
>>
File: lora extract.jpg (64 KB, 1216x413)
64 KB JPG
>>8872922
You can do this much >>8872964 with Comfy's built-in nodes. But unless you do a massive 256 rank "lora" it's very inaccurate for both styles and characters. So I wonder if there's more to it, most people do these extracts with kohya's tool.
>>
>>8873015
hmmmmm nyo i'm not feelin it
>>
>>8873033
massive loras is literally the best usecase for lora extraction tho
>>
>>8873038
But you might as well use the checkpoint directly at that point.
>>
>>8873039
it's easier to control a lora
>>
>>8873040
hmm
cool idea actually, don't use the shitmix at full power

of course it doesn't help if you wanted it for FOTM character knowledge, only style bias
>>
>>8873041
yeah doesn't really help with characters
>>
>>
>>8872682
idk why this is being spammed but i tried it a couple days ago
it sounds good in concept but it has lost knowledge of most artists
>>
>>8872682
never heard of it
>>
>>8873050
but bro the aesthetic™!
>>
>>8873044
why does he look disgusted
>>
>>8873053
he's in the cuck chair
>>
>>8873054
wait so is this actually male pov??
>>
>>8873055
no
>>
>>8873055
it is if you are the camera man
>>
We desperately need more POV netorase gens
>>
No, it's yuri/futa time
>https://files.catbox.moe/ygn9b8.png
>>
>https://files.catbox.moe/pn37y5.png
>>
>https://files.catbox.moe/wi3a2t.png
>>
anima sucks im going back to 102d
>>
https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/diffusion_models
>>
>>8873063
bloody bait bastard benchod
>>
israel'nt
>>
>>8873061
based
>>
>>8872916
>>8872940
>box please
Some of the artists at the end may be different, I've seen this mix in many variations, but the gist of it should be correct.
>>
>nai
yiiiiikes
>>
ouch, looks like local is still so far behind...
>>
File: .jpg (504 KB, 1880x1256)
504 KB JPG
>>8873058
> We desperately need more POV netorase gens
>>
File: .jpg (506 KB, 1536x1536)
506 KB JPG
>>8873073
>>
File: G9imtP2XsAATe9q.jpg (515 KB, 1795x2583)
515 KB JPG
>>8873073
kek

>>8873074
I wanted to do some firefly gens the other day based on this
>>
File: 1777999204508431edit.png (2.05 MB, 1536x1536)
2.05 MB PNG
>>8873074
BRO
Ever heard of a concept known as faceless male?
Disgusting as fuck.
>>
>>8873076
it's still pretty gross
>>
>>
>>8873074
can you gen a close up on the male's face
some anons cant accept the look of her master
>>
>>8873068
For some I didn't see the (You) in /hdg/.
First three are correct, others are pija, shexyo, ratatatat74 (wanted to gen him BLACK!) and opossumachine. I deleted the original gen so I don't have the weights (used different prompt for the upscale).
>>
>>8873076
She deserves handsome man
>>
>>8873074
reminds me a lot of gopro anons gorillas
>>
>>8873077
True, but I love pretty girls pinned down and all that. If anons here weren't into ugly creatures and cuckoldism many gens could be kino, but it is what it is.
>>
>>8873082
>gopro anon
he never ended up sharing that magical lora...
>>
>>8873083
>can do and post X type of gens
>no I won't because of the people here
what a cuck
>>
>>8873088
You missed bro, I post gens all the time.
>>
>>8873089
sure bro
>>
>>8873093
You've seen the horse on horse action, right?
>>
>>8873088
what awful people
>>
File: 1758067207318157.png (2.76 MB, 1344x1792)
2.76 MB PNG
>>
I've psyopped myself into thinking there's an Anime release today, but we all know nothing ever happens.
>>
some anime did release today are you retarded
>>
>>8873100
yes
>>
I am a sociopath, I have hard-on for people's funny reactions to some wrong/immoral thing, I don't totally self-insert, I'm just a little getting inside a cuck's head to get what it feels like to have them girlfriend fucked by some absolutely disgusting prick. At the same time, I don’t feel like a cuck, because I will never have relationships with women, I can only experience lust, I don’t know how to feel love, shame, or affection, I can fuck women, but I don’t feel them as "mine"
>>
yea yea yea move along now shoo
>>
>>8873097
box?
>>
>>8873102
Let me guess, you are also a proud member of the lgtbogpfpxbox360 community?
>>
File: 1764143178720372.png (2.62 MB, 1344x1792)
2.62 MB PNG
>>8873104
https://files.catbox.moe/gcsn9v.png
>>
>>8873107
legit the hottest thing i have seen posted here in months, great work anon
>>
>>8873104
Wow nice it worked, can I try it?
>>8873074
box?
>>
>>8873107
>white background
do it again
>>
>>8873097
thanks
>>8873109
>Wow nice it worked, can I try it?
what? do you come from some shithole that doesn't give catboxes out regularly
>>8873110
white background is based
>>
>>8873111
yeah i come from /hgg/
>>
>>8873110
slopper
>>
i come from a thread down under
>>
>>8873111
>what? do you come from some shithole that doesn't give catboxes out regularly
until recently hgg didnt give catboxes though. not sure what changed
>>
>>8873115
shitpost or retard? probably both
>>
there's many boxes in THIS VERY THREAD that were never delivered btw
>>
>>8873104
Doesn't seem to work=\
Can you try asking him please? I have bad aura I think.
>>
ok enough shitposting. unironically which sampler/scheduler will give me the sharpest edges on anima?
>>
>>8873121
euler a
>>
File: 1772603002265884.png (3.96 MB, 1344x1792)
3.96 MB PNG
>>8873110
sorry saar have a butiful background for your cuck fantasies to feel more grounded!
>>
>>8873123
you cheated
>>
>>8873126
>netorase gen has cheating
wow
>>
Any tags for saliva flying out of the mouth? Need to make the moment of first penetration more expressive.
>>
>>8873128
hehe
>>
>>8873129
projectile ___. also this is a perfect use case for natural language
>>
>>8873129
spitting reaction
>>
>>
>>8873133
connect their urethras with a large tube and make them pee into each other to create a perpetual peeing machine
>>
>any tags
bro... just... write
>>
>>8873135
that's not how that works
>>
kikimetal lora doko
>>
>>8873135
pee is gross
>>
>>8873138
upvote
>>
>>8873131
>>8873132
Thanks, sorta works but needs fiddling.
>>8873137
Currently genning with NAI so NL sucks.
>>
>>8873139
yellow pee is gross. clear pee is fine
>>
File: .jpg (550 KB, 1536x1536)
550 KB JPG
>>
>>8873142
uh. no
>>
>>8873137
what do you mean?
>>
Do diffusion threads make people so mentally ill or they attract mentally unstable people in the first place? It's impressive how fucked up people here are.
>>
>>8873146
yes
>>
>>8873146
novel ai
>>
a real friend of noodles would have baked /h2gx2/ by now
>>
>>8873138
>found their fanbox on kemono but kemono is still dead
fucking pain
>>8873146
it's a combination of both and irony poisoned zoomers. it would be hilarious if the next /hgg/ thread was retitled /hdg/
>>
>>8873143
make her pick the smegma with her finger and have her rub it around in her teeth. all natural breath freshner, comes with free floss too
>>
>>8873146
it's less about it being an SD thread and more about it being a "general" thread.
the only general threads on the entire website that are free of mentally ill schizoposters are extremely small, niche threads tucked away on slower boards.
>>
>>8873146
>Do diffusion threads make people so mentally ill or they attract mentally unstable people in the first place?
Yes and yes sadly, you'll get used to it
>>
>>8873150
Most of their posts are also archived on sadpanda.
>>
noodlefriend save us
>>
File: 1778004017169340edit.png (2.62 MB, 1536x1536)
2.62 MB PNG
>>8873143
She cleaned it.
>>
any artists besides kikimetal a lora would be worth training on?
>>
>>8873157
kyokeer or fujii shingo
>>
>>8873153
i will take the schizos here over the shitfest that is /b/ and the retarded avatartroons on /g/
>>
>>8873156
in a way, you cleaned it too
>>
>>8873156
her pee can ferment into alcohol now
>>
>>8873157
ayakon
>>
>>8873163
we already got that lora tho
>>
has anyone tested fucking with shift when training styles
>>
>>8873159
>i will take the schizos here over the shitfest that is /b/
pedos refusing to go to /b/ and shitposting in revenge whenever board rules are enforced is a big reason for the decline
>>
>>8873158
that first one is absolutely abhorrent, second one is cute

what causes this
>>
File: test.jpg (1.02 MB, 1664x3648)
1.02 MB JPG
>>8872858
It actually worked. Though better captions could probably reduce the style bias. It's kinda pointless too, turns out without facial details every girl is just hair and an eye color.

Also works for actual realism >>>/aco/9191817
>>
File: .jpg (390 KB, 1776x1328)
390 KB JPG
>>8873156
>>
>>8873167
please.
>>
File: 00052-583336035.png (3.86 MB, 1792x2304)
3.86 MB PNG
>>8873155
you didn't enjoy the yuri/futa gens? what else do you want to see?

>>8873159
How different is the /b/ thread from the one on /trash/? /slop/ is filled with drama more than often for my taste
>>
>>8873169
It still won't go into my horny folder until I make the guy white but I can't be bothered right now.
>>
>>8873168
block weights of 0 in the appropriate in/down blocks destroy most style influence

down_lr_weight = [ 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0,]
>>
>>8873171
>what else do you want to see?
/2h2g/
>>
>>8873170
put together the dataset and i'll bake it for the second one, i'm vetoing the first one because i don't bake cringe
>>
>>8873158
cool thanks for the ideas
>>8873166
that's a neat headcanon
>>8873171
the /b/ threads are unusable with nothing but namefags splitting threads and spamming each other constantly. i haven't been on /trash/ much but they seem to just do their thing with the occasional melty at the bbc generals, but that's the case for any board here
>>
>>8873173
Thanks. I had the old ones at some point but figured the layout would be different for Anima.
>>
>>8873173
how do people figure out what blocks to disable for things?
>>
>>8873171
/b/ is filled with borderline cp shit, not cute loli stuff but almost 3dpd and just awful gens. also constant popdick schizo and namefag drama that they revel in.
slop gets invaded by schizos but it doesnt feel like regulars enjoy drama too much. most of the stuff is kinda regular sloppa but sometimes good gens here and there. it's way too fast for me though
>>
>>8873178
by experimenting
>>
>>8873178
there were some articles about it like a year ago, then there was a mix of schizo imaginings and empirical testing, i didn't believe it myself until actual testing, either

>>8873177
the math behind lora are the same
>>
File: 00073-3328221.png (3.65 MB, 2688x1536)
3.65 MB PNG
>>8873176
>occasional melty at the bbc generals, but that's the case for any board here
even the blacked threads hate each other, is kind of funny in some way
as a contrast enjoyer I am sad because no one seems to do it without any of the /pol/ bullshit implication

>>8873179
>/b/ is filled with borderline cp shit, not cute loli stuff but almost 3dpd and just awful gen
as expected from /b/ I guess geg
>>
>>8873179
meanwhile we just get shitpost gens with some noodle masterpieces in between. regular sloppa sounds pretty great to me...
>>
>>8873186
cool concept, shit gen
>>
>>8873188
Noodlekek works hard mixing with utmost precision scientifically the most bland styles.
>>
>>8873168
>It's kinda pointless too, turns out without facial details every girl is just hair and an eye color.
desu anybody could've told you that before doing it. i thought your goal was some 2.5d shit
>>8873171
box?
>>
>>8873187
It's a combination of people being tired of genning and the resident bot commenting negatively on almost every gen.
>>
File: 1762921050449486.png (1.47 MB, 896x1152)
1.47 MB PNG
>>
>>8873176
>neat headcanon
newfriend...
>>
>>8873194
>newfriend...
uh no. i actually very explicit remember what happened that caused people to split to moe and your revisionist bullshit doesn't change that
>>
I miss treech
>>
>redditor vs redditor
woah..
>>
>>8873197
you're a redditor??
>>
>>8873197
sorry guys, i posted in the wrong thread, ignore this.
>>
File: 00226-1538077896.png (3.67 MB, 1536x2688)
3.67 MB PNG
>>8873190
what does this even mean, by this this point you should know I have a mix for every kind of feel/concept and setting that I want to do
from digital media kino to p*interly slop

>>8873191
>box?
>https://files.catbox.moe/rak89o.png
>>
>>8873200
>kana
BASED.
>>
>>8873200
show me her uvula
>>
>>8873202
the room is dark and anything is barely visible, are you stupid or some kind of animal with perfect night vision?!
>>
>>8873203
no i have nogs on
>>
>>8873203
are you senile?
>>
File: 20260505143434-3400898049.png (1.86 MB, 1280x1280)
1.86 MB PNG
>>
>>8873203
My vision is augmented.
>>
>>8873206
it's not senile, it's schizo.
>>
>>8873200
You don't have horny styles. Which is why you have a need to gen blacked to at least feel something from your gens anymore!
>>
>>8873168
Wait, people stopped doing this? I used to do this all the time. Skin Diamond Midna was hot.

>>8871771
>Hasumi Kurea
Her, Shinoda Yuu, and Tsubomi were the shit
>>
File: 1762752020683781.png (1.59 MB, 896x1152)
1.59 MB PNG
>>
>>8873212
uhh, hentai?
>>
>>8873213
xhes drawing it
>>
>>8873211
Anime girls in realistic styles, sure. Not the other way around.
>>
>>8873213
(you) are the hentai
>>
I CAN'T PROMPT
>>
WHERE IS PREVIEW4
>>
what are some good jav with stories of women showing up to interviews to become idols and coerced into av instead
>>
3dpd is garbage, guys
>>
>>8873221
just like your mom
>>
>>8873222
exactly
>>
>>8873216
Results seemed good at the time. To be fair I was doing it when high-contrast 1.5 shading was 'good' and celeb loras and TIs were everywhere. Slightly harder to fuck around with that now.

>>8873221
THAT'S WHY WE'RE TRYING TO CONVERT THEM TO 2-D YOU UNCULTURED SWINE!
>>
File: 00009-3484597226.png (1.12 MB, 1088x1536)
1.12 MB PNG
>>
>>8873221
jav is honorary 2.5d
>>
>>8873226
ok
>>
>>8873226
Finally, fire.inc banger, day saved.
>>
>>8873225
2d is not only appearance, they have transcendental personalities 3dpd will never reach
>>
>>8873226
>non-/h/ slop
>>
bobbleheads are /hgg/ culture
>>
>>8873230
I-I can fix them...
>>
>>8873226
While (You) are getting scolded, I'm behind checking out her panties btw. Have to really crouch, who wears skirts this long?
>>
>>8873233
you won't be able to fix yourself
>>
File: ComfyUI_27991_.png (1.4 MB, 1120x1520)
1.4 MB PNG
>>
>complains about 3d when he is 3d himself
>>
File: QYUWQBUwPPOUtEEjE7QeeZ.jpg (118 KB, 1344x1728)
118 KB JPG
>>
ok nigga and?
>>
is that jav?
>>
>>8873236
No shit dude.
>CANT FIX PERFECTION HURRR HURR
>>
>>8873237
artist mix? she has a cute face
>>
I still can't forget 1106 post, help me guys, it's gonna follows me
>>
>>8873226
into the reeducation chamber you go missy
>>
>>8873227
more like 3.5d because how repulsive are these pug faced whores
>>
>>8873246
KEK
>>
>>8873246
who hurt you...
>>
What the fuck is shift
Can one of you fags be useful for a min and explain it to me
>>
>>8873248
God by giving me eyes
>>
>>8873249
have you considered using ctrl+f for once in your life
>>
>>8873249
scroll up
>>
File: 1767712634964809.png (2.28 MB, 1152x1408)
2.28 MB PNG
>>
>>8873255
what's up there in 2023?
>>
yay slop spambot
>>
just genned an image
>>
>>8873258
same
>>
>>8873258
real shit?
>>
why does the bot spam
>>
I CAN'T GEN
>>
Guys i just put "a picture of my MASSIVE COCK" in as a prompt and nothing came out i think its broken.
>>
how do i get better at prompting NTR?? i use the netorare tag but it feels like it doesnt actually do anything
>>
>>8873265
step one: be brown
>>
>>8873265
NL
>>
>>8873265
because NTR is a concept, it's just regular sex with an added context
>>
>>8873264
Great start – the important thing is you are trying new creative things
Do you want to share your settings – then we can work out a solution together.
>>
>>8873268
>just write! just use tags!!
>but not that one, or that one, or that one...they dont do anything
what the fuck?
>>
>>8873270
to be fair prompt engineering is a subtle and often overlooked art. it needs immense skill and dedication to effectively communicate with ai.
>>
Anyone else ever just sit and read through the thread and wonder what the fuck they're even doing bothering to come here anymore?
>>
>>8873270
listen
just create a mental model of the ai model in your head
it's not that hard
>>
>>
>>
File: .jpg (400 KB, 2048x1152)
400 KB JPG
>>8873270
>>
I stay here and shitpost because you were mean about my gens before
>>
>>8873274
at this point i kinda even enjoy this place's inansity and appreciate the humor
almost worthless for genning but I know everything I need to get the images I want
>>
>>8873281
Same, really
>>
>>8873279
Is that ursa minor...?
>>
>>8873277
>>8873278
her age
>>
How come there are no good lora bakers left here? during the pony era we had an entire rentry full of links and bakers were always on the hunt for novel styles from game cgs and other unconventional sources. bakers also displayed a much higher level of technical knowledge back then. Now we finally have good models and only jeets are left to bake.
>>
what caused the scat spam?
>>
>>8873277
>>8873278
et tu?
>>
>>8873286
you were mean about my loras so I don't post them
>>
>>8873285
18
>>
>>8873286
>bakers also displayed a much higher level of technical knowledge back then.
Not just bakers
You ask a question in these threads and you're always met with either irony-poisoned shitposting or completely retarded answers that amount to telling you to do what you already did
>>
>>8873286
Do you think /ldg/ being the new lora place was a shitpost? Don't know why you choose to rot here if everyone already moved on. /hgg/ is just for penis close-ups and schizos, everyone who's actually serious about the hobby is posting on /ldg/.
>>
>>8873291
when somebody gives a good question i give them a good answer
when somebody asks some retarded shit they would know if they practiced more they get nothing
>>
>>8873286
Because back when Pony was relevant that was the only thing everyone was using. Pretty much the entire thread was using autismmix so it was easier to share things.
Now every single person is either on some random noob ai model or fucking around with anima it makes it hard
>>
>>8873292
>>8873295
it was decent bait until you started including the pony shit then it kinda lost its steam
>>
File: 1764206011929028.png (2.39 MB, 1408x1152)
2.39 MB PNG
>>8873256
Pretty good
>>
bitter schizos chased off every name or recognizable poster. Purists on anon culture doesn't transfer to consistent content posters
>>
https://litter.catbox.moe/gd35bcazhzppe8uh.png
>>
>>8873293
>when somebody asks some retarded shit they would know if they practiced more they get nothing
Then shut the fuck up you retarded mongrel
Leave people who actually want to help answer instead of shitposting
>>
>scat spam
>ponyv6 fried slop
>f*ta
:skull:
>>
>>8873298
>bitter schizos chased off every name or recognizable poster
we never succeeded in getting rid of any namefags tho. I'm pretty sure the few that did leave after a while like miqoanon or ayakon just got bored
>>
>>8873302
miqofag is posting on /slop/ tho
>>
>>8873302
i just dont post images here anymore but ill still shitpost all day because its funny as fuck lol
>>
>>8873305
good head size
>>
>>8873309
too small, it only takes up half the gen
>>
im training my loras on ponyv7 so I am ready for pony v8
>>
>We're still mostly using SDXL/Illustrious model variants
>Most of the old loras still work unless your shitmix is too 'unique'
>Anima isn't 'done' yet so no one wants to commit to baking for it yet
>>
>>8873305
if she were to tilt her head just a bit more to the side she'd end up committing suicide
>>
>>8873314
XD
>>
>>8873314
you should try that
>>
>>8873318
why would you say that?
>>
the biggest shitpost of all was splitting the thread for no reason, still don't even understand what the goal was
>>
>>8873320
I can't wait for highlightschizo to hijack this thread, too.
>>
>the worst avatarfaggot of the thread is the only one who still posts here
monkey's paw bullshit
>>
I never minded highlights, why did you get so schizo over them, did you just never get picked because your gens sucked
>>
>>8873323
let it all out
tell us how you really feel
>>
>never minded highlights
>still posts in the thread specially made to escape highlights
>leaves iafag to rot in /hdg/
curious
>>
>>8873305
>posting the same off-topic shit since 1.5, with the same style since pony
what mental illness this is
>>
>made to escape highlights
do you think people can't see this thread or something or you hid well
>>
>>8873327
please understand if he changed then he would lose his twitter following
>>
>>8873320
about three posters moved and made another thread and you schizos whined about it for a year and ultimately ended up migrating over
>>
>>8873328
retarded argument since it has worked for 44 threads
>>
>>8873321
he would never do that, because that would be accepting defeat in his mentally ill mind. i wouldn't be surprised if he lurks in this thread just searching his name from time to time
>>8873328
is this an /edg/ tourist malding?
>>
can we fucking split please to get rid of these shitposters? this thread has been especially bad its time for a change
>>
>>8873333
Yeah you can go to /hdg/ right now to hang out with IA schizo and the two sycophants who suck his cock
>>
ill follow you to make bad quality posts wherever you go
>>
>>8873333
I'm still voting for a move to /e/. Maybe that will somewhat limit the first few new threads getting relentlessly spammed by the anti-split schizo
>>
a competent jannie would just consistently delete one of hgg or hdg I really don't care which dies
>>
>thread gets slightly faster than completely dead
>HELP SHITPOSTERS EVERYWHERE
>>
>imagine being so hated not even the shitposters want to post in your thread
>>
>>8873334
>>8873332
>>8873326
Why do you keep samefagging? Everybody knows only you are this obsessed with baker anon.
>>
would have been perfect, if not for the stereotypical chinaman tooth the ai gave her
>>
File: k8736p.png (7 KB, 660x119)
7 KB PNG
>>8873340
why are you here instead of /edg/ or /hdg/? not enough attention?
>>
>>8873286
For me it's because loras are only necessary for weird niche shit now. Which I still bake, but can't be bothered to reach the quality level that people itt have come to expect. I can share some trashy ones if that helps.
>>
>>8873344
>I can share some trashy ones if that helps.
Nice of you, but this thread is a lost cause. Basically a waiting room for the next split
>>
>>8873340
>>8873344
Is this the /edg/ dunce who loses his mind whenever someone points out his loras being fried?
>>
>>8873330
this is the real lore btw
>>
>>8873347
in your headcanon, that is
>>
>>8873342
ngl from the thumbnail i thought it was a petrification gen
>>
>>8873305
never gen this ugly specimen again
>>
>>8873350
It's firedotinc self-insert.
>>
if you were willing to be a namefag not a coward you could be like comfy with 10s of millions
>>
>>8873352
is this a discrete shill for X dot com
>>
If X is so popular in Japan, why don't they build their own X that we can then all use happily without fear of getting banned for being too degenerate?
>>
File: 00539.png (3.78 MB, 1664x2432)
3.78 MB PNG
>>8873348
But it is true tho, it was only me and chat
Already posted how it was the first thread the other day
>>
>>8873354
that's misskey. foreigners aren't allowed on the main server though :)
>>
>>8873354
misskey exists. and it's hyper locked down and pretty impossible to make an account on unless you're actually in japan.
>>
>>8873357
what stops someone from just using a vpn? is it a korea situation where you need actual id
>>
>>8873358
you need a japanese phone number or address or some shit like that to sign up
>>
bluesky is open to new users!
>>
>>8873358
there isn't a single viable VPN that isn't blocked on their end. I think instead of blacklisting they're actually whitelisting ISP's.
You should maybe be able to register if you're using tunneling but then you have to find someone in japan willing to let you molest their internet connection(or register for you).
>>
>>8873361
>post something with cleavage
>bsky moderation blocks the image from being viewed by anyone who hasnt altered the deeply buried censor settings
>>
>>8873355
Inauthentic, factory-made pussy, all mathematically perfect and not messy enough.
>>
what if they nuke japan? what would you do?
>>
>>8873355
hi, I'm chat
>>
>>8873366
show me a pov gen of you tuggin'
>>
>>8873365
>what if they nuke japan?
It's called mass migration.
>what would you do?
Only thing I can do as a powerless goy: watch the explosion in slow motion.
>>
>>8873367
nyo
>>
>>8873356
isn't it dead tho, because of no gaijin updoots?
>>
>>8873362
found a really easy way to access the content on the site. just do basic research and you'll learn
>>
>>8873374
accessing content isn't the hard part. getting an account so you can actually post to it is.
>>
>>8873374
based vague poster
>>
>>8873364
it's SDXL

>>8873366
hi chat, don't forget to buy the new merch on store
>>
>>8873375
oh. well i'm not interested in posting so that's fine with me
>>
>>8873374
vagueKING
>>
>>8873354
>If X is so popular in Japan
they're actively working on destroying that popularity btw
>>
>>8873380
The japs or Musk?
>>
>>8873381
Nikita Bier specifically, guy who controls the algo on there. they combined the EN & JP feeds so that both sides saw the others posts wayyyyy more frequently
>>
>>8873382
the nips don't seem to care/mind though?
also the funny thing is I didn't actually notice any real difference. my feed was 90% nip shit to begin with.
>>
>>8873383
they were hilariously seething about anime piracy lately, so kinda outed themselves as bugmen retards
>>
>>8873384
that was largely ragebaiting hyper conservative nip nationalist accounts though. they were doing retarded shit regardless. And there were just as many nips calling them retarded.
also before that there was the dumb peanuts in coke meme that nips were trying and claiming to actually enjoy.
>>
>>8873386
>there was the dumb peanuts in coke meme that nips were trying and claiming to actually enjoy.
because it is good hijo de puta madre
>>
>>
File: 1766858952573965.png (2.1 MB, 1152x1408)
2.1 MB PNG
>>
Sirs, is bibici 3d or 2d? ʕ•ᴥ•ʔ
>>
looks like poojeets are waking up
>>
>>8873396
looks more like the /e/ spammer switching targets
>>
looks like my favourite posters are waking up
>>
meanwhile local is still asleep....
>>
>>8873398
you should do that on /hdg/ instead. they're desperately in need of some images
>>
>>8873402
Nobody wants to post in IA thread.
>>
https://youtu.be/4VTBMznLrWs?si=kJ7VtBWCf9db1biW
>>
Reminder that literally every single post that mentions highlights bro in negative light is made by one person - the spammer schizo. He's obsessed.
>>
>>
which one
>>
>>
>>
>>
File: highlights.jpg (158 KB, 1664x2432)
158 KB JPG
PREVIOUS THREAD HIGHLIGHTS
>>8870706
>>8870707
>>8870709
>>
>>
>>
>/edg/ spammer has lived rent free so badly now one of ia schizo's apostles is gonna spam /hgg/
kek sorry bro this is just the norm here. you can't match his energy
>>
>>8873416
honestly never thought I'd root for a spammer but edg is so incredibly cancerous now that it made me enjoy the show
>>
feels good being a novel ai chad
i come to this thread every couple of months and it helps to validate that i made the right choice
>>
many say this
>>
>>
>>8873427
me on the left
>>
>>8873428
in the phone?
>>
>>8873429
It's called facetime, chuddie
>>
>>8873430
is that why she's bored?
>>
>>8873430
she has an android...
>>
>>8873427
me as the prostitute girl
>>
>>8873429
shes sexting me on discord
>>
>>8873427
@grok what is she watching and provide a link to the video
>>
>>8873435
https://www.youtube.com/watch?v=dQw4w9WgXcQ
>>
>>8873432
let's be real, sex androids for women are gonna be like 95% roleplaying and 5% sex
>>
>>8873427
She's sending me a $400 amazon gift card in this image btw
>>
>>8873438
with death of the author you can imagine anything you want
>>
>>8873439
the man who genned that image died?
>>
always feels weird seeing people reuse/repurpose my prompts
>>
>>8873440
baka my head uncultured anonies https://dn721906.ca.archive.org/0/items/TheDeathOfTheAuthor/the%20death%20of%20the%20author_text.pdf
>>
>>8873443
not reading allat
>>
>>8873444
it's only 7 pages
>>
>>8873445
tldr?
>>
>>8873446
ask grok
>>
>>8873446
just give the link to an LLM
>>
A lot of people seem to dislike highlights, likely because it takes up an image slot to showcase images already posted in a previous thread. It's unfortunate, because highlights help recognize dedicated high-quality posters and show appreciation for good gens. Instead of highlights, I decided to introduce the recognitionlights, in which posters will be recognized for their contribution to the thread. This is still a beta format, so let me know what you think. I will recognize some of the earlier posts and maybe do more later if this format has positive reception
>>8871139
>>8871256
Crazy good stuff, the POV is really nice and the bikini style and lewd body really contrast with the viewer's expectations.
>>8871159
The skin looks wrinklier than an old leather sofa, but I'm not complaining! A very unique and captivating style.
>>8871182
This one has a classic anime feel to it which really adds to the mood!
>>8871259
The colors are hypnotising and psychedelic, a really nice gen!
>>8871292
A classic pose, solid gen overall!
>>8871341
Woah, these are some crazy proportions! But all the more to enjoy!
>>8871342
Very artistic, and a nice use of motion lines
>>
oh this particular schizo is back
>>
>>8873449
kek. i respect the effort. the sad part is there's generals where shitposts like this are real
>>
>>8873449
>intentionally only tag shit gens
kek
>>
>>8873452
he missed plenty
>>
File: 1777545617333293.png (2.5 MB, 1152x1408)
2.5 MB PNG
>>
>>8873448
clearly you have an intuitive grasp on the material you have linked. i want YOUR interpretation.
>>
>>8873452
Seethe.
>>
>>8873456
i'm applauding you, however. keep going.
>>
>>8873455
you're not gonna flatter me into doing what you want
>>
>>8873454
Nice, very painterly.
>>
just ran out of (you)bux. cant give out any more (you)s. sorry
>>
>>8873454
blud has never once ever skipped leg day
>>
excessive body hair is back on the menu boys
>>
>>8873452
I think >>8873449 is a bad actor as he tagged Chadmix's gen and Chadmix is a friend of hdg/hgg culture.
>>
>>8873286
xenobladeschizo finally gave up on this place despite getting the 16ch vae model he was praying for everyday
good for him honestly
>>
>>8873464
i've been here on and off since we were making embeddings. who the fuck are these namefags and why do you keep track of them?
>>
>>8873465
female-brained
>>
>>8873465
there's simply no way you missed the guy who trained literally hundreds of loras for xenoblade nia if you were posting here at that time
>>
>>8873468
maybe if i went into the archives i'd remember but this place occupies such little amount of mental bandwidth there's only a few posters i remember at this point
>>
>>8873468
to be fair that dude was like a ghost, barely noticeable
>>
>>8873469
whatever you say newsir
>>
>>8873463
He's clearly tagging absolute garbage with some good gens sprinkled in to lower morale
>>
>>8873465
>i've been here on and off since we were making embeddings
you mean you've been in /ldg/, not here. no larping buddy
>>
>>8873465
are people pretending to not remember dumb shit or this is some real intellectual deficiency
>>
>>8873475
>calling me newsir
>doesn't remember before /ldg/ existed and everybody was on /sdg/
lol
>>
my gens have always had a clear theme yet ive avoided recognition. i've won.
>>
>>8873470
he was one of the few knowledgeable people who posted on this dogshit general. i benefited a decent amount from his scripts and lora experiments
>>
>>8873474
anybody actually posting images here is already used to all this shitposting to begin with. i doubt anybody's morale is "getting hurt"
>>
>>8873478
nobody will care to recognize you is you are not a lolcow
>>
>>8873482
chadmix anon is highly regarded doe
>>
i'm ayakon
>>
>>8873483
yes, and he is fucking retarded
>>
got a little over 100 bucks on tensorhub from random shit I imported from civit once lol, almost incentivizes me to post more there
>>
>>8873486
make your claim without taking the low-hanging rhyme fruit
>>
>>8873449
DISECTING A TROLL POST: HOW TO TELL WHEN A POST IS DISINGENUOUS
>A lot of people seem to dislike highlights, likely because it takes up an image slot to showcase images already posted in a previous thread.
We start off with a truth (a lot of people dislike highlights), but follow up with false-reasoning (because it takes up an image slot). This is known as "intentionally missing the mark", a classic but trollish use of Cunningham's Law
>It's unfortunate, because highlights help recognize dedicated high-quality posters and show appreciation for good gens.
Lamenting about the loss of highlights (even though nobody misses them) and mischaracterizing their purpose as "recognizing high quality posts" when in actuality they're just spam. This is a technique designed to get users to reply in anger.
>I decided to introduce the recognitionlights, in which posters will be recognized for their contribution to the thread.
The troll post goes on to rebrand the highlights as a 'recognitionlights' while acting like it's an entirely new idea.
>This is still a beta format, so let me know what you think.
An attempt to bait angry replies by openly asking for them
>I will recognize some of the earlier posts and maybe do more later if this format has positive reception
By stating he'll "maybe do more later", the poster hints that the antics are not over and there is still more to come. Those fed up with the highlights are now trembling with a sense of foreboding uncertainty, knowing they could return at any minute.
The poster then goes on the 'recognize' various AI artists with unsolicited praise, which could lead to an uncomfortable and awkward situation due to the nature of a public forum.
>>
>>8873489
do you have aspergers?
>>
>>8873489
grok nobody tagged you fuck off
>>
File: ComfyUI_03552_.jpg (2.18 MB, 1664x2432)
2.18 MB JPG
Can I get some constructive criticism on this gen?
>>
>>8873493
yeah why is she tucking her right arm like that?
>>
>>8873493
the man has lipstick
artstyle sucks
>>
File: ComfyUI_16629_.jpg (1.67 MB, 1664x2432)
1.67 MB JPG
>>
>>8873493
it's fried
>>
>>8873493
save some cfg for the rest of us
>>
File: ComfyUI_38970_.png (1015 KB, 960x1088)
1015 KB PNG
>>8873346
Probably not? imo losing your mind over any online content is questionable at best. They are overtrained tho, still haven't gotten used to how quickly Anima loras bake.

Here's "@kikka-sora" (trigger tag), also known as 橘花 空 on Pixiv.
https://files.catbox.moe/5aov4y.safetensors
>>
>>8873499
ion want yo fried sloppa ahhhh lora
>>
>>8873493
you might as well have just said
>/hgg/ roast my gen
>>
>>8873502
tbqh he proved why nobody posts images here unless they're calloused to the abuse
>>
>>8873503
pretty much every non-noodle image is a shitpost
>>
>>8873503
Can't really complain when he asks for feedback
>>
>ask people to criticize your gen
>people criticize your gen
>see bro that's why this place is so bad
someone used the term female-brained earlier
>>
>>8873496
wow this gen is good :) good work buddy
>>
>>8873493
female is too old
>>
ungenuine hours
>>
>>8873503
same tired bait i've heard since early 2023
>>
genuinely awful hours
>>
ungenuine thread, chud
>>
>>8873511
you're part of the problem
>>
hmmm nyo
>>
so
>>
i jacked off already so i'm just here for the vibes
>>
about that jav lora
>>
>>8873517
ask on /aco/
>>
>>8873518
do they even use anima? also ew
>>
>>8873517
make the dataset, chudnym
>>
>>8873493
the guy is crossdressing, isn't that a female dress? (plus the stockings)
>>
>>8873519
no, they use cartoona
>>
>>8873517
you know the rules
>>
>post old shitty gen made on pony
>farm (you)s
thanks everyone for playing along
>>
we get plenty of new shitty gens that look like pony anyway
>>
imo pony had better aesthetics than anima
>>
kemono being dead sucks so much for lora training bros
>>
>>8873522
I'm assuming he meant the photoreal general, not the cartoon one. They mostly use Chroma.
>>
>>8873529
sadpanda often has more complete archives doe
>>
>>8873532
sadpanda requires me to gamble on seeded torrents
>>
>>8873533
just DDL
>>
>>8873533
ACK
https://i.imgur.com/pxQPYan.png
>>
>>8873531
last time i looked at photoreal it was kinda dogshit, did it improve
>>
>>8873533
>he doesn't know how to just rip the gallery
ngmi
>>
>>8873537
hello sir how do i do that
>>
>>8873534
that requires their retarded currency
>>8873535
guess it's time to treat exhentai like a private tracker
>>8873537
explain a way to download large galleries without GP that is the original files and not the previews
>>
>>8873532
>200 images on kemono
>search exh
>11 images
yeah no, its not a replacement, not even close to it
>>8873540
>guess it's time to treat exhentai like a private tracker
Wasnt it always like so? I have all that and don't even use it because 99% of the galleries I just download with the downloader or a torrent
>>
>>8873540
do not download on weekdays between 14:00 and 20:00 UTC
>>
>>8873543
wait...what the fuck?
>>
>>8873544
https://ehwiki.org/wiki/Downloading
>>
>>8873536
Why don't you look again?

Anyway they split off into Mixed Model General for the SAAS stuff >>>/aco/9172516 which is basically their version of /e/ due to the censorship, but it gets the "style" right. And /asdg/ for the local stuff, which has always struggled with thread identity. Shared between people who use anime-based SDXL merges for that uncanny valley look, proper realism stuff trained on base SDXL/Chroma/Z-Image, and the occasional weirdo posting 3D game renders.
>>
>>8873542
>Wasnt it always like so? I have all that and don't even use it because 99% of the galleries I just download with the downloader or a torrent
i never interacted with it because i never needed to
>>
>>8873539
>>8873540
https://github.com/dazedcat19/FMD2
https://files.catbox.moe/8jtcjb.png
not going to walk you through setting up the account shit because you should be able to do that on your own by now.

tbdesu if you read korean coloring books you should probably want this set up with shit you follow anyway, provided you're not a webview only retard.
>>
>>8873535
How do you even accumulate this much. Just seeding?
>>
>>8873550
Creating & uploading torrents. Seeding is useless.
>>
>>8873550
I uploaded a bunch of galleries, its right there in the image
>>
>>8873554
so basically what i'm hearing is i have a new use for all these curated datsets i've built up and edited for training lol
>>
>>8873535
so THIS Is why ayakon uploaded 4 sets there
>>
>>8873540
I wrote a bot almost two decades ago to play their text RPG and farm currency. You could also just upload a gallery of something that's not on there yet, and it'll earn over time.

Not really useful advice right now, but maybe in ten years.
>>
>>8873555
They prefer complete galleries to curated ones.
>>
>>8873558
i can do that. is this why this site gets a ridiculous amount of ai generated galleries uploaded to it on the regular? i can't imagine that being good for their storage/bandwidth situation
>>
>>8873549
What's the difference between using this and just something like gallery-dl?
>>
>>8873556
god his pics suck so fucking much
it's painful how little self-awareness he has
>>
>>8873559
I don't know about AI. The idea for artists is that missing stuff triggers fomo in people who do not trust the uploader's judgement, your gallery can be replaced if someone uploads a more complete one.
>>
>>8873561
lets see yours, tough guy
>>
>>8873560
haven't used gallery-dl but fmd is just a scraper that's very specifically built to scrape manga format stuff. and in this case has really specific settings/toggles/etc. for websites of interest(in this case being panda). Once you set your login you'll never need to care or worry about that shit again. You can set it to ddl original size images from the gallery and it'll also autoconvert webp to whatever format you want, which is neat.
as far as korean coloring books go you can also just enter a url of whatever series you want to follow and it'll ping it every now and again to see if it's gotten an update, and if it does it'll auto-download whatever chapter got pushed. It's really super convenient.
>>
>>8873563
>>8873305
>>
>>8873567
prove that's you
>>
>>8873563
>>8873454
>>
I like big heads and I cannot lie
>>
sames
>>
File: mas.png (268 KB, 512x512)
268 KB PNG
>>
>>8873583
uwaa… subarashi…
>>
>>8873305
damn, this really upset someone.
>>
>>8873596
kek, wtf
>>
>>8873596
can't wait his autistic tantrums and spamming with blacked grannies
>>
>>8873608
acquire some meds
>>
>>8873611
newfag or retarded?
>>
>>8873612
>>8873611
>>
both
>>
Why are japs like this
>https://x.com/Mick19988/status/2051603186260349407?s=20
>>
>>8873624
it's a vchuba endorsement.
and the heart hand>thumbs up thing is a joke.
>>
>>8873624
why are you like this, nigga
>>
>>8873624
true okinawan nipponjin
>>
>>8873624
you mean based?
>>
if the style lora only looks accurate if i put it at weight 4.0, does that mean it's underbaked? should i up LR or do more epochs?
>>
>>8873631
post your config
>>
>>8873631
>if the style lora only looks accurate if i put it at weight 4.0, does that mean it's underbaked?
yes
>should i up LR or do more epochs?
depends
>>
>>8873633
https://rentry.co/srr87x7k
artist is https://gelbooru.com/index.php?page=post&s=list&tags=piennamekuzi
55 images
using @ activation
>>
>>8873639
Only 20 epochs for 55 images with adamw8bit is pretty low.
>>
>>8873639
55 images at 1 repeat and 20 epochs with a batch size of 4 is 275 iterations. way too low.
>>
>>8873639
you need to up the step/epoch count by like, +1.5x~ or something.
>>
>>8873641
>>8873642
>>8873643
damn how long does it take to train a style? this run took about an hour on a 4090, maybe I should queue them up overnight. i still don't really understand batch size/gradient accum and the impact it has on the output.
>>
why waste all this time on loras when wai base style is better anyway?
>>
>>8873644
idk for your settings but i train at a really high learning rate (probably wouldn't recommend) at 4e-4 and normally aim for 600-1000 steps at a batch size of 4. you also don't need gradient accumulation at that batch size for a 4090. that also takes me around 45 mins to an hour
>>
>>8873644
try this config that I copied from this guy https://civitai.red/user/toghashie441
https://files.catbox.moe/p5apcw.toml
it learns style pretty well. train for atleast 1100 steps at batch size 2. I personally think batch size 2 learn style better than batch size 4
>>
>>8873644
>>8873648
>>8873649
also batch size is just how many images the lora looks at before updating its weights once. the lower teh batch size, the more influence each image has, which helps preserve stronger character/style details instead of averaging everything together too much. bigger batch sizes are smoother and more stable but obviously that averaging means it might not learn the style as well
>i personally think batch size 2 learn style better than batch size 4
that's because it does
>>
i want to steal another AI user's style, what is the best way to do that without metadata
are there any tools that can extract what artists are in a particular AI mix? and dont even bother whining at me about stealing or whatever, all AI is theft so just tell me what tools i can download to do this
>>
>>8873655
yeah train a lora on it nigga
>>
jav lora bake smelling good.
>>
>>8873658
if only...
>>
File: 1768477583665108.png (2.33 MB, 1152x1408)
2.33 MB PNG
Splotchin
>>
>>8873655
Post it so we can steal it together ~ 3~)
>>
i was gonna gen but now there's no cfg left. thanks a lot.
>>
>>8873664
Hai! Show bobs and vegana plz
>>
no one wants to see fried pusy
>>
>>8873680
i do
>>
i want a smoked pussy
>>
File: 1752481087491226.png (2.34 MB, 1152x1408)
2.34 MB PNG
>>
i come back to 500+ new posts
what yall meltin down about now
>>
it's the spammer
>>
cucks
>>
cuck general chuddie
>>
>>
>>8873559
everything is a la carte now via some megacorp, doesn't matter if people upload infinite garbage if no one is looking at it
>>
>>8873644
gradient accumulation is a way to train at higher batch size at the cost of speed. you don't need to do that with a 4090.
>>
>>8873738
apart from the awful style, what is his secret behind these most retarded expressions i've ever seen
>>
>>8873751
it does have that cool abstract art of a guy with his face split into 4 pieces wearing a bowtie on the wall
>>
>>8873751
actual autism
i am fairly certain he was the guy who occasionally would post some drivel like "i hate it when characters' expressions are over exaggerated" on a totally normal image
>>
>>8873738
>prompt masterpiece
>a random framed portrait appears somewhere in the gen
Why is the AI like this, even nais :DD has this problem
>>
>>8873756
because on those datasets "masterpiece" means an actual masterpiece painting or something, most of the time, and most photos of paintings have frames
>>
>>8873756
Pretty sure framed pictures appear when you prompt photo \(medium\)
>>
>>8873751
expressiveH
>>
>>8873761
the thing is that he posted catboxes several times and there was nothing unusual, just ricegnat lora
>>
>>8873760
on danbooru stuff
the issue is what you combine tags with a NL dataset
>>
File: ComfyUI_39049_.png (1.87 MB, 1344x1520)
1.87 MB PNG
>>
^ holy shit resonance cascade is back!!
>>
>>8873774
thout she was a brave and powerful with a tumor in er stomach...
>>
>>8873749
I turned it off and it starts trying to use up 31GB at batch size 2, 1024x.
>>
I trained an anima lora on BLACKEDRAW pics, but while the lora itself can replicate the oversaturated blown out look of the dataset I can't get it to mix with animu artists to create sexy 2.5d realism like I could with noob :(
>>
>>8873786
>I can't get it to mix with animu artists
train loras for other artists
>>
oh i thought BLACKEDRAW was an artist's name...
>>
File: ComfyUI_28070_.png (1.29 MB, 1120x1520)
1.29 MB PNG
>>
>>8873788
oh my sweet summer child...
>>
>>8873786
You need to use a scheduler and interleave some steps with the lora on with some steps with the lora off. You can probably google how to do that on the UI you're using, and if you can't, you probably couldn't learn it anyway.
>>
>>8873793
I think lora loading and unloading at each step is gonna be very slow. It's longer than the diffusion step itself.
>>
>>8873795
Yeah, probably why you want it every "some" steps. If he really wants to do that, I think if he tinkers enough in comfy he can get a multiple k-sampler setup that just loads both models (with and without lora) at once and interleaves the steps. Style mixing doesn't work properly even with baked in artists on Anima, it won't work properly with a shitty lora.
>>
File: ComfyUI_28075_.png (1.16 MB, 1120x1520)
1.16 MB PNG
>>
File: ComfyUI_39022_.png (980 KB, 960x1088)
980 KB PNG
>>8873786
works fine for me just changing the lora strength vs artist emphasis
>>
folks really do be creating gopro loras and not sharing ::skull::
>>
sharing your loras is no different from cuckoldry
>>
yeah? this is cuckold general after all
>>
for me it's cucknew
>>
File: ComfyUI_28089_.png (1.53 MB, 1120x1520)
1.53 MB PNG
>>
>>8873813
it's a "character" lora, xoxo Leah
style is just a side-effect, figured it would be of limited use
>>
File: 1768116143147185.png (1.13 MB, 832x1216)
1.13 MB PNG
>>
>>8873824
You shouldn't do both cloaca and anus at the same time, makes on sense
>>
>>8873825
Animal anatomy connoisseurs in chat :O :O :O
>>
>>8873780
you need gradient checkpointing on, those two are not the same thing. you can do batch 3 adamw on a 4090 just fine, that's the max
>>
>>8873025
gopro lora?
>>
never go all pro
>>
File: 3.png (2.57 MB, 1152x1520)
2.57 MB PNG
>>
>>8873862
the logistics of this are weird, she can't be entirely below him like this at that angle
>>
>>8873862
do real girls like sucking on large man pecs?
>>
File: ComfyUI_00726_.jpg (3.22 MB, 4096x4096)
3.22 MB JPG
>>
>>8873866
It's impossible to impregnate aqua and it's sad, I can't gen sex with her
>>
>>8873871
Have you tried applying mana drain?
>>
>>8873871
she obviously has some degree of control over her powers or else she wouldn't be able to get drunk
>>
>>8873874
Great idea
>>8873875
Would a goddess really want to be impregnated by some mortal
>>
>>8873876
that's why it's hot dumbass
>>
File: 00147-1643397137.png (887 KB, 896x1152)
887 KB PNG
>>
File: file.png (1.53 MB, 1536x1536)
1.53 MB PNG
love the little sketches it puts on (simple background:1.1) when i push rescalecfg too high. Maybe i should collect all these and make a lora out of it.
>>
File: t199.png (2.16 MB, 2016x1152)
2.16 MB PNG
Gave bakariso some of my sdxl "gopro" gens last year, and I noticed he finally decided to train on them for anima (that's not the original gopro which was trained on real-life footage though).
It's not nearly melty enough but style is pretty close for how few gens I gave him.
>>
File: t200.png (2.49 MB, 2016x1152)
2.49 MB PNG
Don't know if I want to gen with it, but some of you definitely asked for it, or maybe it was shiposting: https://civitai.red/models/2587208/gopro-anima?modelVersionId=2906432
>>
>>8873901
he finally updated the anima lora?
>>
>>8873903
No that's a separate lora, not mixed with other aritsts.
>>
that's not fucking gopro lora, fuck off
>>
>>8873901
stop fucking firefly, let her some rest
>>
What is Gopro style? I didn't find it in danbooru.
>>
>>8873902
calling this shit gopro lora is cultural appropriation
>>
WHERE IS THE FUCKING REAL GOPRO LORA >:(
>>
>>8873902
man I always thought it had something to do with the camera brand
3D bias, some kind of depth of field camera effect, etc
>>
>>8873910
it does, this is just some fraud
>>
>>8873910
yes, the real one was trained on screenshots from clips
>>
Good ol' praisins and ronuts
>>
no fucking jav lora
no fucking gopro lora
why even live
>>
File: a003.jpg (694 KB, 1536x2040)
694 KB JPG
>>8873910
Original gopro had (fisheye:5.0)
<- Example
But it was like 3 or 4 gens, he just threw that shit away in the trash after.
>>
chikan and onlyfans loras are also lost media
>>
cprussell can we get another realism lora already?
>>
File: t120.jpg (2.01 MB, 3024x1728)
2.01 MB JPG
>>8873916
And this was my shit bakariso trained on.
>>
File: .jpg (168 KB, 888x1184)
168 KB JPG
https://civitai.red/models/2463593/stuck-in-wall

Any tips for training this concept? Can't find optimal settings, it anyway fucks up, particularly when side view, both in 50 and in 80 epochs. I also tried LoKR factor 8 but results were even worse. 1024, 32/16, 34 images, 1 batch size, adamw8bit, CWR, 1e-4
>>
>>8873927
Maybe when you start sharing your lora you can get some tips, mr .jpg!
>>
>>8873928
Open the link and download it dude
>>
loras*
>>
>>8873927
>34 images
low for a concept lora
>1 batch size
2 might stabilize it a bit
>CWR
que?
>1e-4
any reason it is so high?
2e-5 on the official example lora
>>
>>8873902
>here chud, have the watered down scraps of my magnum opus
im good
>>
>>8873935
>2e-5 on the official example lora
cprussell used constant schedule with 32/32 lora so 2e-5 is a bit low for 32/16 lora with decay schedulers
>>
>>8873927
>adamw8bit
lmfao
>1 batch size
rofl
>epochs
KEEEEK
>32/16
LMAO
Just setup validation loss and stop trying to guess parameters and using shitty advice people will give you little buddy, a good start would be to set alpha = 2x dim, you will have to reduce learning rate a bit with this though
>>8873935
>low for a concept lora
I've trained with less and it works flawlessly
>>
>>8873936
You can have my magnum opus as well, I already shared it before, it's this lora at 0.5 strength: https://civitai.red/models/1229025/coloured-pencils-by-huggy
Except I think nobody wants to gen with SDXL anymore anyway besides noodlekek.
>>
>>8873940
But that doesn't say GoPro, doe?
>>
File: t136.png (3.33 MB, 1824x1344)
3.33 MB PNG
>>8873941
Doesn't mean you can't get some kino gens out of it, does it?
>>
>>8873939
nta butt wat is validation loss?
>>
>civitai
>Error getting file
wtf is this shit
>>
>>8873944
no gopro for you lil bro
>>
>>8873943
https://github.com/kohya-ss/sd-scripts/blob/main/docs/validation.md
>>
>>8873945
fuck you!
>>
there is nothing wrong with using SDXL
>>
there is still no choice but to use sdxl
>>
the example anima style lora by big russ himself uses an @ tag for activation, but i question whether it's a good idea. using any @ leads the model to imagine a style even if it doesn't exist. so style loras with an activation @ tags are unable to be downweighted properly because the hallucinated artist tag will begin to take over at low lora weights.
>>
>>8873824
porse hussy
>>
>>8873943
what >>8873947 said, training without validation loss is a fruitless exercise, if you actually care about getting a good lora you will do parameter sweeps until you get the best validation loss, pick a random dim value (start low like 16 if dataset is small), set alpha to 2x that (https://arxiv.org/abs/2410.21228v3), enable rs_lora so you don't have to tune lr when you change dim for the experiments (https://arxiv.org/abs/2312.03732), now you train 3 times with 3 different learning rates, see which one has the better validation loss, use that as a new basis and train other two loras until you find the best learning rate, now train again changing the dim to see which one gets the better loss, you can change the learning rate again to make sure the new dim has a good learning rate but if you kept the learning rate fixed and changing the dim improved loss then its likely fine already (this will only work with rs_lora, else you will need to sweep learning rate every time you change dim), its a lot of training for no results, but the good part is that after doing that once you will now have a great baseline to all the future loras. The images you pick for validation are also extremely important, if you use a image that is just a small variation of one in the training set it will drag down the validation loss (or make it always go down) making it useless, after getting the good hyperparameters you can disable validation and run the final training on the full dataset :)
>>
>>8873960
>trusting loss to determine if something is good or not instead of empirical testing
rorumao
how the FUCK is the computer gonna know what the gen is supposed to look like? it's retarded. loss is only a warning sign if it goes too high or too low. especially for concepts.
>>
>>8873927
batch 1 is not optimal for concepts, i would suggest 3
>>
>>8873958
That's why I don't like tags with activation, yes. But I think the workaround is like this: to weaken the lora first you downweight the tag, if it's not enough you remove the tag, if that's not enough you downweight the lora.
Downside: if you were to just downweight the lora instead, whatever knowledge lora overwrote would return which is a good thing if you want to use 5 loras at low weight simultaneously without raping the original model. If you only downweight the tag and use lora at :1 you don't get that benefit.
>>
loras with activation tags*
>>8873965
>>
>>8873958
this is a non-issue because not only is the style weakened if you lower the strength as you'd expect, what is exactly the point of adding a style lora if you don't want it to reflect in the style?
>>
>>8873958
The @ is not necessary for a trigger word anyway
>>
>>8873927
also, surely you can find more than 34 images
>>
>>8873960
i want to kill myself
>>
>>8873940
Gopro dataset release when?
>>
>>8873973
Ask that gay dude who baked it! It's probably all male anuses and balls:(
>>
>>8873963
https://github.com/spacepxl/demystifying-sd-finetuning
yeah validation loss is good but not that good. also why the fuck does this motherfucker keep recommending alpha = 2x dim?
>>
>>8873963
>loss is only a warning sign if it goes too high or too low
this. only look at loss values if your results look fucky
>>
"heads" as a unit of measurement is very common when drawing
can i invoke "heads" in the prompt somehow to get a more consistent body ratio?
like for instance 6 heads tall is common for cute female characters and going more towards 7 begins to create a more statuesque woman
>>
>>8873963
>how the FUCK is the computer gonna know what the gen is supposed to look like?
With the validation images?
>loss is only a warning sign if it goes too high or too low
Validation loss never goes too low though (unless you set it up wrong with the example I gave of a image that is too similar/a small variation of one in the training set), are you confounding it with training loss?
>>8873975
>why the fuck does this motherfucker keep recommending alpha = 2x dim?
Appendix H https://arxiv.org/abs/2410.21228v3
Appendix Sec B.2 https://arxiv.org/abs/2405.09673
>>
whole lotta yapping with 0 .toml
>>
post lolikon
>>
Even more yapping with 0 .safetensors. Besides lazylora and fire.inc all the rest of you are a bunch of faggots.
>>
>>8873977
i can guarantee you not a SINGLE NL prompt in any local model measured height in heads.
>>
>>8873977
>can i invoke "heads" in the prompt somehow to get a more consistent body ratio?
Do you think this was included in the captioning of any image whatsoever? Prompting for a thing (or for the removal of a thing in the negatives) only works if it was trained with it
>>
is there a different way then to get consistent height and body ratios?
>>
>>8873985
firedotinc lora
>>
>>8873979
>With the validation images?
That's how it already works. This doesn't do shit besides send you down a completely pointless bunny trail chasing numbers instead of images. Again, why the FUCK would you trust a loss number instead of using come of the most powerful imaging and pattern recognition tools in existence, the human eye? The suggestion to train many different times with different settings is a great one, but trying to delegate that to AI is fucking retarded.
>>
>>8873985
train a lora
>>
>>8873985
use the proper tags for it?
>>
told the model i wanted a girl 4 penises tall and it just worked
>>
retarded unless you're a megacorp, obviously
>>
>>8873990
what are the tags for that?
ive been using the danbooru wiki, but i didnt find any height specific tags
>>
>>8873939
>alpha = 2x dim,
Why?
>>
>>8873994
>https://danbooru.donmai.us/wiki_pages/size_difference?z=2
>https://danbooru.donmai.us/wiki_pages/giantess?z=2
>https://danbooru.donmai.us/wiki_pages/tall_female?z=2
Check those and all the related tags to these
>>
>>8873995
This guy read some LLM papers and thought alpha = 2x dim was the best, and that using a lower alpha would cause massive training instability and couldn't produce a good lora or some shit lol
>>
>>8873987
>Again, why the FUCK would you trust a loss number
Do you know how it works? The validation? Could you provide an example or a mathematical principle demonstrating how a human preference aligned lora has a much worse validation loss to justify the validation being such a terrible proxy?
>the most powerful imaging and pattern recognition tools in existence, the human eye?
The eye recognizes patterns in a single image not across a batch of images.
With that being said you are perfectly correct in that the final assessment should be done by eye, not by the loss (the timesteps of the loss also fit at different speeds, so the validation loss value depends on which timesteps you are choosing to calculate it with), but for finding a good starting point there is no better way, all of those shitty loras people post on civit are made by eye inspection, I'm yet to see a shitty lora with good validation loss.
>>
just genned an image on kirazuri with gopro lora...
i'm speechless.
>>
more the like bitchless lmaooooooo
>>
>>8874009
nope. try again. post hentai THEN ask your question.
>>
>gen image with gopro lora & custom lora
bros... image AGI has arrived...
>>
>>8874006
i already told you that loss needs to stay within an acceptable threshold which is essentially what you regurgitated to me, and you conceded my other point too

you are either baiting or a retarded armchair baker, this conversation is over
>>
>>8874009
realism goes on /aco/ and you also didn't post an /h/ image but instead posted some globohomo slop
>>
>>8874012
>loss needs to stay within an acceptable threshold
You have no idea about the difference between train and val loss do you? Incredible
>>
File: Anima_02245_.png (737 KB, 896x1152)
737 KB PNG
I know this is /hgg/ but the thread I mostly lurk is currently slow right now and I am seeing some anima and lora discussion here so I want to ask here as well.
So I am trying to make a realism lora for anima.
I have hand picked 1300 images mostly on the higher end of quality scale.
I have captioned them and prepended them with @real photo keyword. I went with the keyword route because I want it to be easier to use with tag style prompts. But I can change it to "A photo." or whatever if need be.
So the results are kinda disappointing to say at least and I want to see if anyone with lora training experience can offer suggestions.
These are the parameters I used:
5 epochs with 1300 images totaling 6500 steps run under batch size 2. 1024p, AdamW, LR 0.00003, cosine, WD 0.001, betas=0.9,0.99 96 rank 48 alpha, 0.05 dropout, sigmoid_scale 1.3 max_grad_norm 1.0
>>8874010
Better?
>>8874014
Aco is fucking slow and any training discussion is rare.
>>
>>8873960
>best validation loss
What metric is used to determine the output loss? Surely it's not PSNR or SSIM, right?
>>
chun-li is the ugliest hag of SF and the classic design is way better
>>
>>8873975
also, i read this which is interesting but the entire fucking article is based on a really simplistic LINE GO DOWN GOOD. which is retarded. and he's talking about random cropping like it's a fucking revelation at the end. i fucking swear. everyone should know by now that adding extra closeups, etc., is a huge boost to accuracy. this dude is a fucking moron
>>
>>8874014
>realism goes on /aco/
Tell that to half of this thread salivating over real 3dpd negroes fucking their waifu because cartoon dark-skinned male doesn't do it for them anymore.
>>
>>8874022
yeah but legs doe
>>
>>8873935
>>8873939
>>8873964
>>8873970
Okay, thanks guys. Regarding batch size, should I increase epochs to keep steps the same? Like, 3000 steps and 1 batch, 1500 steps and 2 batch
>>
>>8874026
always use steps, not epochs, because epochs varies with dataset. you do generally need to adjust steps for changing batch size. i think you may just be undertraining some? hard to say. if you want to make the dataset bigger and post it go ahead. my credentials are that i've been training concept loras since early 2023
>>
>>8874026
Think of batch size as just running multiple steps together.
Do not modify the number of epochs. Step count should get halved, and that's expected.
Increase LR by 50% if doubling batch size.
>>
>>8874019
1300 images... I would use cprussell's example config and use adamw 2e-5, constant with warmup, rank = alpha = 32 or higher, and train until I get good results
>>
>>8874021
>Surely it's not PSNR or SSIM, right?
No, its the denoising direction, the image isn't decoded to calculate the loss, PSNR and SSIM would be useless here, CLIP score would make a little more sense though
>>8874026
The relationship isn't linear, 10 steps at batch 1 is not the same as 1 step at batch 10, I don't know what the relationship is, the only heuristic you have to work with is that large batches require more epochs (but less steps), they also require a little larger learning rate
>>8874028
>Increase LR by 50% if doubling batch size.
You may overshoot hard though, theres a limit to this scaling, at some point increasing the batch doesn't allow for a higher learning rate, specially with these comically low batches we use
>>
>>8874022
>ugliest hag
pleonasm
>>
>>8874027
I can make up to 50 images maximum. There are so few high quality images for this concept that I had to make them myself doing img2img screens from koikatsu studio.
>>
>>8874030
>CLIP score would make a little more sense though
The reason I ask is that you are effectively delegating the process of visual comparison to a deterministic tool with very, very flawed perception. Little to no validation method comes close to human perception. I'd have a hard time trusting it.
>>
>>8874032
>5k images for the tag on gelbooru
i just don't believe you
>>
>>8874029
Thanks for the response.
>I would use cprussell's[sic] example config and use adamw 2e-5,
My config is mostly based on that, I cranked LR to 3e-5 due to batch size 2
>constant with warmup
I am using cosine with 10% warmup. I don't really train with constant. I guess I can try switching to constant with some decay steps in the end. It might work faster if high step count is needed.
>rank = alpha = 32 or higher
This disables alpha, why should it be disabled on high ranks? I am just rolling with alpha half of dim because I still don't understand what is supposed to be most optimal in practice.
>train until I get good results
Any ideas for ballpark step target?
>>
>ugliest hag
Like they need a competition for that. They all win.
>>
>>8874034
>>5k images for the tag on gelbooru
>50% of front view
>50% of just the lower body on the white background
I scraped whole gelbooru and rule34 pages, this is how things are.
>>
>>8874035
>I cranked LR to 3e-5 due to batch size 2
cprussell's config use 2e-5 at batch size 4, if you use lower batch size then you should lower the lr, not crank it up
>train until I get good results
Set epoch to 1000000000 or something and test every 5 epochs until you get good results. You don't need to set step target. That's why i recommended constant because who knows when the lora will be done training with a 1300 images dataset
>>
>>8874037
ah, i see. you weren't being specific enough. this problem could have been solved if you had just posted your dataset
>>
>>8874039
tdrussell's config use gradient accumulation, does the LR-batch size relationship apply the same way here?
I interpreted it as batch size 1=0.00002 lr, now I realize I don't know for sure either way.
>>
>>8874043
>Gradient Accumulation is a tweak that does X gradient Accumulation steps for each batch, as such, you can treat it like an extension of your batch size with minimal increase in VRAM usage. For example, Gradient Accumulation steps of 2, with a batch size of 4 would give you 2 batches of 4 per step, effectively making it a batch size of 8.
Global batch size = gradient accumulation x gpu batch size (x number of gpus)
>>
File: 1778074531463321.jpg (393 KB, 1735x1270)
393 KB JPG
Neat, in /adt/ someone shared a new front end for Comfy alongside a full and free Anima artist style list >>>/g/108765823
>>
>>8874049
I'm glad they picked /adt/ to share that, it's a good thread.
>>
File: .png (2.51 MB, 1280x1791)
2.51 MB PNG
>>8874042
I'm gonna scrap at least 40. Damn, there were masterpiece frames in haitokukan's doujinshi. So much cum has been spilled. Kami be with you...
>>
>>8874049
>new
that's literally just NAI's UI. lol
>>
>>8874049
>front end for Comfy
just use swarmui
>>
>>8874049
Thanks, will check /adt/ later
>>
>>8874049
Why does every gen in that style explorer have fisheye effect?
>>
>>8874019
Anima can already produce realism with just a prompt, though it doesn't look great. I would suggest using such a prompt in your dataset captions, along with a trigger tag.

Don't fuck with training settings when you're unsure about the dataset. You should only change one thing at a time, else you won't know which one helped or harmed the result.
>>
>>8874049
>Quality prompts applied
Oh so that's why it doesn't look like Tsunako
>>
>Anima can already produce realism with just a prompt,
I know but
>I would suggest using such a prompt
I don't know any that can consistently do that.
Also wouldn't this just weaken the TW?
>You should only change one thing at a time,
I know but trying everything one by one is going to take a really long while.
>>
>>8874058
>Anima can already produce realism with just a prompt, though it doesn't look great.
its 3dpd
>>
>>8874064
I am an imbecile and as such I forgot to quote >>8874058
>>
>>8874065
The fact that it often makes weird cgi like images for realism prompts is another reason I wanted to create a @style.
>>
>>8874019
>I have hand picked 1300 images mostly on the higher end of quality scale.
make a torrent of the dataset and link it
>>
>>8874073
For license related reasons, this is unfortunately not possible.
>>
LOL
>>
>>8874074
>license
who the fuck cares lmfao
>>
File: 1748948911476473.png (2.85 MB, 1248x1824)
2.85 MB PNG
>>
>>8874019
I only made one realism experiment so far, in >>8872858
where just "photo" became a reliable trigger tag for the style
>>
>>8874073
what could you possibly be training on for this to be an issue..
>>
>>8874076
Not you and me but I if I succeed at training it I might upload it somewhere and if it receives sufficient attention people who may or may not own rights to certain images in my dataset that may or may exist might not like it.
Very low chance I know but I am paranoid.
>>
kekmao what a cuck
>>
>>8874080
Are you new to AI? If someone asks, you didn't train on that dataset. Somebody else shared it on 4chan who was absolutely not you. All that photos you trained on you took yourself with your gopro hero.
>>
>>8874079
I am not training on anything illegal but intellectual property can get uncomfortably serious sometimes.
>>8874081
You can call me a cuck million times. Better than finding myself on court.
>>
>>8874079
ponyxl outputs
>>
>>8874080
4chan is an anonymous website. Also you are not the only one who trains realistic loras/finetunes for anima
>>
>>8874080
lol are you serious. you have to go back to l eddit. thought someone would save me the trouble of having to scrape some high quality jav and take screenshots
>>
>>8874086
there's apparently an /r/ thread where most real people loras end up
>>
>>8874087
>taking screenshots from videos
Ewwww they have chroma subsampling you know?
>>
There are JAV (and not only JAV) photosets of every single porn bitch in existence in 8K+ on pornolab.
>>
Is there any Kirara Asuka dataset?
>>
File: Gaveupmu.jpg (399 KB, 1856x1280)
399 KB JPG
>>
>>8874091
Many of these photosets are same JPEG 80 or whatever quality image you can find on the internet, just much higher res but the quality isn't particularly top notch when you zoom in.
JFYI.
I am super autistic about quality and I couldn't find any place in the internet that delivers consistently good quality NSFW images (SFW is easier to find.)
>>
>>8874096
bruh. you made it sound like you were training on your own photos as a photographer but you're just an actual retard lol
>>
>anima lora fails to pick up grain/brush pattern
How big a vae do we need???
>>
>>8874099
I don't think you can train a model to reproduce literal noise. Maybe I'm wrong.
>>
>>8874100
>the way artists draw textures is random and there is no consistent stylistic differences between artists
Are you serious anon?
>>
>>8874100
A lot of bigger models can like z image, but i always found smaller models struggle with this kind of stuff
>>
>>8874102
He/You also mentioned grain though.
>>
thread challenge - male focus, women out of frame, large cock
>>
>>8874105
done >>8872568
>>
>>8874105
Almost every gen on /trash/ is huge cock focus, why not use the appropriate place?
>>
>>8874098
He's aiming to become to next big Civitjeet please understand
>>
>>8873864
i'm not super buff but i've had women do that often, i didn't think it was that common either
>>
>>8873864
She's sucking on a nipple though which stimulates your precum production and makes you hornier.
>>
why do some artists cover their art in an obscene amount of chromatic aberration
>>
>>8874114
hides imperfections and adds aesthetic™
>>
>>8874114
I did it along with painterly texture to hide SDXL's awful 4channel VAE. Not sure why real artists do it.
>>
>>8874115
im going to train an upscaling model specifically to remove it out of spite
>>
>>8874117
>im going to train an upscaling model specifically to remove it out of spite
there are like hundreds of those with the purpose of fixing that already
>>
>>8874118
there are zero for removing chromatic aberration alone
>>
>>8874084
>You can call me a cuck million times. Better than finding myself on court.
you're not a cuck, just an idiot. there's a difference
>>
>>8874114
Same snobbish reason photographers use shitty cameras that deliberately add that effect.
>>
>>8874077
nice
>>
>>8874074
>>8874080
You should fuck off to Civitai then
>>
>>8874126
the most ironic and amusing part about this hobby is you have these people that beg for advice, use tools developed by others, and train on data created by others yet dip like rats when asked to return the favor
>>
>>8874126
He already did after getting all the answers he needed:3
>>
>>8874129
Well, another piece of garbage added to a sea of it
>>
>artist only has 18 images worth training on
this bake might be scuffed...
>>
>>8874135
>pad dataset with cropped close-ups
>train lora
>gen with lora
>add to dataset
simple as
>>
>just jeet the fuck out of your dataset bro
>>
Cropped close-ups are fine tho
>>
>>8874140
did i say use wai and add a few hundred images to the dataset? no. datasets on ai gens train perfectly fine, stop being retarded
>>
There is still no amen pose \(meme\) for noob vpred btw
>>
>>8874151
lora*
>>
>>8874151
Everyone keeps their .safetensors to themselves here, you'll just have to learn to bake.
>>
>>8874153
I will never be a nerd who masturbates to graphs and numbers
>>
>>8874151
after using anima and its 16ch vae i've lost all motivation to train anything on noob
>>
>>8874156
Same, even though I like noob styles and style mixing better (and gen speeds too) I just can't go back...
>>
anima loras soak up styles like a sponge. you dont need noob
>>
>>8874158
yeah but he has a point that anima style mixing is nowhere near as effective as noob. still though if i have a style on noob i actually like i just train it for anma
>>
>>>/trash/82673414
>>>/trash/82673506
New business model just dropped
>>
>>8874151
You do not need a lora for that, just prompt ;^)
>>
>>8874163
You need it for the consistency and correct hand pose
>>
>>8874163
>>8874164
yeah best i could get is like 1 out of every 5 gens with proper anatomy on the hands with that pose. also anybody have a link to that lora still?
>>
>>8874151
Just generate something else bro.
>>
>>8874177
But the memes...
>>
File: 00081-3302474050.png (1.94 MB, 1248x1824)
1.94 MB PNG
how do i ejaculate from my dick with such vigor like this?
>>
>>8874151
I could make one but I'm not sure I care enough to spend 5 minutes making a dataset, 10 minutes to train it and then another 5-10 minutes testing it, especially since testing it requires me to swap models and fiddle with settings.
>>
>>8874179
Zink, abstinence and edging.
>>
zinc
>>
>>8874181
But you can shitpost and argue about parameters and papers for hours?
ok cool...
>>
>>8874179
by watching lolicon
>>
what's the difference in approach between training a style and concept anyways (besides dataset)? it's been literal years since i've bothered training a concept
>>
>>8874189
concepts are either significantly easier or significantly more retarded depending on complexity with no real in-between.
>>
>>8874185
no that's what you do while training
>>
>>8874189
For concepts you need pics with as wide style variation as possible to not bias the style too.
>>
>>8874194
that goes without saying--my question was geared towards parameters. do you just use the same for style training?
>>
>>8874185
fine whatever but I'm not testing it.
https://files.catbox.moe/q6i5in.safetensors
amen pose, bent over, from behind, ass focus
consider dropping weight to like .8 or something and you can probably drop it lower when upscaling, but fuck if I know for sure I'm not testing it.
the dataset took longer to make than I wanted because everything on gelbooru is fucking AWFUL. just about everything is done by a garbage human being of an "artist" that only draw overweight hags.
anyway if it's a garbage lora that doesn't work well you have no one to blame but yourself for begging.
>>
>>8874189
style loras are the easiest (just lower lr and add more epochs), characters are in the middle, and concepts are the most difficult.
>>
>>8874205
that didn't answer the question at all lol
>>
is there anything better than iopaint for removing watermarks from datasets
>>
>>8874189
>besides dataset
nothing
>>
>>
>>8874220
Sir, i can't self-insert to this pov because the cock isn't vantablack.
>>
Oh, noodlekek stole this too >>>/e/2953567
Well, his 3days were deserved.
>>
>>8874225
holy fuck are you the art police
>>
>>8874225
doing gods work anon
>>
>>8874225
and that was ages ago too like WTF are you doing my guy
>>
>>8874225
exposed his ass
>>
>>8874203
Thanks sir! pls upload to civitai so i can test it :3
>>
Now I know why he doesn't have a pixiv. He'd be called out like this every day because all his 5000 images are unoriginal lazy cheap copies!
>>
File: 00194-257893603.jpg (911 KB, 1536x2560)
911 KB JPG
I need to gen KAngel or Yuki screaming into the void or something
>>
noodlekek kiwifarms thread when?
>>
File: file.png (25 KB, 784x191)
25 KB PNG
HUH??
>>
how big of a rock do you live under
>>
I legit had no idea what it was that
>>
>>8874225
how do you steal a gen
>>
deadnooodles is turning in his grave right now
>>
>>8874064
>I know but trying everything one by one is going to take a really long while.
do you want to know or not?
i spent 8 fucking months testing shit to make a passable stasis tank lora on sd1.5
>>
>>8874114
it's a fad
>>
>>8874203
Well it's not that different from the anima lora, only complain I have is that it's not that flexible with the perspective of the pose but is not really an issue for what I intended to do so, thanks
I can work with this
>>
>>8874239
Take someone's hand-drawn image and i2i with 0.2 denoise - literally every single noodlekek's "art"...
>>
>>8874222
it's my pov, you can be jealous
like an indirect ntr
>>
>>8874248
imagine using i2i for pose transfer and I sometimes doodle (heh) my own references
>>
>>8874250
Nobody's gonna believe you. Every image that has no reference to another's art is just an art that hasn't been found yet.
>>
>>8874246
I mean, take a look at gelbooru and look through the potential possible dataset. there really isn't much flexibility in perspective like you'd want to think there might be. And there's a good chance any sort of variation you might discover is going to be tainted by fat hag disease which has a good chance of ruining the whole train by forcing a bodytype.
>>
>>8874252
True, true, I consneed

>>8874253
Hmm... I can try to do some gens with anima with a little more variations of the perspective and more cute and funny proportions
Would be fine if I do them at base res, right? no need to upscale those at all right?
>>
43 imgs are enough, I tired. Kabeshiri dataset for those who want to try to train one

https://files.catbox.moe/3i45qw.rar
>>
>>8874253
that's the point of AI retard, it can fill in the blanks and adapt a concept to go beyond what it was trained on... if you trained the lora right instead of just downloading a toml from somewhere and thinking he can do the right without studying
>>
>>8874257
>>8874042
>>8874027
Here guys
>>
Just so a genner on /trash/ asking a session link from requester so he can keep genning more for him in return for more praises.
>>
???
>>
>so
saw, what's with all the typos today
>>
Baker-sama, please check the oven as I think the jav lora is done!
>>
>>8874263
Still 25 minutes to go
>>
>>8874259
>one furry image in there
bruh

>check booru
>plenty of perfectly fine images in the first 10 pages
?

>tagged with, i shit you not "girl is stucked in wall" in the middle of the prompt
man you are shockingly incompetent
>>
LOL
>>
kidney inflation lora when https://www.pixiv.net/en/artworks/144465726
>>
>>8874256
synthetic data isn't a whole lot better. either way if it works better on anima you're probably going to be better off genning base res on it and controlnetting it while using the lora as reinforcement.
either way that lora was only 512 steps at batch1 with a dataset of 4 images. While I did keep the dataset as varied as I could(has two cropped ankles, two full body, one upskirt, one long dress, one pair of jeans, one pair of bike shorts, one has character looking back), I still really don't know what you'd actually want out of variation. It's a really stiff pose and 99% of artists aren't actually doing anything to it other than giving the girls a texas diet.
>>8874258
I am not putting in more than the absolute bare minimum amount of work for a LITERAL dumb ass meme lora that I don't even want to use.
>>
>>8874269
I am not clicking that
>>
>>8874269
e621 sisters, our response
>>
>>8874269
why are real """artists""" like this
>>
>>8874269
eva beatrice is cute but that's pretty ugly ngl
>>
>>8874265
>plenty of perfectly fine images in the first 10 pages
It's too far from the truth.
>agged with, i shit you not "girl is stucked in wall" in the middle of the prompt
And?
>>
>>8874279
you're supposed to use an already existing tag and then put it at the start and click the "keep tags" setting, idiot. that's why it didn't work. you did a pretty good job curating the dataset.

ESL grammar doesn't help shit. stop doing it.
>>
And learn some English or make an LLM proofread your shit, retard...
>>
>>8874281
why are you a racist bastard you motherfucker
>>
>>8874281
That concept doesn't have own tag actually, "through wall" is too general
>ESL
I don't care nigga
>>
>>8874270
>you're probably going to be better off genning base res on it and controlnetting it while using the lora as reinforcement.
that's what I thought as well
thanks anyway
>>
it's fine, i'm a man of my word, you'll get your lora tomorrow morning, just know you're a fucking retard who couldn't even be bothered to learn the basics even civitjeets have figured out
>>
>>8874291
I don't force you to do it dude, don't think you're doing me a favor, i just share it
>>
>>>/trash/82676580
I just checked and this funny gen is a completely original composition. Unlike noodlehack's fraudulent "art".
/trash/ > noodlefraud
>>
>>8874276
true, but then would you be any better if you could draw
>>
why are you crossposting your own gens you moron
>>
>>8874264
finally can't wait for my 3dpd slop!
>>
>>>/trash/82677187 OK if this composition is also original I'm moving to /slop/.
>>
>>>/trash/82675214
aaah, ahh, mistress~
>>
I've been genning some more pov sauna and I keep running into weird seeds where the floor is covered in water. Looking through the booru there's a whole bunch of pics tagged sauna that look more like hotsprings/onsen
https://danbooru.donmai.us/posts/11139352
Is that a real thing or just users being dumb?
>>
https://files.catbox.moe/6hkn02.png
>>
head way too small
>>
>>8874305
looks very pregnable
>>
>>8874304
no tagging is just inconsistent
>>
>>8874304
anyone who actually believes *booru tags to be universally "high quality tagging" is an absolute dumbass.
>>
baking a style lora rn
>>
>>8874305
can you make the legs thinner
>>
File: Untitled.png (409 KB, 587x768)
409 KB PNG
>>8874220
>>
chat, can we hit the fabled 3k
>>
>>8874305
Serious question: what's with grey spot on her ass?
>>
nyo
>>
I can keep dunking on noodlekek's ineptitude for another 1000 posts, yes.
>>
>>8874320
aids
>>
>>8874318
wow, highlight material
>>
business idea: penislights
>>
>>8874322
why are you like this
>>
File: 00003-1258630568.png (2.22 MB, 1344x1728)
2.22 MB PNG
>>8874319
that record has been broken here for years
>>
we need a tuggin’ leaderboard
>>
>>8874320
mongolian spot. it's a bruise-like birth mark that shows up sometimes for asians. There's a bit of a weird meme that westerners who would try and adopt asian children would see it and think it was child abuse.
it's a meme in the series that marii has one. here's a random webm of a scene where it becomes the focus of a joke.
https://files.catbox.moe/mqunhc.webm
>>
stop sexualising my huge black cock
>>
so about that jav lora...
>>
>>8874328
don't remember it ever crossing 3k, it was close during noob releases
>>
getting to 3k is a sign of shitposting decline not when there would be interesting stuff
>>
>>8874334
no shit
>>
im trying to have noobAI write something specific as bodywriting and i've seen a few gens with accurate text but i cant figure it out. Is there a specific detailer or something?
>>
>>8874339
do it with photoshop
>>
it's nai
>>
yeah im thinking nai won
>>
>>8874339
i think only novel ai can do text
>>
I can do text <--- look
>>
>>8874345
this post was generated with novelai
>>
>>8874345
waow
>>
how he make text wtf
>>
>>8874340
I know thats an option but i could've sworn ive seen gens with accurate text from the get-go.

>>8874344
Doesn't nai prohibit NSFW gens though?
>>
>>8874349
you can get some text with anima just prompting for it but for SDXL you'll need to use an external tool
>>
https://files.catbox.moe/4zkfeq.png
>>8874349
>I know thats an option but i could've sworn ive seen gens with accurate text from the get-go.
if you mean like simple short words then with enough gacha on noob you can still get it to roll it once in awhile but nothing consistent
>>
>>8874349
novel ai is better at nsfw than any available model right now
>>
>>8874350
>>8874352
thats kind of ass, thanks anyway anons.
>>
>use two loras on their own
>coherent gens
>use them together
>extremely melty gens
cant have shit
>>
>>8874358
seems like it's time for a dual lora then
>>
>>8874358
nai is only $25 friend
>>
File: 00005-2805185524.png (1.32 MB, 1248x1824)
1.32 MB PNG
>>8874357
>thats kind of ass, thanks anyway anons.
i mean what did you expect? on noob if it wasn't raw txt2img luck it was most likely inpainting which is pretty easy anyways
>>
>>8874361
box?
>>
https://gelbooru.com/index.php?page=post&s=view&id=13993908 how do I describe this with NL?
>>
>>8874364
wtf
>>
>>8874365
Oh sorry, I know pee is not everyone's cup of tea (heh), replace id with 13993909 if it's too unpalatable.
>>
File: 00006-2805185524.png (1.94 MB, 1248x1824)
1.94 MB PNG
>>8874363
https://files.catbox.moe/z7kx1s.png
i haven't uploaded the loras yet
>>8874364
those piercings are disgusting
>>
>>8874368
i don't agree with you
>>
>>8874369
the nipple piercings are weirdly placed
>>
File: 00008-2805185524.png (1.61 MB, 1248x1824)
1.61 MB PNG
>>
>>8874372
But this is not mine
>>
>>8874373
check your balls
>>
File: 00074-2286396323.png (629 KB, 1088x1536)
629 KB PNG
>>
>>8874380
stylebox?
>>
how can someone find this style arousing?
>>
ask his thousands of loyal x followers
>>
>>8874382
not going to be of any use to you for style since I'm not uploading the lora but if you're a promptlet who can't get a girl stuck in a hole in a wall for some reason I guess it'll be useful.
<https://files.catbox.moe/pktmm2.png>
>>
File: .jpg (373 KB, 1536x1536)
373 KB JPG
>>
>>8874386
i was just curious if it was firedotinc. bakariso's lora is good enough anyways
>>
>>8874387
my pov
>>
>>8874386
>who can't get a girl stuck in a hole in a wall for some reason I guess it'll be useful
Now try from side or from above
>>
>>8874389
are you raiden shogun?!
>>
>>8874389
You will help her raise her dark kitsune cubs
>>
ugh you're all so fucking disgusting
>>
File: .jpg (111 KB, 1282x973)
111 KB JPG
>ugh you're all so fucking disgusting
>>
>>8874390
depends how from above/from side you want to go?
<https://files.catbox.moe/ki2ib1.png>
<https://files.catbox.moe/ix89wi.png>
if you're wanting like, 90 degrees from side you'd have to try and do that manually for the most part anyway since there's a whole of like, 3 references or something for that on *booru anyway.
>>
>>8874394
anal
>>
https://civitai.red/models/2604674?modelVersionId=2924690
Canny ControlNet-LLLite for Anima
>>
>>8874392
anything for my queen
>>
>>8874368
thanks!
>>
File: 1763340957705041.png (1.8 MB, 1152x1408)
1.8 MB PNG
>>
>>8874404
>digital camera
>in teyvat
>>
why are my loras melting
>>
>>8874408
it is really hot these days
>>
just me and you in here huh
>>
i am you, take the pills
>>
(You) are (Me) and baba is (You)
>>
then who is the bvll?
>>
(You) and (Me) and Her under Bull
>>
File: file.png (77 KB, 239x267)
77 KB PNG
>>
>>8874418
Naisma mugen shitmix when?
>>
File: 1762447995072465.png (1.3 MB, 832x1344)
1.3 MB PNG
https://civitai.red/models/2604424/nova-anime-am
your new slop finetune, sir
>>
>>8874419
ask the bvll not me
>>
File: 1751693076673825.png (1.78 MB, 1152x1408)
1.78 MB PNG
>>8874407
>>
nice. very ayakon.
>>
>>8874425
Very dmd 4 steps, sir
>>
>>8874421
i'm good, brownie
>>
why do browns love that slop style so much
>>
it's our coprophilia culture, chuddite
>>
send bobs and vagene
>>
good anima gens doko
>>
on my pc
>>
i keep them under myforeskin, so that they're safe from the americans
>>
what stage of penile cancer you at rajesh?
>>
mental illness
>>
>>8874421
Try this one instead, it really makes the colors pop
https://civitai.red/models/2596298/
>>
just popped my anal cherry
>>
>>8874458
Looks like the best anime checkpoint ever. I cannot wait to try it.
>>
only miqoanon knew how to really make the colors pop
>>
does it hurt when u get your boob sqeezed?
>>
>>8874471
depends
>>
>>8874474
show bobs
>>
why is there so many videos on twitter about indians dying on dumb ways, it was oddly disturbing at first but now it's kind of funny
>>
>>8874479
they record themselves doing it
>>
>>8874479
there are too many of them and they began to infest every corner of the world and everyone is becoming sick of their deprived nature
>>
>>8874479
There is a billion Indians, they all have smartphones they are constantly filming with, they live in a country with shitty infrastructure and they seem to show extreme apathy towards death, be it their own or someone else's.
>>
>>8874479
if you interact with specific types of content on twitter the algorithm adjusts and feeds you more of that content.
"interact" is as benign as expanding an image or letting a video play for a few seconds.
so the reality is is that you started watching them of your own accord and now twitter is just facilitating and feeding you more.
t.have never seen a single video of an indian(or anyone else for that matter) dying on twitter at all.
>>
>>8874479
they're "people" obsessed with literal feces, what do you expect?
>>
>>8874483
I know i know but twitter algorithm is really aggressive jesus, I was enjoying my feed of cute and lewd chinese cartoon girls getting cocks of all sizes and then a single video of a dying indian changed it all

>>8874480
>>8874481
>>8874484
it was really interesting to see how many of those could have been easily avoided, some felt like legit suicides but turns out people there are just like that, crazy
>>
>alpha = 2x dim
I've read a lot of dumb shit but this one is so dumb it has to be intentional trolling (not allowed outside of /b/).
>>
>>8874500
Definitely a perversion of what alpha was intended for originally. But I can totally imagine it working out because of some weird AI quirk.
>>
>>8874049
>we have invokeui at home
just use invoke at that point, jesus
>>
>>8874503
Isn't invoke dead now that Adobe has bought them?
>>
>>8874504
no? the last stable ui release was at the end of march
>>
File: file.png (43 KB, 1299x206)
43 KB PNG
>>8874504
>>8874505
the paid service (basically just renting their hw and using an invoke instance hosted on their hardware) is what's gone, the actual ui you can host yourself like every other ui ever is still alive
>>
fucking finally i can transition to anima
https://civitai.red/models/1171727/
>>
>>8874503
fuck off kurumuz
>>
>>8874516
memes aside, was it difficult to reproduce this look before?
>>
>>8874518
no, just spam all the shitty disgusting pony score tags and you'll get same looking slop
>>
it already works okay without a lora but here you go anyways idk

https://litter.catbox.moe/9cps7v7s3k8wo0l5.jpg
https://litter.catbox.moe/hkquz8a2g1nqvbw1.safetensors
>>
>>8874519
Not true at all, the scores go towards a western painterly look. https://litter.catbox.moe/5uud3mwteuj2vbxm.png
Same as they did back on Pony btw, at least before applying autismmix bias. This is much closer to the "AI-Generated" tag.
>>
>>8874516
Looks nothing like the original ntr mix >:(
>>
>>8874526
what the fuck is this guy mixing? Are there finetunes of anima coming out enough that there's material to mix, nowadays?
>>
>>8874528
Loras obviously.
>>
>>8874529
And people just accept that shit? Couldn't you do that on the fly if you really need (ai-generated:1.2)?
>>
>>8874528
There was this >>8872861
The Nova and WAI guys also claim to be doing finetunes now. I assume tiny in scope, but they can't really make merges when there's nothing to merge with.
>>
>>8874531
Anima is actually comparable to XL in size, plus it benefits from all the transformer tricks people have been collecting out there. Once we get the finetuners going, we can likely see some interesting things coming out.

You just won't see much results from lazy dataset building. Most of those won't modify the currently baked weights much, you'll need to put thought into it.
>>
File: ComfyUI_28214_.png (1.69 MB, 1120x1520)
1.69 MB PNG
>>
kind of sad how nov 15 will mark 3 full years of local waiting for a decent non-slopped model
maybe it'll happen in 2030
>>
>>8874530
Yes majority of people will eat literal shit, especially if it has a popular brand attached, like ntrmix or WAI. This has been exploited for profit in many areas, music industry for instance.
>>
>>8874528
>Took dataset from my favorite ai imgs generated by ntr mix
>>
>>8874530
Not quite the same, but you could use a slop lora at low strength.

I guess the thinking is if you're gonna keep it enabled 24/7 you might as well have it in the checkpoint. Or, there's no thinking at all.
>>
>>8874532
In size yes, but you're effectively training everything in double resolution because of the better VAE. Just kike genning takes about 3x as long, training should too. Except loras don't need nearly as many steps for whatever reason, not sure if that applies to full finetunes too.
>>
anima is actually better than novelai, i used both
>>
>novelai out of nowhere
>>
>>
>>
v5 soon
>>
>>8874553
They will just make it even more "sketch, unfinished". I still have to use v4 for my realistic sexy henti, because they removed the capability from 4.5 entirely!
>>
>>8874551
male womb tattoo hits really weird
>>
>>8874560
all the posters here need one (except me)
>>
>>8874556
it's hard to go even more sketch than 4.5
I hate the usual "realistic" ai gradients but they really went too far in the opposite direction
>>
i love lying on the internet
>>
me too
nai is good
>>
i also have a womb tattoo btw (i am male)
>>
File: .jpg (173 KB, 888x1184)
173 KB JPG
>>8874523
It works pretty fine as mine, but it's still too sensitive to any token associated with upper body composition. It's the problem I tried to solve, but anyway fine
>>
File: .jpg (391 KB, 1536x1536)
391 KB JPG
>>
File: ComfyUI_28240_.png (1.93 MB, 1120x1520)
1.93 MB PNG
>>
>>8874516
Does it know e621?
>>
>>8874571
just put those in negs dumbass
>>
>>8874560
ai is literally the succubus draining us dry it is what it is
>>
File: 1762434336057822.png (2.42 MB, 1152x1408)
2.42 MB PNG
>>
>>8874629
Ayakon's composition and faces:3
>>
(cute covered nipples tho)
>>
imagine trying to ape ayakon of all people and somehow manage to do even worse :skull:
>>
obsessed
>>
>>8874629
finally some cute bois
>>
>>8874630
ayakon could never :3
>>
>>8874516
>fucking finally i can transition
average /h/ user
>>
cucks
>>
>>8874644
where
>>
https://civitai.red/models/2585622/ultrareal-fine-tune-anima?modelVersionId=2925904
Realistic finetune trained on preview 3
>>
>>8874652
let's see
>>
Someone exlain to me how does the same brain can activate neurons for 3dpd and fucking peak 2d girls
>>
>>8874387
pressbox
>>
3d activates my neurons only if it's jav
>>
>>8874656
i could explain, but you clearly wouldn't understand
>>
>>8874652
javlora baker was pulling our leg.. he said 25 more minutes..
>>
>>8874652
Looks like it was trained on 240x320 .jpgs
>>
>>8874661
real life is 240x320 .jpg
>>
>>8874656
Just like with 2d anime girls, I have never had sex with 3d girls irl, so I still have some desire for them
>>
>>8874663
I agree that some attractive 3d may exist, but it looks like 3dpd fags into any humanoid with long hair and vagina, for example gens on that shit finetune page above, they are literally the opposite of 2d
>>
>>8874656
i love women
>>
>>8874668
proof?
>>
i hate women
>>
>>8874672
same
>>
i love women
>>
i love men
>>
>>8874674
same
>>
3d fetish is mentail disorder
>>
i hate men
>>
>>8874670
my mom was pretty nice on the whole
>>
3d fetish is mentail order
>>
>>8874680
proof?
>>
>>8874652
it's kinda blurry
>>
>researching realistic tags
>realistic is more like realistic illustration so shouldn't be used at all
>jav doesn't work and just makes anime style for no apparent reason
uh???
>>
File: .jpg (261 KB, 864x847)
261 KB JPG
>researching realistic tags
>on anime art imageboard
>>
Who doesn't enjoy an occasional sweetonedollar masterpiece on an anime board?
>>
>>8874692
i must investigate
also, it helps with artist negs
>>
oh yeah nevermind the context was that realistic tag is not realistic enough
>>
>>8874693
it's pedophilia so it's anime
>>
>>8874696
the most "photo" ones are "real life" and "photo (medium)" btw
>>
and "cosplay" of course but i already figured that out
>>
Speaking of realism there is already multiple realism snakeoil on c*vitai. The last time I used it you had to use 2 loras iiirc and other shit but it just werks.
>>
bau bau?
>>
>vtumor
>>
we love vtumors here, peak NTR and NTS material
>>
>>8874690
Using NL and starting with "photo of" usually works, also don't use quality tags
>>
our oshi ironmouse doesnt like AI
>>
ironmouse is okay but I don't enjoy her content
>>
who cares about ecelebs following online moral panics
anti ai virtue signaling is trending to the point of being ridiculous
>>
so true sis
>>
people are still melting down about ai use?
>>
>>8874736
you have no idea lol, we are worse than any actual pencil pedo circle out there for some
>>
>>8874737
honestly I thought it would die out as people would use it more, but I guess some people are loud enough to continue the trend
at least around me every normie uses ai either chat or fun image, there is zero outcry
>>
you are meant to hide you use AI now get with the times
>>
File: 1777033641021407.png (1.09 MB, 832x1216)
1.09 MB PNG
>>
>>8874662
real life also runs at 30 fps but gens run at 30 spf
>>
>>8874728
Wonder what Juniper thinks. Stopped watching her since she spent way too much time drawing on-stream and that's zzz content. Shame, she's kinda hot irl too.
>>
>>8874743
Where's Connor and his cuck chair?
>>
>>8874751
more like where's her IV drip
>>
>>8874727
i already tested quality tags and know they're all animecoded

>>8874728
obviously because it hurts her business
>>
>>8874652
NIce, very sugoi!
https://litter.catbox.moe/awuocd85hdigsp4m.png
>>
>>8874752
>he still believes ironmouse was ever sick
>>
>>8874757
Ok i actually want to try this finetune with character loras now
>>
>>8874757
seeing a 3dpd after watching anime girls all day feels like some bizarre fever dream
>>
>>8874757
no way this is AI
>>
>>8874757
I had less plastic-looking babes with 1.5
>>
>>8874758
Tdlr on vtumor lore?
>>
>>8874771
she has some immune deficiency so she gotta stay inside all the time
some people think she's lying but there's no hard evidence either way
>>
File: 1774395256919612.png (1.84 MB, 1152x1408)
1.84 MB PNG
>>
>>8874773
just farted (i am brown). once that air wafts over she's done for
>>
>>8874049
has anyone tried mooshieui? it seems kinda interesting as a long time forge user, it has alot of the useful addons built right in so no need to fiddle with all the menus and extra junk.
>>
wow what an organic post
>>
>>8874776
Meh. Swarmui is more mature as a comfyui frontend. Also this is not a webui so I can't use it remotely
>>
>>8874771
>>8874773
>she has some immune deficiency so she gotta stay inside all the time
actual lie lol
she has kids btw
>>
>>8874773
You are genuinely braindead goycattle if you believe that. But you are a filthy goyslave in any case if you ever followed any vtuber.
>>
why do you niggas know so much about some random ass vtumors
>>
i miss noob artist mixing
prompt editing mixing sucks the fun out of randomness of style mixing and mixing 5+ artists is awful now
>>
>>8874788
>mixing 5+ artists
do browns really?
>>
>>8874786
i'm not parasocial i just like the funny clips and the boobs
>>
>>8874774
nice shallow penetration there
>>
>>8874757
metadatabox?
>>
>>8874397
Thanks for the canny controlnet for anima. I can now make shitposts twice as fast as before!
https://litter.catbox.moe/o9xkh1.jpg
>>
is there a strategy behind activation tags? i notice some bakers substitute numbers for letters in them. i'm training a character and am finding that different activation tag names seem to have a large impact
>>
>>8874838
The "strategy" with using numbers instead of letters is making it intentionally not match the official booru tag, or not match any existing tokens. The assumption is that whatever training the model already received somehow goes against the training you're trying to do.

imo it's very annoying to use and for little if any benefit
>>
>>8874774
catbox?
>>
>>8874795
>>8874845
Probably a gacha roll using "imminent penetration".
>>
File: 00025-3686768009.jpg (513 KB, 3328x1152)
513 KB JPG
>>8874847
It is not a gacha roll, "shallow penetration" is a real e621 tag and yes, you can also get something like that with combinations of other tags
>>
so this objectively proves that e621 dataset is needed for anima
>>
>>8874848
Yes but that genner is on anima for a while, you are the only sdxl-kek left:(
>>
File: 00281-2674952075.png (858 KB, 832x1216)
858 KB PNG
>>8874851
>you are the only sdxl-kek left:(
not my fault anima isn't quite there for (Me) yet
>>
>>8874852
Yeah, because you can't use t2i and have to CN-steal everything.
>>
File: 00326-1875313630.png (875 KB, 832x1216)
875 KB PNG
>>8874853
damn brat, stop it
>>
>>8874852
you can easily get shallow penetration on anima though...
>>
>>8874861
?
>>
>>8874862
reading comprehension?
>>
File: ComfyUI_39343_.png (1.12 MB, 832x1216)
1.12 MB PNG
>>8874770
True I'd rather use my Sambase mix, or Chroma. But it does make a great base for img2img, with Anima's prompt comprehension and full booru tags.

>>8874847
On noob the secret (or one secret) used to be prompting "vaginal, sex, glans" instead of imminent_penetration, because that's poisoned with a lot of shit like https://danbooru.donmai.us/posts/11316513. But such minute details tend to work differently on every new model base. Anima doesn't even recognize glans, but there's probably some NL way to describe it.

>>8874861
Open to suggestions
>>
File: grid-0003.jpg (803 KB, 2688x3456)
803 KB JPG
>>8874871
"just the tip, the tip of the man's penis is against her pussy, cleft of venus, shallow penetration, vagina"
took like 2 minutes in the middle of testing a lora
>>
>>8874872
>shallow penetration, vagina"
vaginal*
>>
>>8874872
How the fuck did cleft of venus become part of this formula?..
>>
>>8874874
that was just a leftover tag i forgot to remove from the gen i adapted
>>
>>8874872
Cumrag users not beating the promptlet allegations..
>>
>>8874878
Not with NL, anyway
>>
decline u.u
>>
shut your whorish mouth
>>
the population of darknessfag should decline
preferably by jumping off a cliff
>>
Literally fucking WHO?
>>
cucks
>>
>>8874841
correct post
>>
>>8874872
real prompter
>>
>>8874874
why wouldn't you prompt for it?
>>
>>8874943
it usually looks weird with penetration
>>
that's what you get for not putting e621 in the dataset
>>
>>8874944
should be okay with just the tip but yeah I guess you're riight.
>>
>>8874946
i gen a lot of loli so i know
>>
>>8874841
>The assumption is that whatever training the model already received somehow goes against the training you're trying to do
im training a character now and am finding this logic to be somewhat true. the character's default outfit isn't distinguished from their alternate outfits which causes bleed if you're trying to train an all-in-one character lora on a character the base model partly knows. and no, models aren't smart enough to know what 'alternate outfit' means
>>
>>8874948
the only useful "generic" tags like that are popular character cosplays (like tifa or someone), hair down, and alternate breast size \(larger\) (which doesn't really work that great without other tags anyway)
>>
>>8874949
a reasoning model like gpt-image-2 would be able to understand it. something like "character A has character B's hairstyle" works fine, it's just the tech we're playing with is vastly smaller and dated. so the tech is there, we just don't have it (yet)
>>
>>8874950
pointless talking about what-ifs for megacorp tech we'll probably never see
>>
>>8874951
we were on track but all the local model providers pulled out to join the megacorp, so there's nothing left releasing anymore. noob can't even find a good base model for their edit project because qwen 2.0 abandoned local
>>
>>8874953
>qwen 2.0 abandoned local
you can blame cumragUI and their api nodes for that
>>
>>8874953
ai still doesn't make money so local is obviously first on the chopping block
>>
there's nothing to chop, lil pup
>>
>>8874958
stelle down latinx
>>
nai sub
>>
>>8874960
if they make a better model
>>
nai v6 soon
>>
Francesca is kind of cute ngl
>>
File: 20260508004737-148951236.png (1.28 MB, 960x1408)
1.28 MB PNG
>>
>>8874845
https://litter.catbox.moe/f88v0c2dittjshys.png
>>
File: ComfyUI job offer.jpg (518 KB, 1024x1536)
518 KB JPG
>>
>>8874969
i thought this was a position in NovelAI incorporated?
>>
>>8874967
this but with tanya
>>
>download character lora
>the accessory details are all melted
hmmm... does that mean it's fried, or underbaked?
>>
>>8874973
it means not enough crops and closeups and painstaking tagging and probably underbaking. accurate character loras are insanely annoying especially for gachashit character designs
>>
wouldn't happen on nai
>>
>>8874974
i only baked on illustrious once. is there no way to mask details on the image to tell it what is what? like gold_brooch and painting a mask over it so the shitty 1b model isn't hallucinating and guessing at what tag refers to what?
>>
nais :DD
>>
>>8874976
Imagine doing that for every accessory in a dataset of 200 images. I would rather kill myself
>>
>>8874973
Also post download link pls
>>
>>8874976
That's generally a non-issue unless it both looks very different from a regular gold_brooch and you need it to be recognized as such, not just baked into another outfit piece. You need close-ups because of VAE compression, just like you need "close-ups" in the form of inpainting/hiresfix when genning such details. That's what causes the melted details, and that's why nobody cares about illustrious anymore.
>>
File: 1757976130570408.png (2.17 MB, 1152x1408)
2.17 MB PNG
>>
COWGIRL POV ME ONE MORE TIME
>>
Can anima gen wingjob?
>>
those gens are boring as fuck
>>
it's bargain bin ayakon
>>
duh
it's not nai
>>
That gen? Don't care for it. And?
>>
no one does
>>
>>8874997
yeah
https://litter.catbox.moe/66iyq62w6xo7xvx2.jpg
>>
>>8875007
looks painful
>>
>>8875007
>no metadata found
fake and gay
>>
>>8875007
twin tower-zuri where
>>
>>8874993
same pony-tier slop for years
>>
>>8875015
at this point one might think he actually likes the look
>>
>>8874999
at least ayakon was entertaining.
this thread has no lolcows anymore :(
>>
File: 1767724161999315.png (2.16 MB, 1152x1408)
2.16 MB PNG
>>8874997
Not really
>>
well if the retarded poojeet says so it must be true!
>>
>>8875027
Why are you like this
>>
>>8875027
I'm pakistani.
>>
>>8875034
GMS!!!!
>>
>>8875034
lol
>>
>>8874943
I don't fear the female vagina in all it's glory like you fags!
>>
Painterly bro gets the hate because his gens are female focused. Like in that just the tip gen the guy doesn't have legs below knees and bro just doesn't give a fuck and posts it as is because 1girl is all that matters. And this makes the gay community of this chat seethe violently.
>>
otoko gens, wuhteva happened there?
>>
>>8875043
true true you are so real for this post sister
>>
i'm actually gonna make one of my doujin ideas fuckit
>>
>>8875043
t. certainly not the painterly spammer
>>
You guys find a way to shit on every single image poster. Kinda impressive how self-destructive this place is.
>>
maybe if you had managed a single non-garbage image in all these years of spamming slop you wouldn't keep getting shit on for it :^)
>>
>>8875053
Composition is fine, just CN it into your favorite style like noodlekek. All he ever learned and he's the most prolific genner of all 4chan.
>>
It's incredible how brown he actually is.
>>
which one
>>
>if I call everyone around brown, maybe they'll think I'm white
lol
>>
>>8875056
noodlekek?
>>
xhe didn't like that one kek
>>
They’re bringing drugs. They’re bringing crime. They’re rapists. And some, I assume, are good people
>>
uh oh, the paki is getting uppity
>>
he's actually a beaner, chud
>>
>>8875064
maybe 3%
>>
Why do you guys hate the retarded brown spammer so much
>>
i don't see any spammers here
>>
>>8875075
It's one anon who doesn't like that there is a genner who consistently gens images with no blown out male asshole visible so he complains about him all day.
>>
>>8875074
curious
>>
>only one anon doesn't like my boring fried slop
kekerina of time
>>
yes we know you hate all images posted in these threads, very original
>>
we like good gens such as noodlefriend's though
>>
>>8875084
Pretty much. Usually if it's boring you just pass over it, not have a mental breakdown from overflowing rage?..
>>
>>8875088
Sounds far too sensible for /hdg2/
>>
>>8875088
Nobody else spams their garbage like that.
>>
>>8875091
this and it's been the exact same shit since pony like come the fuck on
>>
Uhhh, what's wrong with Pony?? Some of my best gens were made with PonyXL v6.
>>
>>8875091
>>8875092
literally nobody else is posting. his 2 fucking gens are the only pics in 7 hours. yeah maybe they're sorta boring but so what? do you need every piece of slop to be an inspiring masterpiece? mental.
>>
>>8875091
i think they're nice
>>
>>8875094
>do you need every piece of slop to be an inspiring masterpiece?
nta but yes? why would you purposefully aim for mediocrity? the entire purpose of art should be attempting to elicit a reaction. the worst thing you could do is be boring. its arguably better to have shitty looking images than one that is completely boring
>>
>>
>>8875084
>kekerina of time
I say this so please desist.
>>
>>8875099
Cute cityscape.
>>
oh yeah i just remembered i updated forge and the stealthmeta stopped working
>>
>>8875097
provide constructive criticism then instead of calling everybody brown for no reason. shitposters here just want to bark on images
>>
>>8875102
Pancho really turned on us huh?
>>
>>8875104
no? it's the extension's fault it's not upkept
>>
>>8875102
>not using MooshieUI and just pressing the "Both" button in metadata section right on the genning page
ngmi
>>
>>8875094
>do you need every piece of slop to be an inspiring masterpiece?
if noodlefriend can do it what's your excuse?
>>
>>8875105
No need to sweep for him, anon. We're no longer a Pancho general. Only touchdown russell.
>>
File: 1748025048738778.jpg (58 KB, 407x405)
58 KB JPG
>>
>>8875099
Hert age?
>>
>>8875107
most of noodlehack's gens are stolen, he is creatively bankrupt
>>
>>8875111
so true. would you like to post another fried pov cowgirl mr creative?
>>
>>8875110
this is a 4000 year old demon
>>
>>8875111
hey retard

all ai is stolen
>>
>>8875112
you've been exposed several times already
>>
noodlekek's biggest fan
>>
I want a cool name for pixiv =\
>>
Is there a controlnet for text so I can trace over someone else's cool name?
>>
>>
>>8875117
same
never came up with anything that sounded right
>>
fire.ink maybe?
>>
dios mio.. el hombre cafe
>>
File: 00028-2445338695.png (3.81 MB, 1496x1992)
3.81 MB PNG
>>
noodlecuck meltie
>>
paki meltie
>>
>>8875126
on the daily
>>
trying out AI genning on pixai becuase I dont have graphics card on my pc, just a igpu. looks better than what I expected after tweaking some stuff and experimenting.
>>
darknessfag status?
>>
>>8875132
c*mfy cloud also offers free gens btw
>>
Is it possible to be an effort genner with pixai? Suppose you got your melted base res sloppa and you want to upscale it but there is a size limit for freekeks. You run your animesharp or whatever locally, then do an i2i on the left and the right part of the image separately, then inpaint the seam. Except how do you inpaint the seam? I can do it with NAI because I can just 100% reroll it, but will it properly get rid of it if you just do 40% denoise or something?
>>
>>8875137
Best to just use Anima checkpoint and not worry about upscaling. I don't know what the pixai limitations are. But you do not need much denoise to fix a seam that's similar colors on both sides, the tone difference is key.
>>
>>8875136
How hard it is to set up? I went with pixai because it seemed like the one with less friction, but aside from that, the credit spending is really steep, and you have to see fucking ads on their app to get a decent amount of credits per day.
>>
>>8875141
it's a node based ui so it may be confusing for beginners, but you can just use available workflows instead of making your own. also a basic text2img workflow is very simple
>>
>>8875137
Actually I just remembered brittle and I think he did 9000 inpaints on base res sloppa for many of his images.
>>
cute otoko no ko dicks inpainted over 9000 times
>>
I did some cute astolfos the other day, yeah
>>
File: 1769442162079081.png (2.24 MB, 1152x1408)
2.24 MB PNG
>>
OVNIS ARE REAL?!
>>
>>8875162
She has her feet in his face doesn't she?
>>
>>8875167
what's it to you where she has her feet, weirdo?
>>
>>8875169
I like feet.
>>
>>8875167
looking at the angle of her legs theyre phasing into the poor dudes chest
>>
>>8875169
Plus if she didn't have feet in his face how did he cum?
>>
>>8875177
maybe her legs are amputated below the knees and the dude has a fetish from playing katawa shoujo
>>
File: 1778252870265174edit.png (3.51 MB, 1528x1840)
3.51 MB PNG
>>8875180
:(
>>
educate yourselves idiots
>https://exhentai.org/g/3925242/a881bb12c2/
>>
>>8875187
kek
>>
>>8875023
looks like a wingjob to me
>>
>>8875188
Thanks, too hard to remember all this though, which AI can I feed this to so it can automatically reference these when generating next time?
>>
>>8875187
Wait how did you do this?
>>
>>8875191
lol
>>
>>8875192
They shout at me when I tell.
>>
>>8875187
what did you use to outpaint?
>>
>>8875194
>>8875195
it's the cuck NAI painterly fag
>>
>>8875195
NAI at free res (outpainted area is slightly wonkier than original because it was generated at lower res).
>>
>>8875198
i spit on you
>>
>>8875200
Jealousy bad.
>>
Anyway, you are all free to outpaint with Anima and show your results!!
>>
File: file.png (1.34 MB, 1152x888)
1.34 MB PNG
>>
I am not a cuck, I don't touch others gens
>>
I also don't want to catch the brown
>>
good job, the creature is angry now
>>
the goblin?
>>
>>8875204
if she farts he turns brown. this is how brown people are made
>>
>>8875162
Can you confirm this >>8875209
>>
File: AHAHHAHA.png (30 KB, 653x376)
30 KB PNG
>>
>>8875210
Yes it's true
>>
:skull:
>>
>>8875211
why is novelai japaense
>>
Good morning sars! ( ͡~ ͜ʖ ͡°)
>>
File: file.png (1.28 MB, 888x1152)
1.28 MB PNG
>>
>>8875215
dis what happens when she sharts instead of farts
>>
>>8875215
finally, a gen by a white man
>>
>>8875211
What is aetherroom
>>
>>8875220
cai but nai flavored.
(they dropped it because they're absolute dogshit at llm's)
(just like they are with post sdxl image gen)
>>
i hate novel ai so much when are they going bankrupt!!!!!!!!!!
>>
>>8875224
I mean they aren't. they still have h100 cluster shit.
they have a really easy exit strategy in that since they own the hardware and rack space they can just rent shit out with super low overhead and it's just free money.
>>
two more weeks
>>
>>8875224
It's the same guy that loves the bbcpred gens but hates the wholesome painterly gens. Very sad individual...
>>
would they open source nai v4.5 if they went bankrupt?
>>
so fucking tired of seeing nai images it makes me sick to my fucking stomach
>>
File: i_0049.png (2.13 MB, 1856x1664)
2.13 MB PNG
i love anime!!!
>>
ew
>>
>>8875225
they don't own shit lol
>>
>>8875233
they own you every single day seeing how rent free they live in your tiny head lmao
>>
>>8875234
why aren't they paying rent if they are so rich?
>>
>>8875235
the rent for that coreweave cluster is too high already
>>
kinda funny that nai would be forgotten here for the most part if the creature didnt lose its shit every time nai is mentioned kek.
just 25 dollars btw!
>>
;^)
>>
what does the cantonese in this image say
>>
>>8875238
why did you crop his asshole?
>>
>>8875239
yess, i love shitting on the bed so much
>>
>>8875239
"noodle, bobblehead and oekaki go in negative"
>>
I want a new noodle baked thread, local only, hold the /hdg/ culture.
>>
You idiots could be on the non cucked website but noOoOOoO you like your cage a little too much
>>
2h2g waiting room
>>
File: 1778259454987153edit.png (2.43 MB, 1856x1408)
2.43 MB PNG
>>8875230
Bro, her anus got split in half and is now peaking from both sides of dude's dick.
Fixed it and cleaned up some other dirty spots.
>>
>>8875248
el basado
>>
>>8875246

why would I go to the website with pedophilia when I'm not a pedophile?
>>
>>8875248
peak
>>
>>8875248
Why are you so fucking racist?
>>
>>8875230
>>8875248
bro what is that cut :skull:
>>
>>8875246
birdschizo
>>
>>8875248
damn, her pussy juices are way too acid
>>
>>8875250
anime is for pedophiles
>>
>>8875257
then why are you on an anime image forum baka!!
..unless ..
>>
>>8875248
You wanna say sex with a corpse is not a dirty spot
>>
>>8875250
make your own board then
>>
>>8875246
treech is dead tho
>>
>>8875261
we were only like 6 people at most there, jordach and tree included
>>
>>8875258
>anime image forum
>can't even post actual anime images
cucked site for cucks
>>
>>8875262
that's way too much jordach per capita
>>
>>8875263
you can post them on /b/
>>
>>8875265
>have a board called hentai
>actually have to go to the random bullshit board to post hentai
epic
>>
could always post on /ldg/
>>
>>8875266
>have board rules that have been in place for 20 years
>cry and scream and shit and piss and cum about them every single day for eternity
epic
>>
"no loli" is a fine rule by me
"no girls that aren't ancient landwhales" isn't
>>
>>8875270
pretty much. but of course we're ignoring that posting girls with less than double D's getting you banned has only been an issue in very recent years because le rules never changed!
>>
>>8875099
>>8874967
>>8874368
>>8873791
how are any of these "ancient landwhales"?
>>
>>8875272
shhhh we are still pretending that loli isnt allowed
>>
>>8875272
you know what, if every single one of them stay (especially last one) I might start posting here again
>>
>>8875272
people have gotten banned for posting those girls before. no one's pretending the jannies are consistent
>>
why was >>8873305 deleted but >>8871645
is fine
>>
small breasts isn't directly equal to loli, you may sometimes get away with it but most of the time they got deleted
>>
>>8875277
noodlefriend>oekakitroon
>>
>>8875277
the creature has a seething hatred for oekakibro's success as an artist and spam reports him whenever possible
>>
>>8875281
>the creature
which creature? you made up 3 different ones by now
>>
>>8875278
yeah the line is completely arbitrary and depends on which janny gets summoned by the shitpost reports
>>
>>8875282
birdschizo
>>
all you
>>
>>8875283
>which janny
I thought /h/ only had a single janny
>>
>>8875286
i wouldn't know or care
>>
>>8875287
Cared enough to whine about it though.
>>
>>8875282
dont blame me that others use that word like that now
>>
>>8875288
some anon made a post and i responded to it as part of the conversation. are you autistic or something?
>>
>>8875286
probably meant mod. I'm pretty sure if something gets enough reports it gets elevated to mod queue and most of them just treat it as an auto-ban for the reported reason without even looking at it.
>>
>>8875290
he thought it sounded cool just let him have that one
>>
>>8875290
Cared about it enough to chime in despite not being a part of the conversation originally.
>>
>>8875277
It's a fucking imageboard
why does every single image have to be hentai
>>
>>8875294
because thats what the rules say? are you retarded or something?
>>
>>8875294
why are you dilating over straight up avatarfagging
>>
>>8875294
because that's the rules you fucking retard
>>
nai creature having multiple melties you love to see it kek
>>
>for free
>>
>pretending to care about the rules in bad faith
you're not subtle
>>
>>8875302
is she fucking herself?
>>
>>8875300
wow this is a new tactic bravo
>>
>>8875298
Is the nai creature this claude guy or the momoira guy or someone else? we need a schizo mindmap
>>
hehe i was just ragebaiting chat!
>>
You are being ironic right now but not giving a shit about rules and laws is extremely brown coded, I hope you realize that.
>>
>>8875309
he does care about rules but he'd rather the rules be whatever he likes
>>
>>8875309
you've used this one before can you try rephrasing it to be less obvious?
>>
>>8875309
>You are being ironic right now but not giving a shit about rules and laws is extremely brown coded, I hope you realize that.
>UK exists
lol
>>
I'm 95% sure I didn't get my first ban for "lolicon" until after naiv3 released bee tee dubs
>>
cucks
>>
>oekaki garbage gets deleted
>meltdown
always
>>
>>8875313
american website owned by a nip.
britbongs, cancucks and upside down prison islanders have no place here.
>>
>>8875314
image gen was awful and nobody with eyes in their head bothered with it before naiv3 released
>>
>>8875266
Could also just post hentai here.
What's the worst that could happen?
>>
>doesn't care when art is destroyed
How extremely pedestrian.
>>
>>8875318
yeah my point is if you have a society of retards that follows all rules without question you become as cucked as the UK is
>>
>>8875320
>Could also just post hentai here.
hilariously ironic when you clearly made a yuri gen and inpainted a dick and cum just so you could post it here lol
>>
>>8875320
you could get reported for being an antagonistic shithead
>>
>>8875322
brownoids don't think that far ahead
>>
>>8875319
Meh. v3 was the first model that could gen actually good pictures, but it was still crazy fun even all we could gen was based64 slop. The tech itself was like magic, even if the pics looked kinda bad.
>>
>>8875322
I thought your point was that "lolicon" is illegal and treated 1:1 with CSAM under british law.
(this is factual, by the way, and is the same in canada and australia - so if you travel to any of those places do not bring thinbooks and make sure all devices are completely media free and with no saved logins or you're one search in an airport away from potential jail time)
>>
>>8875328
>I thought your point was that "lolicon" is illegal and treated 1:1 with CSAM under british law.
that is a part of it yes but that's just an example of a larger problem with the country
>>
firedotmelt
>>
>>8875324
and take a 3-day break that will probably be good for you in the end
>>
>>8875322
That is a surprisingly leftist take for this general
>>
>>8875323
No it's a proper threesome, I just like to watch.
>>8875324
>antagonistic
In what way? What's wrong with contributing images?
>>
>>8875336
>I just like to watch
based NTR/NTS enjoyer
>>
>>8875335
>libertarianism is leftism
this is the result of a generation raised on brain rot
>>
>>8875336
>>8875337
>I just like to watch
isnt that technically all porn?
>>
>>8875340
correct, we all are NTR/NTS enjoyers at the end of the day
>>
the hentai is only straight if there is no women in it, then you can't be cucked
>>
>>8875335
???
>>
>>8875346
he thinks firebombing a tesla dealership is resisting tyranny
>>
>>8875347
yeah seems like it
>>
File: aOAoLYMaAEDAme.jpg (151 KB, 1664x1856)
151 KB JPG
>>
>>8875349
wrong tab, bro
>>
>>8875339
It's the result of reading yesterday's news. All political philosophies are irrelevant new, there is only a binary choice.
>>
>>8875350
I just felt bad with how little image posts there were
>>
binary code
>>
when is someone gonna make a tile cn for anima
>>
there are only 11 types of people in the world that understand binary, the ones that do and the ones that don't
>>
>>8875354
when the model releases
>>
>>8875357
what if it never releases
>>
>>8875354
when slowpokerussel releases the final model or admits that further finetuning is useless and we have to cope with v2/v3
>>
>>8875358
Mat 6:34 bro
chill
>>
>>8875356
*clap clap*
>>
>>8875358
i wouldnt worry about that
when is the last time a local model was promised to release but didn't ever end up getting finished?
>>
>>8875363
anglebottonless
>>
>>8875358
then some other group will pick up the model, inject a shitload of e621 and irl images into it, and call it something related to nai then call it a day
>>
File: 1768383392826553.png (2.28 MB, 1280x1408)
2.28 MB PNG
What a bunch of goobers
>>
>>8875366
slenderman?
>>
>>8875302
>>8875312
>>8875329
>>8875345
>>8875367
janny asleep?
>>
>>8875369
offtopic got deleted doe
>>
>>8875369
From the recent janitorial activity I concluded that janny REALLY dislikes furry!
>>
>>8875367
wait, there is new final fantasy?
>>
>>8875372
no i'm happy about it actually. maybe it'll wake 'em up
>>
>>8875369
>furshit deleted
based
>>
File: 1748489664267237.jpg (249 KB, 600x900)
249 KB JPG
>>8875369
who?
>>
>>8875372
oh no...
>>
pls stop using adam8bit, scrounge enough video rams to accommodate CAME
>>
pulchra until the jannie likes her
>>
>>8875380
came doesn't seem to require more vram than adam8bit for me. also came8bit exists
>>
>>8875380
>pls stop using thing that has been proven to work, scrounge enough video rams to accommodate random snake oil with a prettier graph
>>
>>8875380
I'm only using w8bit because adamw proper doesn't fucking work on this shitty ass easyscripts fork.
>>
>>8875382
CAME is better than Adam8 and converges faster than full Adam
>>
>>8875378
ed helms son, will ferrell daughter, steve carell intersex sprog or seth rogen otherkin offspring?
>>
Why is the anima training script in the sd-scripts instead of the musubi-tuner repo?
>>
>>8875366
Is she having feet in his face now or not??........
Can you make it a male pov so I can be certain?
>>
(plus if you make it male pov you won't need to generate proper male legs you are struggling with so much)
>>
>>8875384
yes it does i just updated even
>>
>>
>>8875391
whats going on with big bro's left hand
>>
>>8875391
based orcbro, I was in fact planning on doing some orc x elf gens later
>>
>>8875392
i don't see it
>>
>>8875394
i don't even like orcs but your gens have nice imagination to them and make me want to copy
>>
>>8875396
looks like his pointer finger is backwards
>>
>>8875398
ehh...i guess?
>>
>>8875391
elf should be male
>>
>>8875400
your post has some words in it but i can't seem to understand it
>>
File: 1778251075520124.png (3.76 MB, 2048x2048)
3.76 MB PNG
Well, well, well, looks like someones finally trooned out and made themselves their own blue general logo(because blue board) >>>/g/108779476 ? Oh, and they went ahead and added /hgg/ to their thread neighbors list, looks like /adt/ came here to stay.
>>
Not my problem anymore lolol
>>
>>8875403
Cute logo, good job?
>>
It seems /hgg/ the famous "lora and tech discussion general" has its days numbered.
>>
>>8875390
nah just ran update.bat and it still NaN's after the first iteration on my end.
>>
>>8875407
that means you fucked up your settings
just restart and test every setting 1 by 1 won't take you more than 20 minutes or so
>>
>>8875402
i'll let my large brown cock do the talking then, bend over.
>>
>>8875403
What is the topic of posting in this new general?
>>
>>8875409
just to double check your retardation, you're training on anima, correct?
because you are quite literally the only and first person to claim that adamW actually works without NaN'ing. so post a toml or fuck off.
>>
why does /a/ not have an AI general yet
>>
>>8875406
People treated it as a joke, but the /h/ general really had been slowly decaying since mid-2024. The thread split only sped things up.
>>
>>8874799
Sorry, but I can't share my BLACKEDRAW lora.
https://litter.catbox.moe/h26dic5a5ehkbs8n.png
>>
>>8875384
>>8875407
i don't see why it wouldn't work. the fork just calls adamw directly from pytorch. are you training in full fp16/bf16 precision?
>>
>>8875419
finally, non-pedo anime
>>
>>8875415
https://litter.catbox.moe/3td47rmxdp3yppuu.toml
>>
when are they going to do a flux2 vae conversion of anima?
>>
>>8875416
the /a/ mod banned it in 2022
>>
where is catgoddess
>>
we used to have a thread per day before splitting
>>
>>8875431
with 90% of the gens as /e/
>>
you can just go back you know
>>
>>8875431
I know
but most posters were only here for the activity, when you cut it in half you end up losing a lot more than just that
>>
>>8875391
Can you make elf ears longer? Orc (me) is gonna want to molest them while splitting her in half, and he (me) would prefer having a handful, not just a tiny little plaything for a finger or two.
>>
it was the fault of the meanies nit picking gens
>>
>>8875437
nyo
>>
>>8875440
I am a sad orc today ._.
>>
i can't go back... i don't have a time machine...
>>
>>8875420
training/saving in full fp16
>>8875424
>apparently training in bf16 makes it no longer NaN
right. this shit's dumb.
well whatever. w8bit can go back to trash where it belonigs.
>>
>>8875431
it's ok. next thread we'll just rename in to /hdg/ what ia schizo will probably seethe
>>
>>8875421
It's about time we take our hobby back from the paedophiles.
https://litter.catbox.moe/j02aeoc97co24zme.png
>>
>>8875443
why did you train on full fp16 if your gpu supports bf16 lol, full fp16 training has a lot of problems
>>
>>8875446
based, don't let pedo tourists claim our hobby
>>
File: 1763454857548888.png (1.44 MB, 1152x1536)
1.44 MB PNG
>>
>>8875451
can you make the sky darker
>>
>>8875451
bro hasn't been unlock yet
>>
>>8875451
MYSTERY NIGGAS
>>
>>8875452
>night
>girl is lit up like a christmas tree
i doubt he can kek
>>
>>8875451
Who's that pokemon?
>>
>>8875418
>thread is active
>thread has lora bakers
>thread has acceptable levels of spam and meltdowns
>thread has no highlights
current /hgg/ is better than the last 2 years of /hdg/ and only falls short of pre #10 /hgg/
>>
>>8875448
because I'm retarded and didn't really think about it. and while in trouble shooting I swapped to w8bit and it worked just fine, which was weird, but with the amount of issues that branch of easyscripts has had I just didn't question it and didn't want to deal with dumb shit anymore.
>>
>>
I want the highlights back
>>
said no one ever
>>
>>8875451
Wow. And the retards here are saying painterly anon is boring and uncreative???
>>
you are just mad you never made highlights
>>
paki meltie continues
>>
>>8875468
but why aren't you in /hdg/. they have highlights over there!
>>
>>8875465
This and unironically. I think highlighter anon did a good job at displaying the best genners. I know some anons get angry when their gens don't get into highlights but we shouldn't pay any attention to their trolling.
>>
>>8875464
Marsbox?
>>
>>8875464
is this the frickin Sailor Mars?
>>
>>8875468
bro every time he put me in the highlights I had to make another gen in the same style to tell him to kill himself
>>
>>8875471
fake news, no new highlights in the diddybluden /hdg/
>>
How come this thread is now more active than /hdg/ if everyone loves highlights so much?
>>
>>8875479
i get paid to post here
>>
>>8875479
this is the question they will always avoid
>>
>>8875479
what does it have to do with highlights? hdg doesnt have highlights either
>>
>>8875479
shitposters need that shot of dopamine we can only provide
>>
>>8875479
>/hdg/ equivalent to why the jews were kicked out of 109 countries
>>
>>8875488
nice try bro this post is too high quality for that loser. the best part is i know he's lurking here reading this too
>>
>>8875462
well adamw8bit worked because it uses the fork's own custom version, while adamw is the unmodified version from pytorch. so you can thank the fork for making adamw8bit work for you.
>>
this board never recovered from the loss of ayakon
>>
>>8875488
yes, please
we need to reward effortposters to spite nogens
>>
having your thread quickly fill up with slop images isn't a mark of pride btw.
>>
This is literally a ai slop general
>>
this is the noodle masterpiece and lora discussion general. the slop thread is next door. fucking immigrants
>>
>>8875451
it's missing that je ne sais quoi
>>
If I wanna see a bunch of random ai images I can just turn on generate forever and wildcards. Only purpose of ai threads is discussion and maybe sharing of loras
>>
>>8875503
personally i got almost 6k of fappable images i saved from anons over the years, it's nice to scroll through them every so often
>>
>>8875503
plenty of times ive seen nice artists and ideas by other anons in older threads. not so much anymore though, nowadays this thread is mostly for funposting
and this isn't 2023, what the fuck is there to discuss anyway? how many repeats and what scheduler for the umpteenth time? wow so engaging
>>
unironically hope we'll get highlights next thread
>>
>>8875488
why are you larping as him? is this really what you life have come to?
>>
File: 1761024171551871.png (2.03 MB, 1216x1472)
2.03 MB PNG
>>
>>8875511
I was more interested in the shadow dude route.
This "getting all worked up over the shadow dude then calling her bf to quench the thirst" route is a bit anticlimactic...
>>
>>8875511
trex arm obscured by leg
>>
>>8875515
its building up to a ntr route
>>
she's like 14 :skull:
>>
>adds aged up tag
ok now what
>>
>>8875520
>old woman, wrinkles, sagging breasts
>>
the real problem is that she's from a 3dpd game
>>
>>8875519
don't tell this dude to look up the age of most female anime characters
>>
anime women are based on 3dpd women
>>
>>8875521
you joke but if you like to bake having a hag gf is a great source of free yeast
>>
>>8875523
anime is for pedos after all
>>
>>8875525
why are hagfags so fucking disgusting
>>
>>8875519
Joe Rogan sisters, our response
>>
https://litter.catbox.moe/50j515.png
>>
>>8875526
Not on my watch!
https://litter.catbox.moe/ba9u9sqiu3ycn8jt.png
>>
>redditbuddy
>>
>>8875532
is he wrong, tho
>>
>>8875530
>4chanbuddy
based
>>
bad day to have eyes....
>>
>>8875531
more like that's ????
>>
>>8875531
didn't know BA had collab with Final Fantasy
>>
>>8875530
me on the right
>>
>>8875530
your handwriting is cute
>>
>>8875542
thanks please follow me on X (formerly twitter)
>>
What the matter pedo, head not bobbled enough for you?
https://litter.catbox.moe/2p6hxdd39hiw3iiu.png
>>
>>8873608
damn that projection was crazy
>>
>>8875530
damn that's a really good way to describe redditors' sexuality.
>>
cucks
>>
File: ComfyUI_39468_.png (1.31 MB, 920x1160)
1.31 MB PNG
>>
https://litter.catbox.moe/rk5qhdaftr49covp.png
>>
>>8875553
please, feed the /aco/ thread
before it dies completely
>>
>>8875211
kekerina of time
I forgot this was even a thing but do recall everyone thinking it would be the ultimate chatbot experience. We just had to let JUST.. WRITE cook. Seems the kitchen is closed.
>>
>>8875454
TEXT NIGGAS
>>
>>8875531
3dpdbox
>>
>>8875554
What request did they not fill for you, buddy?
You've dropped this in at least three different boards now.
>>
>>
>>8875554
Do it yourself.
>>
File: HFGIYxZbYAA4SIC.jpg (30 KB, 625x351)
30 KB JPG
>>
>>8875562
Link the other two?
>>
>>8875554
fuck off, pedo
>>
>>8875531
>karin's face turned into an actual nignog
lol
>>
>>8875569
This, she is supposed to be Okinawan (aryan Japanese with dark skin)
>>
If futa Karin fucked Asuna, would that be blacked?
>>
>>8875570
baka my head people don't know about okinawans
>>
>>8875567
/b/ and /trash/ at least, maybe another /aco/ thread. Sorry, not digging for them. Wait long enough and he'll do it again. Can't help himself.
>>
japs are brown in general
>>
>>8875561
https://litter.catbox.moe/tlp367.png
>>
>>8875575
share it so i can shitpost with it
>>
>>8875573
I was the one asking, if some tourists do arrive it's best to know where they came from.

Those boxes are offtopic here anyway.
>>
>>8875575
Was the "UltraReal finetune" no good?
>>
>>8875583
nta but i found it only okay
>>
Interesting, more ControlNet for Anima are popping up, this time Cabal Research >>>/g/108782301
>>
help
>>
why are localweiners so obsessed with controlnets
>>
>>8875590
We'll see, the previous one was worthless.
>>
>>8875583
It doesn't give me as much of the overexposed shitty camera look and the colors tend to be a little less saturated.
https://litter.catbox.moe/syjlwbmu7qi5in94.png
.>>8875576
https://www.mediafire.com/file/1407i1ihn3dqufz/BLACKEDRAW-anima-test-2.safetensors/file
>>
>>8875591
help with what chud
>>
>>8875595
That's CHUDDIE to you, chud.
>>
chudette
>>
>>8875419
metadatabox without the BLACKEDRAW lora?
>>
https://litter.catbox.moe/l5dtjavwrpo66tzy.jpg
>>
File: file.png (1.98 MB, 1120x1440)
1.98 MB PNG
>>
>>8875603
based futachad
>>
>>8875622
Vaginal penetration looks like genner knows much more about penises than pussies though (typical of /h/?)
>>
>>8875625
let me guess, not messy enough?
>>
>>8875626
No, it looks like there's weird swollen meat in place of labia minora. Also most of it is spread open for no apparent reason even on the outskirts where the dick isn't pressing against it. Makes it look like a rubber vagina that's just always like that with dick inside or without.
>>
pack it up boys, the vagina police is here
>>
>virgin incels argue about what a vagina looks like (they obviously have never seen one irl lol)
>>
>>8875629
I can make it shorter: it looks like shit.
>>
>>8875630
>virgin incel projecting
>>
ultra realistic vaginas look disgusting
I'd rather just see a cute slit/innie
>>
>>8875633
It will have to open up to take your dick. And then you'll see all the horrors of it anyway.
>>
the closer a thread to collapse, the more pedohysteria and black poop gens there are
>>
>>8875633
based cleft of venus chad
>>
>>8875630
i've seen run over animals, tho
>>
>>8875633
>realistic vaginas look disgusting
skill issue https://gelbooru.com/index.php?page=post&s=view&id=13141266
>>
>>8875647
I wish he didn't mosaic Kagami's gaping hole though. I wanted to see it in it's full glory.
>>
>>8875650
>he
it's a female, also nangong yu one is the best but fucked by mosaic
>>
nigga over here talking in tongues :skull:
>>
It's over, I think the shitmix is better than base preview3....
>>
>>8875665
oh yeah? which one?
>>
imagine if there was no decline
>>
>>8875673
kirazuri
>>
>>8875678
shouldn't have bullied the loliposters
>>
>>8875679
it rapes styles so it sucks
>>
>>8875681
just use lora retard
>>
>>8875681
Styles are already fucked from the start. FUCK styles
>>
File: 00118-2691197221.jpg (620 KB, 2304x1664)
620 KB JPG
>>
glad they could poop together
>>
Are they indians?
>>
>>8873547
What's weird is someone who has a beef with anybody using Sambase >>>/aco/9187171 >>>/aco/9187207 and recently Anima >>>/aco/9193750. As a newfag in the past, some other anon claimed anything is ironically fine as long as they didn't need to be pointed and redirected i.e. anime style or character to /h/; cartoon to csdg, so on. Semi realistic anime models were even there. Someone though had to get snarky, not constructive with somebody and even newfags testing out Sambase. Why not those actual non realism gens then. Another anon clashes and insists on only strictly realism posts. There you go, thread identity crisis.
>>
CRISIS?!
>>
AT THE DISCO!
>>
erm... can't you just take whole e621 dataset, prepend @621 to it and make a lora or a finetune on anima. that way it learns new tags and if it fucks up the style too much just put @621 in negs.
>>
That part.
>>
cant we just train on the entirety of civitai so we can put it in the negs?
>>
>>8875730
We've already determined that it doesn't fuck up styles when you train it on something else. First with deviantart and ye-pop, then with this "ultra real" finetune still being able to do anime artists. You don't even need to neg out the other datasets, people just do that out of paranoia.

The main issue is it's millions of images, nobody other than russell has the hardware to attempt it and he doesn't seem to care.
>>
>>8875733
it's already in. just put (score_9, ai-generated:1.8) in negs
>>
let russ cook
>>
>>8875735
>>8875730
I would not recommend putting style heavy tags in negatives. It produces a style bias in a specific direction, not away from your one thing in all directions. Which is what people seem to expect.
>>
I trust TouchDown Russel implicitly and respect his ML prowess.
>>
File: 1746867965177816.png (1.44 MB, 1216x1472)
1.44 MB PNG
>>
has anyone tried kirazuri + WAN 2.2
how is the result?
Can share workflow?
>>
>>8875734
>The main issue is it's millions of images, nobody other than russell has the hardware to attempt it and he doesn't seem to care.
based russell not poisoning his model with furshit
>>
>>8875741
the ntr route...
>>
>>8875741
something is extremely wrong with the dude's proportions/perspective here
>>
File: 1763884585068773.png (1.67 MB, 1152x1536)
1.67 MB PNG
>>
>>8875744
I just explained that the model is immune to poison in the first place
>>
>>8875749
sorry bro you're gonna have to ask jordach for your e621 farting on prey finetune
>>
>monster shit
>3dpd game slut
it's /d/ or /aco/?
>>
>>8875752
it's /h/
>>
i think it's a fair cycle that furries get shat on. every drought year up pops a furry 'project' that amounts to nothing but compute-burning garbage (chroma $200k).
>>
jordach will bake it if you fund his next fursuit
>>
i can send him a dead rat
>>
>>8875742
>kirazuri + WAN 2.2
What a fumble. This is a touchdown russell general, chudette.
>>8875763
>the pakistani
I thought he was mexican?
>>
if i see someone talking about pakistan i immediately assume it's brownie indian hands typing it. normal white people shouldn't be aware of third world shithole countries like paki
>>
good morning sirs
https://litter.catbox.moe/8wjf0l4jsl9vxlho.png
>>
>>8875768
he literally self reported though??? >>8875034
>>
>>8875768
this
>>
>>8875769
can you turn the cfg down, it's melting my eyes.
realism is like cfg 2.5, not 5
>>
>>8875769
Give us a Juri, BLACKEDLORAanon.
>>
>>8875774
with or without *that*
>>
>blackedlora running rampant
>javlora still in the oven
b-bros.. I need a good CODE to cope tonight.
>>
File: test.jpg (2.53 MB, 7527x1743)
2.53 MB JPG
>>8875590
Ok, this one actually works. Though still very awkward to use compared to any sdxl controlnet. Lowering the strength doesn't work as I would expect, rather than becoming less strict overall it just starts ignoring thinner lines altogether.

workflow: https://litter.catbox.moe/j1qrvypiz7c1hv9f.png

Another issue is because it relies on a lora, just enabling it it has serious side effects even with the controlnet set to zero strength. picrel is base, lora only, reference, and then 0.25, 0.5, 1.0 strength. Even the 0.25 seems too strict on the primary lines, which also ends up affecting the style. Maybe a different line preprocessor not canny would get slightly better results.
>>
>>8875775
With but make it a nigress as Juri is more of a dom.
>>
>>8875776
jav lora was never real, bro...
>>
why are localweiners so obsessed with controlnets
>>
>>8875779
nyooooooo..
>>8875780
>ugly bastard, gokkun, humiliation
Based.
>>
>>8875773
Those gens are less fried than the real photos
>>
>>8875783
made you reply
>>
>>8875765
>kirazuri
The what?
>>
>>8875769
Why isn't that ass being gaped by a monster BBC?
>>
>>8875789
has the good morning fag ever done penetration
>>
>>8875791
>>8869573
>>8848603
>>
>>8875628
it's the artist mix it's a bit weird
>>
>>8875749
immune is the wrong word, you still need to neg them out, but style negs work rather well in anima
>>
File: 1750684892743952.png (2.82 MB, 1216x1472)
2.82 MB PNG
>>
>>8875798
You don't, as long as you prompt something else for style.
>>
>>8875803
yes you do. what kind of retarded post is this? not having style negs in the dataset are noticeably less anime, even if you prompt an artist
>>
https://litter.catbox.moe/su56qgp6mgk0f5qj.png
now imagine if we had actual jav lora
>>
File: 1758217770797753.jpg (456 KB, 1216x1472)
456 KB JPG
>>8875800
FTFY
>>
>>8875800
>>8875805
why is it so cfg 30?
>>
>>8875807
that's not the pakistan flag
>>
>>8875805
this shit looks like sd1.5
>>
>>8875810
Meant for this >>8875800 post
>>
>>8875812
oh you got me confused
>>
>>8875812
sd 1.5 looked less fried and had better backgrounds :skull:
>>
also had better traditional media texture
>>
>>8875805
>now imagine if we had actual jav lora
Anon please.. I'm just barely holding on.
>>
>jav lora
i'm still waiting for gopro
>>
>>8875819
you got a fake gopro lora
>>
>>8875819
Just use the blackedraw lora at low strength?? It's the same thing.
>>
>>8875805
reminder people only like jav ironically
>>
you should ask for loras on /ldg/ instead. theres nothing but retarded shitposters left in this cesspit
>>
>>8875824
nah thats you
>>
>>8875826
I don't like it at all
>>
>>8875825
looool not a single one of those troonix boomers know how to train a lora. they're the same retards who spent a year coping with flux.
>>
>>8875825
This. Everyone is having fun and baking great loras on /ldg/.
>>
I was having so much fun in /ldg/ that I came over to this thread to take a break from fun lol
>>
>>8875829
You have never baked a lora, shill. /ldg/ actually has people interested in local development, training and discussion unlike this mental asylum.
>>
>>8875825
the general that doesn't know what sd-scripts and LoRA_Easy_Training_Scripts are, and keeps recommending the anima standalone trainer? yeah no thanks
>>
>>8875832
>shill
Reminds me of when we used swear words in elementary school, not knowing what any of them meant.
>>
>>8875831
same lol
>>
>>8875835
Subhuman who says "troonix" is most definitely also a saas shill and a proprietard.
>>
if hgg is so good at lora baking where's the jav lora huh?
>>
>>8875837
That may well be, but you're the one shilling /lgd/ right now.
I don't get why you'd even want these posters in your general.
>>
all i said was ask for loras there, it's not like im directing the paki to shit up that thread...
>>
>>8875839
I see people who are genuinely interested in lora training and I would like to rescue them from this dumpster.
>>
>local development
AHAHAHAHAHAHAHA
>>
You are fucking listen, I am here just squatting with all my brothers.
>>
that's kinda gay
>>
imo /ldg/ is really cozy. give it a try, it's not even trolling. you guys seem to be interested in jav/real girls anyway so it makes sense to me
>>
>>8875833
>the general that doesn't know what sd-scripts and LoRA_Easy_Training_Scripts are
there's no way this is true, right?
>>
>>8875848
I agree, that would be a win-win if those who are interested in JAV and real stuff leave for your shithole.
>>
blackedlora bros.. I think we can all read the room.
>>
it's interesting the only schizos related to ai generals i've seen go into other boards and try to astroturf are /ldg/ and /adt/. why are you retards on /g/ so mentally ill?
>>
>>8875853
Local is stagnant and this is how they cope
>>
File: 1746819372985758.png (1.03 MB, 832x1216)
1.03 MB PNG
why does Anima love dark-skinned male so much? I already use light-skinned male and put dark-skinned male in the neg (negpip)
>>
not true
>>
>>8875855
just look at the dataset dog
>>
>>8875853
Take a look at your joke of a general again
Do you think you have the right to call /g/ "mentally ill" after this?
>>
>>8875854
it's unironically less stagnant than it's been for quite some time thanks to anima existing (pony to illustrious and then noob to anima were the worst stretches). i feel like all of these retards are newfags that haven't come to terms with how local always lags behind. if they think this is bad they would have lost their mind if they saw how fast things were progressing back in the nai leak era all the way up to pony compared to now
>>8875860
>Do you think you have the right to call /g/ "mentally ill" after this?
yes
>>
>>8875647
ok this looks good
but that's innie
>>
File: 1759678150411690.png (2.62 MB, 1472x1216)
2.62 MB PNG
>>
>>8875855
Try using normal negs for skin color. Also neg interracial.
>>
>go to /ldg/
>it's just the same shitposting as here except without any discussion of lora training or hentai with the addition of avatarfags calling each other out
can't believe this faggot got me to waste 1 minute skimming through that shithole
>>
>>8875866
dunno they seem to actually post loras there and discuss the news, AI feels fresh and interesting again almost. hgg seems like a joke compared to them now
>>
>>8875867
There's lots of exciting SaaS news there by the looks of it. Lots of discussion about GPT-Image-2, NovelAI, and Nano Banana Pro
>>
>>8875741
Box it senpai, I'm gonna need to make more.
>>
>>8875867
Comfy and the Big Russ posts there afaik. Imo it's the most relevant AI general nowadays
>>
>>8875870
remember when hdg was the main hub for anime posters? how the mighty have fallen
>>
you guys should do this skit in /edg/ instead
>>
hdg, wuteva happened there?
>>
>>8875869
https://litter.catbox.moe/mxax3dscdsahzlkk.png
Also the correct version since I derped up the prompt at first https://litter.catbox.moe/buf2im2g9rc9ufdk.png
>>
>>8875863
nice.
>>
just bought an opus level subscription to novel ai where is the best place for me to post the art that im making?
>>
>>8875879
/hdg/
>>
right here, man!
>>
>>8875879
/ldg/
>>
>>8875879
here. we are in desperate need of good gens...
>>
>>8875879
Patreon, sir
>>
Should I take offense when people call me a paypig when I post NAI gens if I'm not actually paying??
>>
>>8875874
Thanks but wrong image...
>>
>>8875885
Technically speaking, we're all paying the price when we use AI as it's quickly destroying our planet.
>>
>>8875887
That's lil timmy's problem.
>>
pretty sure indians destroy the planet faster
>>
>>8875885
If you keep shouting so loud every retard will eventually jump on the train and then you won't be using it anymore.
>>
>>8875886
gdi https://litter.catbox.moe/8kv8iki1m0blniju.png
>>
>>8875887
Is it really that bad or it's just redditor bullshit?
>>
>>8875891
Nice.
>>
>>8875892
just liberal anticorpo reddit bullshit, nothing ai is doing is any worse than what server farms were already doing
>>
>>8875895
this
>>
What would a JAV lora even look like, they don't have a distinct unique look besides having shitty flat lighting.
>>
>>8875899
cool mosaics for the aesthetic
>>
>>8875895
This, these leftists just hte America and want us to lose to the CCP.
>>
>>8875895
Same thing, just a lot more of it. Projected to hit 10% of global electricity usage early next decade https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai

But >>8875892 it's a temporary issue. Normally the price of electricity and clean water would be enough to prevent over-use, as it does for metallurgy, chemistry or any other business. The problem is most AI ventures are still in the pre-enshittification phase trying to lure in more customers, they sell their service at a loss. Often fractions of a cent per query on massive LLMs.
>>
>>8875899
ugly japanese people
>>
>>8875902
god the enshittification phase AI is gonna be so fucking bad
>>
>all this jav lora discussion but no jav lora baking
doko?
>>
ask /ldg/
>>
>>8875906
anime general chudster
>>
you need to pee
>>
someone is training ip adapter for anima
https://huggingface.co/Wenaka/anima-ip-adapter-dataset
>>
>>8875914
who cares about zero-shot cope when lora is a thing?
>>
>>8875914
when are they going to train TIIIIIIIIIIIILEEEEEEEEE!!!!!!!!!!!!11
>>
THISSSS
just train the heckin lora
>>
>>8875915
>buzz farming lorajeets when their livelihood is threatened
>>
File: 1762132508389182.png (3 KB, 138x64)
3 KB PNG
>>8875915
imagine waiting for 4 hours just to goon to the new gacha slop
>>
Don't care.
Just. Train. The. LoRA.
>>
what makes you think you can tell me what to do
>>
why do vramlets always go rabid when loras are mentioned?
>>
>>8875906
Provide a dataset and I will bake. You don't even need to tag it, I'll just run JoyCaption on it.
>>
>>8875928
Not sure, considering Anima uses less VRam on average than the XL models. -vramlet
>>
>>8875933
For the same latent size, but those are 2x larger on anima.
>>
>>8875919
also not including the time spent gathering and tagging the dataset. also the rebake(s) because the first run(s) sucked. or the time spent on the artists and concept loras you now need after raping the base knowledge of the model with the other loras...
>>
>>8875914
is ipadapter STILL the best cope for consistent original characters? anima is way more creative than xl so I wind up getting slight variations in characters which makes them hard to use between prompts
>>
>>8875939
More reliable to train a weak tiny lora on your gen(s)
takes like five minutes
>>
>>8875939
if you want consistency, train a lora
>>
>>8875944
>>8875945
How about we move away from 2022 shitware instead
>>
>>8875946
Loras are a part of my identity now, don't think I can do that.
>>
>>8875946
>2022 shitware
it's a small model on top of the larger model. it's just the fundamentally best solution and will always be. apart from outright training the entire model it'll always outperform all other solutions if quality is of any concern. it's like saying that wheels are 10000 B.C. shitware
>>
>update sd-scripts
>finetune config suddely needs twice the amount of vram and I can't use adamw anymore
thanks @machina
>>
why are loratards like this
>>
>>8875949
so just revert
>>
File: 1749438917360842.png (2.3 MB, 1280x1408)
2.3 MB PNG
>>
>>8875952
Haha =) Very nice, man. I really like the humor and clever use of text capabilities of the new model.
>>
Retard. Machina's fork of sd-scripts hasn't had any new commits for a month. Are you trying to use upstream sd-scripts with lora easy training script
>>
>>8875953
it was drawn by hand...
>>
darknessfag was less cringe...
>>
>>8875954
I was still on the now deleted anima branch
>>
>>8875957
Hmm, if you update the ui, the update script should also update the backend and sd-scripts. This shouldn't happen unless you manually ran git pull to update or are still using an outdated ui.
>>
>>8875938
wow, such a cute and clean style, I wonder how the poster who came with such mix idea is like
>>
noodlefriend is a naigger???!!?!
>>
>>8875963
you're new as fuck if you didn't know this
>>
im selling all my noodlecoin
>>
Hi! My name is Takeshi Yamayata, I am overseas anime fan! I just learned about all of this so please be gentle...! I hope we can be friends!
hentai yeah!!!
>>
I really like big boob anime chicks and cute colours, what you like??
>>
>>8875966
noodlekek really fell off..
>>
I can't do this anymore, I have no idea what newfags say nowadays
>>
>even noodlekek gave up on local
its never been this grim
>>
uncs really paid $25 for this back in the day?
>>
>I can't do this anymore
if only it was true
>>
>>8875974
uncs pay $25 for that today
>>
>>8875974
It was the shit back on the day you know, a true escape from pony hell

>>8875975
I mean the act of course
>>
>>8875974
There was nothing else for hentai
>>
there is nothing wrong with pony
>>
i cant wait for nai v5 bros
>>
>>8875978
unironically if you do a better upscale this would look better than your newer gens
>>
>>8875985
it looks like shiny flat ai slop no matter how much you upscale
>>
>>8875986
how's pakistan this fine day?
>>
File: 00082-3696648765_result.png (2.54 MB, 1664x2304)
2.54 MB PNG
>>8875985
>this would look better than your newer gens
Pose is good but you clearly have no idea what you are talking about, I was getting a second hand embarrassment while browsing my old NAI gens
>>
>>8875987
>pakistan
again only browns can identify SEA or african countries by name. to whites it's just the third world
>>
What's with the uptick in NAI trolling? Is it because of what happened with AetherRoom?
>>
>>8875990
again >>8875771
>>
>>8875990
That's only because we've fucked our educational system, fellow retarded cracker.
>>
>>8875993
>uptick
???
>>
>>8875990
Why are Americans so proud of their stupidity and ignorance?
>>
Uh oh turdie meltie
>>
File: file.png (112 KB, 1583x243)
112 KB PNG
>>8875997
This thread has 60% more NAI trolling compared to the previous one.
>>
>>8875998
We're trained for it. The c-suite wants to keep us stupid, need more serfs
>>
Can we reach 3k this thread? We're almost there bros
>>
>>8876009
paki spammer is gonna fuck it up
>>
File: 63079719.jpg (167 KB, 832x1216)
167 KB JPG
>>
File: 63627333.jpg (110 KB, 832x1216)
110 KB JPG
>>
File: 63856666.jpg (123 KB, 832x1216)
123 KB JPG
>>
>>8876012
>>8876013
there he goes
>>8876011
>>
no please stop spamming we are so close to 3k
>>
why is darknessfag like this
>>
summoning groidspammer/groidy/remaketifaspammer/scatspammer/sailormarsspammer to save this thread
>>
>>8876018
I wonder what happened to Sailor Mars bro
>>
File: 1762856005725328.png (2.54 MB, 1280x1408)
2.54 MB PNG
>>
>>8875998
Why are you so much poorer if you are such a smart boy that can name every irrelevant shithole country in the turd wordl?
>>
Look at how the uneducated 'white' man seethes. He has been thinking about this for almost half an hour.
>>
>>8875998
>why can americans tell the difference between some fag from portland and a chud from alabama but can't tell the difference between person A that lives in a shithole on the opposite side of the world and person B who lives in an adjacent shithole on the opposite side of the world?!?!?!??!
it's almost as if removing the locality removes all meaningful distinction.
>>
why is the paki like this
>>
Noodlekek could only generate images back in nai3 days, nowadays he only does recolors.
>>
rent free
>>
25 dollars?
>>
>>8876028
this is why controlnet is an anti-feature
it kills creativity
>>
imagine noodlekek's power when nai v5 is out
>>
1000 a month, one room, might have bugs
>>
>>8876036
>it kills creativity
he says while genning the 867th white background bobblehead 1girl standing for the day
>>
>bobblehead out of nowhere
>>
>>8876038
You think he still has money for it? Between videogame addiction, genning (with CN), watching vtumors, does he even have time to make money?
>>
why are localweiners so obsessed with controlnets?
>>
>>8876044
it's only 25...
>>
>>8875948
It requires way too much setup and is way too finnicky to be called "the best solution". Whatever NAI uses for character reference is much better in that niche since it gives similar results at a microscopic fraction of the time and effort. It's literally NAI's only good development.
>>
>>8876048
slopper, retard, skill issue
>>
nai creature isn't taking the aetherroom announcement very well it seems
>>
>>8876042
Not sure why you brought him up, but firedotinc is indeed the most creative poster here.
>>
>>8876048
Built in couple with automatic advanced or mask zoning, probably an ipadapter in there too. If that's worth 25 bucks to you, go for it.
>>
>>8876054
this, i still remember the hole in a wall with splotch of cum gen, true virtuoso of digital mixed media
>>
File: 00140-474227639_result.png (1.44 MB, 1664x2304)
1.44 MB PNG
>>8876044
geg, you almost got everything right
almost
>>
JUST
BAKE
A
LORA
>>
>found noodlefraud's twitter
uh oh
>>
>come to lora baking thread
>mad when people like loras and talk about baking them
what causes this retardation?
>>
>>8876062
post it
>>
>>8876056
>Built in couple with automatic advanced or mask zoning
What's a variant of regional prompting has to do with character references?
>If that's worth 25 bucks to you, go for it.
I didn't say that, I don't pay for NAI anymore, but I can't ignore how often I used character reference when I did and how much I miss it on Anima.
>>
>>8876065
>miss it on Anima
L O R A
O
R
A
>>
>>8876065
>has
got*
>>
i thought the first 3k thread would be more exciting
>>
>>8876067
I miss it BECASUE it wasn't LoRA, with none of the hyperparameter settings, dataset collection, dataset curation or trial and error training that takes time.
>>
>>8876069
It isn't the first thoughbeit
>>
otoko no doko
>>
>>8876071
it is and you still dont have proof otherwise
>>
>>8876070
you miss it because you're a retarded shitter or retarded saas shill, yes we know
>>
>>8876065
Focused character references on specific bodies/regions, prevents bleed, better details.
>>8876067
This guy's not wrong, but he's a little excited.
>>
>>8876072
you know, last time I told you to post yours you didn't even after I did so that's why I haven't post cute boys in a while
>>
>>8876063
this is hentai generation general, not loratroon circlejerk thread.
>>
>>8876073
pretty sure it isn't. I swear we've had one hit either 3200 or 3500 before.
good luck finding it though and there's a good chance it doesn't exist due to the nai faget reporting old archived threads as containing CSAM and getting them completely nuked(without any form of review).
>>
You would be able to do character reference on local if we had an inpaint model.
You'd be able to do inpaint trick and since local doesn't have retarded limitations like free resolution you would be able to use it very easily to reference any character you want.
>>
>>8876078
we havent ever had a 3k thread
>>
>>8876080
Misinformation or ignorant.
>>
>>8876077
you got lost somewhere on the way to /hdg/ it seems. this thread is for lora and noodlefriend appreciation
>>
>>8876081
says the guy who has literally zero proof other than "uhhh im pretty sure! i swear! its real! but if it doesnt exist its because any evidence of it has been erased by evil people!!"

you are retarded lol
>>
>>8876078
You're confused. At some point going over 2k was considered an achievement. That's why you remember 2.4k threads as if they were 3k+ threads. Reminder that hdg used to be 800-1200 posts per thread before civitai ponytroon invasion.
>>
One thing is true, /hgg/ never had 3k.
/hdg/ had 3k+. I'm not "pretty sure", I know for a fact, I was there.
>>
>>8876085
>/hdg/ had 3k+
unless you can prove it, that literally never happened as i was also there and we never hit 3k
>>
>>8876085
>I was there
and you should go back there
>>
>>8876085
Post the 3k+ or shut the fuck up, retard. Hdg never reached 3k.
>>
>>8876087
There was no /hgg/ yet so you were also there lil' bro.
>>
now I wonder if NAI isn't sitting on a huge stack of character loras that they apply during regional/couple
>>
>>8876091
based nai 3k post
>>
>>8876091
>regional/couple
such an extreme cope thinking their custom architecture is anywhere close to being similar to regional prompting or forge couple lol
>>
>>8876091
maybe they just train one on inference with their 8 billion nvidia gpus
>>
i was here
>>
3k bros... we finally did it. im so proud
>>
Do you remember where you were when /hgg/ thread #044 made AI history?
>>
>>8876095
You were here for a nothingburger since it's not the first time:3
>>
this ones for you cathag
>>
>>8876091
they're too retarded to do that.
they could legitimately do this and apply it on the back end and then have a real easy, low effort monthly update post going "Hey we added all these characters!" to sell more subs.
>>
>>8876098
it was my first time.
>>
>>8876091
clearly that's the case just like it was for artists on v3, clearly
>>
this is my third 3k. pretty mundane by now
>>
>>8876100
thats what they do for styles. no way a model can just remember styles in tags like this.
each time you prompt for an artist, a lora is automatically applied to your prompt.
>>
lol
>>
localkeks are so funny
just because they are stuck on 10 year old tech that somehow means everyone else is lmao
>>
>>8876104
>na
>styles
lol
lmao
>>
>>8876104
lol no, if they did that then v4.5 would actually be good at artist mixing
>>
>>8876093
It uses almost the exact same syntax. Pretty sure the only 'custom archetecture' is some form of automatic image detection/region creation during the process. Which is impressive, honestly. I never hated NAI, but the mild obfuscation always kind of put me off.
>>
>>8876106
they genuinely cannot fathom not having access to the latest toy and more importantly not having it for free for once in their lives
piracy (and especially how easy it is to pirate) plays a big part in this, at first it was just "i'm poor so i'll pirate" but now it devolved into "i paid for my pc/parts so i won't pay for data (software, media, etc)"
>>
>>8876108
this, artist mixing is impossible without Lora technology. lora means low rank adaptation, neat tech isn't it?
>>
Good morning sirs, there you go! ヽ(^o^)丿
>>
>he paid for his pc parts
bruh :skull:
>>
>>8876109
>It uses almost the exact same syntax
my uncle's prius uses the same kind of batteries as a tesla therefore it's a tesla
>>
>>8876110
My mom got me my pc few years ago, who the fuck buys their own computers?
>>
And the TES/astroturfing they can go in the hole with Comfy for that one
>>
i threw up
>>
>i paid for my pc/parts so i won't pay for data (software, media, etc)
i say this
>>
the elder scrolls?
>>
Thread Engagement Specialist
>>
on average how long does lora training take on a 4090?
for a good lora not a slopped job
>>
>>8876112
Do you do rimjob gens?
>>
>>8876112
blackedraw lora edition?
>>
samefag-kun...
>>
any use case for not posting 5 images right now?
>>
>>8876128
aiming for 3.5k
>>
>>8876129
won't happen unless you nuke pakistan
>>
>>8876128
I'd need to make them look good and that takes time.
>>
>>8876131
cuck
>>
>>8876130
Wtf is your problem dude?
>>
>>8876134
he's indian
>>
>>8876130
pakistan president here. i'm giving iran our nukes
>>
>>8876122
he's engaging the wrong thread tho
>>
>>8876134
(You)
>>
Painterly bro is white unlike most of (You).
>>
lmao
>>
>>8876136
based based based
>>
4k thread when
>>
>>8876143
there were multiple 4k threads, chud
>>
if you don't remember the 11k thread you're a filthy tourist naishill
>>
that thread when we hit 100k? now that was the real shit
>>
This 3k thread doesn't even count since it took 6 whole days. Back in the day we would burn through 2k threads every day and 3k took at most 2.
>>
for me it's 67k
>>
you weren't in the first 3k? so what? get over it
>>
40k bros?
>>
deus vult
>>
the juu stole all the 3k threads :skull:
>>
File: file.png (917 KB, 1570x1471)
917 KB PNG
>200B
Yup, I think local won.
>>
>masturbating to numbers
get out fucking nerd
>>
>>8876154
finally. 200 finger gens
>>
>>8876154
Imagine how realistic those cats in spacesuits on the moon will look.
>>
>>8876154
199B of the parameters are for making sure it remains ethically censored
>>
>>8876154
HOLY SHIT!!! LOCAL JUST BROKE THE RECORD FOR THE WORST PARAMETER-PERFORMANCE RATIO!
>>
>>8876163
>implying that only the TE for gpt doesn't have at least triple that amount
>>
hidream wai will save local
>>
File: 00564.jpg (742 KB, 2176x2944)
742 KB JPG
>>
File: 00562.jpg (762 KB, 2176x2944)
762 KB JPG
>>
File: 20260312184149.jpg (379 KB, 2496x1152)
379 KB JPG
>>
File: 20260312181917.jpg (390 KB, 1152x2496)
390 KB JPG
>>
what causes this
>>
wow the new nai must have just dropped
>>
When you read "what causes this?" do you stress the word this? It seems the most logical.
I stress the word "causes" and I can't fucking read it any other way, just doesn't sound right.
>>
I emphasis the mental picture of the crying s()yjak posting it
>>
>>8876181
I wouldn't want to hold them in my mind. Visiting their website, it was an actual mental asylum. This place is like a stroll in the park now.
>>
File: 1774655820841337.png (2.42 MB, 1280x1408)
2.42 MB PNG
>>
>>8876185
she's only pretending, human cocks are too small for her
>>
make a new thread within the next 5 minutes or i'll make highlights
>>
>>8876189
oh no
>>
/2h2g/ no turdies allowed edition
>>
>>8876191
but noodlebro is honorary white (light skinned mexican)
>>
>>8876189
As a poster of gens myself, this possibility does not worry me too much.
>>
>>8876188
Which cocks are the correct size? Horses don't exist in Uma world.
>>
>>8876196
mine
>>
In all honestly I should do a new board and thread on the other place but you all will be under my rules and I would have to play the devil's advocate to determinate what stays and what not
It sounds both good and bad at the same time

>>8876196
>Horses don't exist in Uma world
You have bioengineered horse girls and technology adapted to them, how hard would it be to have horse sized cock implants
>>
>>8876199
>the other place
nobody will follow you there.you tried is 3 times already
>>
>>8876196
Then shouldn't they also have horse men? Like in Arknights.
>>
>>8876199
>bioengineered horse girls
Headcanon
>>
>>8876199
>new board
And then what, you gonna ban loli there? Or what's gonna be the difference with /hdg/?
>>
>>8876204
>you gonna ban loli there?
nobody actually wants loli anyway, its like two pedos total out of all the posters here who try and gaslight everyone
>>
>>8876205
I don't care about loli either, I'm just curious about the grand plan for new board, because there's already a board.
>>
i'll let pulchra (but no other furries) be posted
>>
>>8876207
I'm not posting on ayakon's board thanks.
>>
I'm gonna do it
I'm going to train the TE
>>
>>8876201
>nobody will follow you there.you tried is 3 times already
I haven't tried for real yet, the fact that you refuse to use the existing threads is a different history

>>8876203
Was the first uma out there in the wild, then?

>>8876204
I won't ban anything just force a spoiler on the really bad stuff (like really extreme things) and yes, while the residents there said themselves they are also ok with other types of gens, let's be real, how many of those are there lol
>>
astraGOD is back??
>>
>>8876210
So it's the same as /hdg/. I'll just keep posting in /hdg/ until they resurrect treech or something.
>>
>>8876212
then post in /hdg/ and fuck off from /hgg/ already
>>
>>8876215
No.
>>
>>8876210
>Was the first uma out there in the wild, then?
Why wouldn't that be the case? How do you bioengineer a hybrid out of an animal that doesn't exist in the first place?
>>
every 4chan alt that somehow thinks 4chan's rules are too restrictive is always an absolute shithole
>>
>>8876210
I don't post to infinite because it's dead, I don't trust that shithole and turning on a vpn just to post feels dumb, awful and is really inconvenient.
>>
>>8876219
it's /hgg/, the superior thread that all the /hdg/ browns migrated to for some reason
>>
>>8876223
reminder it's jannie singular, we know from the leak, maybe they are in a good mood as their HRT was approved by insurance
>>
>/hgg/ is totally the thread with no shitposting we split to get away from /hdg/
>/hgg/ ironically ends up being the first thread to hit 3k text posts because of how much more shitposting there is here
like poetry
>>
>>8876209
If true, with what, and on what model?
>>
>>8876221
>turning on a vpn just to post
Split tunneling.
>>
>>8876223
It's the usual cycle:
>everything is chill and nothing gets reported
>pedos overdo it and have their gens nuked
>pedos get mad and report everything below a D cup to drag the thread down with them
>meltdown ensues
>>
>>8876225
Not the first. Will never be the first. History revisionism.
>>
>>8876225
>first thread to hit 3k
won't be true no matter how often you repeat it
>>
/hgg XX Generations Ultimate/ when?
>>
>>8876226
anima
>why
because I want to see what happens
the worst that could happen is I waste 2 and a half hours of my gpu's time.
>>
Attention whores and mentally ill posters ruin everything, we could have ONE fucking AI thread but they/them ruined it with the split. ( ͠° ͟ʖ ͡°)
>>
>>8876237
fuck off bbc schizo
>>
>>8876234
>why
Didn't ask why, man. I fully encourage the use of free will. Curious to see what happens.
>2 and a half hours
Fuck, I thought that process would be shorter by now
>>
>>8876237
where bibici
>>
controlnet obsessed localweiners had a huge meltdown i see
>>
>>8876240
tbdesu I'm not expecting much other than a larger filesize. Maybe the outputs will be more schizo? who knows. the loss values seem about exactly in line with what I would expect on a unet only train. I know people were claiming that training the TE made it forget a lot of things but honestly the character memory of the model right now isn't the most stellar and various stages of unet only training fitment can also cause it to forget character details.
>Fuck, I thought that process would be shorter by now
despite being a smaller model and requiring less vram it's somehow slower than training SDXL.
I blame that we're stuck using shitty vibecoded trainer implementations.
>>
I'm here
>>
i'm not
>>
>>8876247
who?
>>
私が来た
>>
Anyone preparing highlights?
>>
highlight this
*unzips dick*
>>
>>8876246
I don't think it's really forgetting, felt more like NL prompts stopped working properly. Well I only tried one lora back on preview1
>>
4k posts when
>>
>>8876254
most of the NL is pretty gimmicky to begin with. People don't realize that inference will grasp a lot of connotations just from the tag list and how characters are normally portrayed.
And the times it does work is usually still RNG dependent and unet only lora still have a pretty good chance of breaking it, anyway. Even simple things like defining text in text bubbles or on a shirt will often only get part of the text and garble or repeat parts of the rest.
>>
This thread is pretty much identical to pre-split hdg, the only difference is highlights
>>
Am I morally guilty if I paid for NAI textgen back when it was actually novel? Is it slightly better since Turkiyee is majority Muslim?
>>
The what
>>
>>8876258
you can thank all the immigrants for that
>>
>>8876123
still curious
can be for anima or sdxl models, i just want some numbers
>>
>>8876246
Isn't TI training heavily reliant on the VAE? And does it work on the other Qwen models?
>>
>>8876260
it would be based if novelai was an israeli company, but giving money to muslims is awful
>>
we don't want to give money to companies that belong of friends of pakistan only to friends of the West like Israel and Bharat.
>>
what an odd and specific thing to say on a hentai board
>>
a pakistani posted 90% of pictures in this thread. show some respect please
>>
oy vey he's noticed...
>>
jeets and jews, sharties are obsessed
>>
can someone just delete local ffs???
>>
I can do whatever i'm white
>>
>>8876257
Sure, but you can do an XYwith the lora's TE effect as the only variable. And see that it fucks up comprehension.
>>
>>8876273
i shall speak to your government about this.
>>
>>8876272
Actually no, they can't. That's one of the main advantages.
>>
gonna gen hentai using my new Steam Controller 2
>>
New Bread
>>8876282
>>8876282
>>8876282
>>8876282
>>
lol
lmao even
>>
why is the OP /aco/
>>
>>8876287
it was made with pdxl
>>
shan't be posting in paki thread
>>
real thread doko? do I have to bake myself?
>>
>>8871530
>>
yeah do it
>>
>>8876292
>do I have to bake myself?
Yes, please climb into an oven.
>>
>>8876253
https://litter.catbox.moe/hcxl4z2ipf48spmf.png
>>
>>8876308
why are you like this
>>
>>8876312
Boredom mostly
>>
>>8876312
he is based (sometimes)
>>
>>8876312
Was the emphasis on "like" or on "this"??
>>
>>8876308
me riding down the bike trail with no brakes at 100 mph headed right between them
>>
Can't think of a gen for op... someone box me one
>>
>>8876345
>>8876308
>>
>>8876348
really? well if you insist...
>>
>>8876349
Sorry bro we are already 35 posts in in the other thread, I'm committed now. Can't just start from 0, wouldn't feel good.
>>
>>8876349
just use >>8876112
>>
>>8876357
>>8876357
>>8876357
>>
>>8876349
>Jeetsplitting
yeah bro go for it, make sure whatever gen you put for the OP is BLACKED though so everyone knows which thread to use
>>
>>8875709
>Identity crisis in asdg
I was reading it thru bro. It's a childish shill war going on. A shitting on Sambase, A enraged with an Illustrious poster who didn't use Adetailer and chasing the presumed lost tourist to MMAI, B shitting on Chroma, C shitting on Flux Klein, D lamenting to one guy asdg is nearly dead lol. Hardly anyone posting. Newfags wouldn't feel like posting if checkpoint salesmen are having shill wars in a fairly dead thread. Just folks posting pics once a day to just activate the thread from inactivity. Ranging from 3d graphics to proper realistic model still shat upon. Dunno how they would react to semi realistic Illustirous anime models though.
>>
>>
File: ComfyUI_28398_.png (1.31 MB, 1120x1520)
1.31 MB PNG
>>
When is Mugen Tensei comic out? This anthology seems to be the most unpredictable and always comes out later than the others.
Also wtf 3k replies what is the reply limit on this board?
>>
>>8877598
>https://litter.catbox.moe/hcxl4z2ipf48spmf.png
workflow :) ?



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.