[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / 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
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1794006123486353.png (3.48 MB, 1024x1536)
3.48 MB PNG
Previous: >>110254936

Still running local Edition

>UIs
ComfyUI: https://github.com/comfyanonymous/ComfyUI
SwarmUI: https://github.com/mcmonkeyprojects/SwarmUI
KritaAI: https://github.com/Acly/krita-ai-diffusion
Forge Classic/Neo: https://github.com/Haoming02/sd-webui-forge-classic
InvokeAI: https://github.com/invoke-ai/InvokeAI

>Image Models
IL: https://rentry.org/illustrious_checkpoints
Anima: https://huggingface.co/circlestone-labs/Anima
Krea: https://huggingface.co/krea/Krea-2-Turbo
NovelAI (paid): https://novelai.net
Moss: https://huggingface.co/Comfy-Org/MOSS-Image

>Video Models
H3: https://huggingface.co/Comfy-Org/MiniMax-H3
Turbo lora: included above

>How to generate anime:
image : https://rentry.org/comfyui_guide_1girl
video: https://rentry.org/wan22ldgguide (No H3 guide yet)

>Style Explorers
Anima: https://anima.mooshieblob.com/ | https://animadex.net/
Illustrious: https://tagexplorer.github.io

>Checkpoints, LoRAs, Upscalers, & Workflows
https://civitai.com
https://civitaiarchive.com
https://tensor.art
https://openmodeldb.info
https://www.seaart.ai
https://www.liblib.art/

>Resources:
Share Metadata: https://litterbox.catbox.moe
Img2Prompt: https://huggingface.co/spaces/fancyfeast/joy-caption-beta-one
Online metadata viewer SD/NovelAI: https://spell.novelai.dev
Catbox/Metadata Userscript: [Gist] README.md
Output cleanup: https://rentry.org/RemovingDiffusionGunk , https://www.mediafire.com/file/vipr23exc5htmnt (batch processing python script)
Inpainting Guide from an Anon: https://files.catbox.moe/fbzsxb.jpg >>106520607

>/adt/ logo:
https://files.catbox.moe/pakfgt.png
>>
File: ComfyUI_36108_.png (1.7 MB, 1408x1728)
1.7 MB PNG
>>110266186
mossimage got released in a time where I don't have unlimited mobile data baka
otherwise I was gonna make a human swastika or t least a 4girl square
>>
>>110266209
Download a VPN and use the online huggingface space you habe max 2 gens per IP
>>
File: kishipost 6.png (1.32 MB, 1024x1024)
1.32 MB PNG
>>
File: ComfyUI_35871_.png (2.18 MB, 1408x1728)
2.18 MB PNG
>>110266231
local or bust, son

>>110266278
no ass difference?
>>
>>110266186
Not a local vs SaaS thing, but this is not OP material. Nice idea on theory, then the curls, clouds and dress details repeat everywhere and it ends up as GPT slop. Nothing against the author but this just isn't his best gen. Pick a better gen from the same anon if you want to alternate betweem SaaS and local OPs.
>>
who asked
>>
File: correction.jpg (1.09 MB, 2176x1472)
1.09 MB JPG
Get it right.
>>
File: ComfyUI_27004_.png (2.17 MB, 1728x1536)
2.17 MB PNG
thread theme, kiss, and for those who gen 1girl, tongue out

>>110266300
bring back image collages
>>
File: ComfyUI_36159_.png (1.3 MB, 1280x1392)
1.3 MB PNG
>>110266408
paying theme tax to myself
1boy
>>
File: kishipost 7.png (1.32 MB, 832x1216)
1.32 MB PNG
>>110266292
>>
>>
>>110266278
Nice abs on 10/10. Also, see >>110266481.
>>
>>110266292
Nippaaa, can I have your workflow miii
>>
File: kissu.jpg (735 KB, 2176x1472)
735 KB JPG
>>110266408
>>
File: kishipost 8.png (796 KB, 832x1216)
796 KB PNG
>>110266481
>Proud of what?

Her magnificent chest of course
>>
sfw?
>>
File: ComfyUI_07493_.png (2.69 MB, 1440x1920)
2.69 MB PNG
>>110266661
its the default anima workflow idea, just prompt better and do not create a tag soup

>>110266693
rape-y dyke
>>
>>110266727
She's just recharging her magical energy.
>>
File: ComfyUI_36168_.png (1.37 MB, 1440x1010)
1.37 MB PNG
>>110266749
I charge mah laser the same way
>>
>>
>>110266703
Well, as long as she's happy with it, I guess ...
>>
>>110266693
could you lower the quasarcake please
>>
>>110266794
It's already lower than it was. But maybe, sometimes.
>>
File: ComfyUI_36064_.png (1.98 MB, 1408x1728)
1.98 MB PNG
denji after hearing how much you'll pay for a night of playing risk the board game
>>
>>110266693
10/10 pls post workflow so I can goon
>>
>>110266825
Thanks, here's a box:
https://n.uguu.se/tdbNUxps.png
It's NAI though...
>>
>>
>>110266760
I give up
Its demoralizig seeing NAI gens as a poorfag

Its over, fuck this life I'm gwtting a jov
>>
>>110266991
I am a different person and I agree, we localfags are such cucks
>>
>>110266991
Don't give up anon! Just train a lora :^)
>>
NAIfags are barking extra loud after getting BTFO by Comfyanon
>>
File: kishipost 9.png (706 KB, 832x1216)
706 KB PNG
>>110266991
>as a poorfag

You can't be that poor anon,it's like 10 for the lowest tier and 25 for the highest
>>
>>110267033
How?
>>
File: 1777187481613396.png (3.57 MB, 1792x2304)
3.57 MB PNG
>>
File: 00204-6838225.png (1.16 MB, 832x1216)
1.16 MB PNG
>>110266991
Don't be like this anon, found this Moss anime finetune with 23k images and it looks promising.
https://civitai.red/models/2530372/slivanimix-moss-imageanima
>>
>>110267043
NTA, but for image gen only the highest tier is worthwhile.
>>
>>110267048
By typing a lot of seething cope and attacks then hitting the Post button
>>
>>110266965
V6 when?
>>
File: kishipost 10.png (932 KB, 832x1216)
932 KB PNG
>>110267117
I know, just saying you can't be THAT poor
>>
>>110267124
But why Comfyanon?
>>
>>110267117
Post gens.
>>
>be local finetuner
>cry for years that sdxl is a dead end and we need a new base model
>new base drops, good VAE, good TE, promising
>immediately feed it the exact same ultra slopped SDXL dataset
Can locals explain this mindset to me?
>>
Is it MOSS issue or Civitai issue there is barely any loras? I remember Anima having much more when it came out.
>>
>>110267143
He just happens to be the most recent in a long run of local genners that make better stuff than NAI.
>>
>>110267159
Moss seems to be getting loras more or less as fast. What method are you using to measure this?
>>
File: ComfyUI_36186_.png (1.64 MB, 1344x1728)
1.64 MB PNG
>>110267164
who is comfyanon?
>>
>>110267164
That's a lot of words yet not a single example
Curious....
Is this ''comfyanon'' in the room with us?
>>
>>110267164
Interesting, is there anywhere to see Comfyanon's gens? If he's backing Kohaku or Noob DiT, I can sleep easier knowing he gives a shit about anime.
>>
>>110267178
>>110267187
NAI lost in a straight-up match:
>>110262164
>>110263282
>>
>>110267159
>>110267167
this is completely normal tho moss base right now isn't even half as trained on anime or artist styles as anima preview 1 there is no usecase for training loras on moss yet
>>
>>110267167
Just a feeling. I remember there being lots of style and character loras in first few weeks of Anima.
Can't say same about MOSS.
>>
>>110267206
I have Nai V20 running in my 1GB of DDR2 RAM, not going to show it tho, just trust me bro kek.
>>
File: fweh.jpg (691 KB, 1472x2176)
691 KB JPG
>>110267142
I just don't want anyone to try subbing at a low tier only to learn that they can make like 20 images with the sub, because those tiers are for the shitty text gen only. It's a trap to mention them in the context of image gen.

>>110267146
Okay.
>>
>>110267206
But that gen is from cui_ not Comfy tho
>>
>>110267206
That looks like chatgpt nonsense both look bad and NAI worse bc the prompt is probably trash.
>>
Localkeks aren't sending their best huh?
>>
>>110267206
How do you know that's comfyanon?
>>
>>110267043
25 bucks a month is a lot for a subscription though. Im used to subs being like 5 or 8 maybe at worst 10
>>
>>110267222
Children grow up so fast...
>>
>>110267231
>That looks like chatgpt nonsense
No it doesn't. Why the obvious cope?
>>
>>110267251
It obviously does. It's highly saturated and the curvy patterns have leaked everywhere from the clothes to clouds and even into the torii.
The NAI gen sucks and its probably badly prompted but at least it doesn't have those kind of artifacts.

Both are still shit dog but sure local wins that one.
>>
>>110264761

why are you so obsessed with me lil bro? Post some gens
>>
>>110267289
It's not oversaturated and I've never heard of GPT putting swirls on things the way a lora might do.
>>
File: kishipost 11.png (700 KB, 832x1216)
700 KB PNG
>>110267241
Guess I don't view it as much since it's the only sub I have for anything
>>
>>110267294
Nta, but why are you using illu and Anima separate like that. What do you use for inpainting?? I liked your gen
>>
>still running local edition
>most of the gens are NAI
lolcals pulling a Lost Cause, cute
>>
File: ComfyUI_temp_tgxnz_00061_.png (1.62 MB, 1280x1024)
1.62 MB PNG
>>110267312
>>110267294

I liked it too
>>
>>110266794
???

>>110267043
Shizune has no right being that sexy as a Plain Girl(tm).
>>
>>110267155
Most localkeks pay for Netflix, Spotify, and even Claude Code to vibecode their custom nodes.
>>
still waiting on moss-anime and h3 image
>>
>>110267289
>It obviously does. It's highly saturated and the curvy patterns have leaked everywhere from the clothes to clouds and even into the torii
NTA but imho that is what looks unintentionally great about >>110262164. Definitely more visually interesting than the boring "correct" ones anyhow.
>>
>>110267368
>>
File: ComfyUI_36202_.png (1.42 MB, 1472x1408)
1.42 MB PNG
>>110267387
baka just gen a skeleton
>>
File: popsicle.jpg (984 KB, 2176x1472)
984 KB JPG
I really wish we could just post gens without the console warring.
>>
File: ComfyUI_36107_.png (1.48 MB, 1408x1728)
1.48 MB PNG
>>110267460
sudden urge to force feed those skeletons
>>
File: ComfyUI_01764_.png (2.48 MB, 2048x768)
2.48 MB PNG
>>110267312
>>110267340

Thank you, guys. So far I've always done only Illu for everything. But I read someone on twitter saying they used Anima for backgrounds because Ill makes random people in them for no reason (which is true). So I decided to give it a shot and it gave me the samples in this picture. Then I tried to inpaint the girl sitting on the chair but couldn't quite get a decent result, so I started using control net and messing with the denoise level, when I set it to 0.92 the gen I posted came out and I somewhat liked it, Still not the ideal background though, but the style looks a bit softer for some reason.

Funny note: Yesterday they tried to hack my Instagram, something that has never happened to me before. I guess some anons REALLY hate me. Nice try though
>>
>>110267460
Well anon, some of us have more fun with the console wars and the yapping.
>>
File: kishipost 12.png (1004 KB, 832x1216)
1004 KB PNG
>>110267352
When everything looks exotic a plain girl becomes the unique one
>>
>>110267460

This one is extremely close to the sun, anon.

Here's some overcooked whatever the fuck this is. Maybe they'll ignore you and you may fly under the radar
>>
File: ComfyUI_36209_.png (1.9 MB, 1280x1728)
1.9 MB PNG
>>110267568
can you do characters from other IPs in that style? mitsuru kirijou from persona in that bikini perhaps?
>>
>>110267552
Motherless behavior. Nothing but disdain for cowards.
>>
>>110267589
>>
Anyone has experience using 3D software to set up poses and scene and then render AI over that? What software should I use? Any ready to make poses? Preferably sexy ones. And couples having sex.
>>
>>110267306
":3" fits Anzu very well.

>>110267460
Just ignore it. Or take part ironically and have fun with it.
>>
>>110267622
NTA but Kobeni from CSM in that style in a bikini, all shy 'n shiet?
>>
>>110267589
Another one
>>
>>110267652
>>
>>110267460
post with a name so we can filter you
>>
>>110267667
Thanks, Kishinon, that rocks! She's so frickin' cuuuuute!! Makes me wanna hug her till she stops shivering (and then starts shivering again 'cause she feels my raging boner press against her tummy).
>>
File: snow.jpg (863 KB, 2144x1467)
863 KB JPG
>>110267580
Obviously, it's a summer gen. Here's something to cool down you.

>>110267641
>Or take part ironically and have fun with it.
I could do that in the /h/ threads already.
>>
>>110267673
>we
Who made you king, buddy?
>>
>>110267673
>we
lol
>>
>>110267688

ToT Illya-chan!
>>
File: ComfyUI_36212_.png (1.57 MB, 1280x1728)
1.57 MB PNG
>>110267653
https://www.youtube.com/watch?v=jEI3N9kIyP4
>>
>>110267673
>>110267688
NTA but genuinely consider doing this. Not so that this twerp can filter you (though that's a nice benefit, him no longer replying to you) but so we can search your posts in the archive if need be.
>>
File: ComfyUI_36213_.png (1.59 MB, 1280x1728)
1.59 MB PNG
>>110267673
who is we, sucka?
>>
>>110267701
>>
>>110267622
Damn this surprisingly works
>>
>>
File: ComfyUI_31303_.png (1.41 MB, 1344x1414)
1.41 MB PNG
>>110267714
impressive slots desu, it does copy the og's style quite well
>>
>>110267043
>>110267306
Kurenai pls
>>
>>110267775
One's gonna fall on her as the tangerine tower tumbles and rudely wake her up, calling it now.
>>
>>110267796

nah i believe in her
>>
>>110267688
>I could do that in the /h/ threads already
what about posting a nipple or two in /e/
>>
File: makim.jpg (204 KB, 1248x1824)
204 KB JPG
>>
>>110267081
boobaslop
>>
File: ComfyUI_36058_.png (1.62 MB, 1408x1728)
1.62 MB PNG
>>110267831
don't be a pussy now, son
>>
  File "C:\StableDiffusion\ComfyUI_windows_portable\ComfyUI\main.py", line 268, in <module>
import execution
File "C:\StableDiffusion\ComfyUI_windows_portable\ComfyUI\execution.py", line 17, in <module>
import comfy.model_management
File "C:\StableDiffusion\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 364, in <module>
total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
~~~~~~~~~~~~~~~~^^
File "C:\StableDiffusion\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 213, in get_torch_device
return torch.device(torch.cuda.current_device())
~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "C:\StableDiffusion\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py", line 1357, in current_device
_lazy_init()
~~~~~~~~~~^^
File "C:\StableDiffusion\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py", line 609, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

HELP!
>>
>>110267851
>windows
ngmi
>>
>my hobby is troubleshooting my hobby
Enjoy the saturday, lolcal
>>
File: ComfyUI_36203_.png (1.19 MB, 1472x1408)
1.19 MB PNG
>>110267851
open a new profile and pick cuda properly anon, your torch is for cpu I believe
or if you're in a non-cuda hardware do not use a node that requires cuda like rtx nodes
>>
>>110267842
>... quoth the coward.
>>
Rtx 3060 masterrace here. What is a profile? Is that a Comfy thing? I'm downloading a fresh Comfy portable btw, I don't want to think.
>>
>>110267898
>>110267868
>>
File: ComfyUI_36223_.png (1.97 MB, 1280x1728)
1.97 MB PNG
>>110267873
I posted a huge astolfo cock, what did you post?

>>110267898
download desktop, baka idiot garbage, and go thru the setup process and pick the option for your hardware
>>
>>110267865
>never seen anyone that enjoys working on cars or tinkering in general
>>
>debugs the hobby in his free time
>then debugs the output in his free time
>>
File: purple.jpg (409 KB, 896x1152)
409 KB JPG
does everyone here like vtubers??
>>
>>110267914
A clothed loli playing in a pond/puddle. But I'm not the one criticizing other's risk tolerance.
>>
File: ComfyUI_36230_.png (1.67 MB, 1280x1728)
1.67 MB PNG
>>110267971
acolyte you silly nigger
>>
>>110267959
Vwhat?
>>
>>110268086
I'm right tho'.
>>
>>110267959
A 3DPG puppeteering a 2D avatar and doing the voice is cringe and tacky. I'll admit the character designs are interesting though, mostly the outfits.
>>
File: cui__22155_.png (2.1 MB, 1024x1536)
2.1 MB PNG
>>110267228
Yes it is my gen and some of my 'living fossil' workflows still save as cui_. And no I am not Comfy. I can barely even vibecode without wanting to shoot myself
>>
>>110268174
You should still use the cui_ filename, is funny
>>
File: Slop.png (2.03 MB, 832x1216)
2.03 MB PNG
>>
>>110268427
>wolf
>wearing clothes
???
>>
>>110268427

hey dont be mean to my wife, bad economic luck shall come your way
>>
>>110268427
Are those green fences in wildlife parks a Euro thing? They're the default here.
>>
File: 1782209516831316.jpg (855 KB, 1792x2304)
855 KB JPG
>>110267959
fuck no but anya and mint fantome are cute
>>
>>110268427
aw0o
>>
File: BP - Kobeni 01.1.png (3.24 MB, 1536x1024)
3.24 MB PNG
>>110267667
>>
>>110268476
The retvrn of Rakvn to SDXL?
>>
>>110267959
No. E-thottery taken to a whole new level (and made available to IRL trannies, jeets and who knows what else).
>>
>>110268635
Successful trannytubers are very rare, because voice changers still suck and most are unable to train their voices enough.
>>
File: naralowres.jpg (3.6 MB, 2048x3072)
3.6 MB JPG
>>
>>110268701
Oh let the smallbrains be frightened, it's entertaining.
>>
>>110267959
Vtubers = product is the person behind it= trash

Gacha = written characters inside a story = not SO trash
>>
>>110268706
Where sunset >:^(
>>
>>110268750

I'm trying to quit the sunslop for a bit, don't want people berating me for that, as much as I like the sun
>>
>>110268701
>>110268724
Toupee fallacy in action.
>>
>>110268757
>listening to the plebes' kvetching
ngmi
>>
>>110268769

don't worry ill make sure to add a supernova on the next one to make up for it
>>
>>110268762
I've done a lot of research on voice changers, they do still suck.
>>
>>110268793
sus
>>
>>110268706
Good, you are finally studying closeups, smart move.
>>
File: moss-image_04040_.png (1.61 MB, 1024x1024)
1.61 MB PNG
>>
>>110268808
They can't do blowing air in the mic, they can't really do coughs, laughing and singing also mostly don't work, even with the newest AI ones. And that's on top of speech usually sounding unconvincing or artifacted, but that can require a somewhat trained ear to detect. Real female voices also have much better dynamic range than manipulated male ones.
>>
File: doreja3lowres.jpg (1.01 MB, 2048x3072)
1.01 MB JPG
>>110268840

nigga I made this a month ago fuck are you talking about, I don't like closeups too much
>>
>>110268852
Purdy.

>>110268862
The joke here was you "[doing] a lot of research" on them. Also, good for you but I doubt the average Vt*id simp would be able to tell.
>>
>>110268894
There might be stealth ones among indies without many viewers, but for a popular one, enough people would notice and it would be talked about.
>>
>>110268852
This is the NAI killer? *lols in anlas*
>>
>>110268934
*kills you in python*
>>
>>110268982
*kills you with a python*
>>
>>110267368
>h3 image
it's not happening lil bro

chinese culture btw
>>
File: file.png (2.28 MB, 1120x1440)
2.28 MB PNG
>>
>>110269053
b... but moss is c-chinese, that means..
>>
>>110269067
:D
>>
File: ComfyUI_35892_.png (1.54 MB, 1408x1728)
1.54 MB PNG
>>110269067
she ate the entire bakery
>>
File: anima_aestheticV11_00748_.png (2.74 MB, 1512x1944)
2.74 MB PNG
>>
File: ComfyUI_30293_.png (3.36 MB, 2064x1696)
3.36 MB PNG
>>110269080
willy wonka "you get nothing" speech
>>
>>110269106
Cute
Bulbasour vibes
>>
>>110267959
I hate western vtubers and everything that has to do with a western community around vtubers.
Japanese vtubers far away from western EOP viewers are fine.
>>
>>110269159
neat, another anima preview 3 chad, i didn't like it when tdrussell started messing with the cosmos backbone though
>>
>>110269183
I've yet to try preview 2. Someone said it had better prompt adherence than 3, but I also saw someone say 1.0 had better prompt adherence than 1.1 which after much testing was complete bullshit.
>>
File: file.png (1.21 MB, 1120x1440)
1.21 MB PNG
>>110269097
The cake is cakin'
>>
File: migu.jpg (315 KB, 896x1216)
315 KB JPG
>>
File: ComfyUI_36327_.png (729 KB, 860x1109)
729 KB PNG
>>110269349
tel aviv impressed
>>
File: train.jpg (466 KB, 2176x1472)
466 KB JPG
>>110267914
Nice view.
>>
>>110266424
Your gens give me flashbacks to that manga about two childhood friends, one moves away, comes back years later at the same high school as a trans girl, lands in his old friend's class and immediately tries to seduce him. Anyway, your style got the same libidinous eyes.
>>
File: anima_aestheticV11_00757_.png (2.98 MB, 1512x1944)
2.98 MB PNG
>>
>>110269460
miyu would never do that
>>
File: teta.jpg (329 KB, 896x1216)
329 KB JPG
>>110269359
>>
>>110269469
Omg a Jigglytroon!
>>
>>110269518
it's a shame that the cotton candy color scheme has been corrupted for you
>>
>>110269471
She clearly is doing it thoughbeit? It's photographic proof.
>>
File: decency.png (1.85 MB, 2176x1472)
1.85 MB PNG
>>110269460
I agree with >>110269471, Miyu is a good girl and would only flash white panties to passengers.
>>
>>110269467
>to that manga
NTA but name?
>>
>>110269557
i-it's AI! I can tell by the pixels, or at the very least modified since it's faded around the character and phone
>>
>>110269359
migu stands for migical gurl

>>110269599
welcome back alice, NTA
>>
File: sdfg5545.jpg (949 KB, 2560x2560)
949 KB JPG
>>
>>110267492
>So far I've always done only Illu for everything.
I'm curious why you're still primary (or at all) using illus over just Anima. Is it the lack of tooling/ControlNet for the latter or what?.
If so then please share some nice illu workflow for comfy onegai.
I was using InvokeAI at some point
>>
>>110267959
I deeply look down on vtubers and specially and most of all the people that actively consume them or what their streams. The only vtuber that I may get a pass from me is that chibi one from the crazy eyes that insults in Spanish that I see occasionally clipped on Xitter.
Other than that they're all whores, and that's fine. if I was a girl I would be a massive, absolute whore, total and complete bunnyblue shit. I just hate the people that enable them more than anything. Why can't I live an easy life like that? Being a man sucks
>>
>>110269349
The cake was a lie!
>>
>>110269524
NTA but true. However, imaging that to be some faggy femboy instead of a loli gyaru ain't half-bad either.
>>
>>110269930
no
>>
File: prequel.jpg (635 KB, 2176x1472)
635 KB JPG
>>110269460
>>
>>110269882

I'm not entirely closed to the idea of moving to Anima, but I would have to play with it for a bit to get the same vibes I'm getting with ILL, so it's a matter of convenience mostly. I can definetily see the potential of Anima, but there are thing that still don't convince me. For my latest picture I decided to try making a quick gen in Anima then run ILL on top to get it up to my style. As you can see the result is nowhere close what I usually do, but then again it's the pure model, no overbaking with 20 loras on top.

Like I know a Michelin star restaurant is good, but I prefer eating burgers in the restaurant that hasn't cleaned the grill in 20 years because that's where the flavour is.

The workflow I use you can find it in CivitAI by the name ComfyUI Image Workflows" by Legendaer. He also has workflows for Anima and they are pretty good imo

From now on I will most likely use Anima for some things and then turn to ILL to get the details done, depending on how good the Anima results start coming I may switch to it.
>>
>>110269881
slop central just called
they are offering you a job at civitai
>>
>>110269951
Yes.
>>
File: cold1.jpg (416 KB, 768x1216)
416 KB JPG
>>
File: big frenda.png (841 KB, 1024x1536)
841 KB PNG
>>
>>110270153
where mugino
>>
File: ComfyUI_08479_.png (1.18 MB, 1024x1024)
1.18 MB PNG
>>
File: mugino nilitsu design.jpg (131 KB, 414x1200)
131 KB JPG
>>110270275
trying this one but they're being extremely retarded about it
>>
File: keilowres.jpg (3.61 MB, 2048x3072)
3.61 MB JPG
yeah i'm liking the Anima>Illustrious idea so far
>>
>download nochekaiser outfit lora
>expect the outfit
>get the outfit and some random body proportions that are off model
Is this a tagging issue? Like, the body features weren't tagged during training, so they got absorbed into the outfit concept? Or is it overcooked? I'm asking more for the lora making side so I don't repeat the errors myself.
>>
>>110270648
nice bambi
>>
>>110270798
no proper tagging
no diverse enough dataset
>>
>>110270817
Thank sirs
>>
>>110270676
You should consider Krea 2 if your style stays the same. It's better at style transfer, and you'll spend way less time checking gens for inconsistencies. Cui_ did it.
>>
>>110270848

When you say "if your style remains the same" do you mean I should use it in conjuction with other models or should I just move entirely to it?
>>
>>110270798
Always thought it was a thing, even with Nai
>Shimakaze cosplay comes with a breast size debuff
>>
>>110270883
I trained a LoRA of my old Illu style on both Krea and Anima, and Krea 2 captured it far better.
>>
>>110270973
Interesting... I thought about it but I didn't think I could create a lora with my own gens. I thought the noise would fuck it up. Guess I'll have to learn to train a lora then
>>
>>110270959
With NAI 5, putting the character of the person you are cosplaying and the franchise in the UC works very well for isolating only the costume, though.
>>
>>110270983
>and the franchise in the UC
Ha, never thought of that, thanks!
>>
>>110270981
Regarding the diffusion noise, I don't think Qwen's VAE captures it much or is influenced by it. You can always curate the dataset. In my case, I removed the backgrounds with Klein, filled some with solid colors, and tagged them. The good thing is that Krea is so smart that if you have sloppy things from Illu, Krea forgives them and renders them well even with the lora active.
>>
>>110271003
Franchise in UC worked for me if the character is from a different franchise than the one you are using.
It also has a better effect if you put them in the positive prompt with a weight like -0.5~-1::wuthering waves::, which is more powerful (and destructive) at isolating them. For example, Wuthering Waves has a bias toward making characters slender with long necks and small eyes, with -0.5~-1 I can mitigate that.
>>
>>110270275
delivery, sorry for the basic multiple_views white_background
(resubbed to nai after 2 months so I can stop posting unfinished corposlop)
>>
File: mara2lowres.jpg (2.79 MB, 1638x2458)
2.79 MB JPG
Decided to go back to the drawing I posted yesterday. My girl came out too cute to just leave her hanging like that with such a plain background, so here's my latest attempt with some Anima inpainting
>>
>>110271226

>ask for a favela view
>street lights
>>
>>
>>110271112
>resubbed to nai after 2 months
B A S E D. Welcome back, anon!
>>
>>110271335
I regret doing it, instead of being distracted by making dumb pics I could be continuing the neverending videogame backlog.
>>
How do I trick Anima into generating copious amounts of explicit porn? Is it even trained on it, has anyone managed to extract at least a nipple out of it?
>>
>>110271493
what do you mean "trick"?
https://litter.catbox.moe/axzdyu.png
>>
File: chubby_gyaru_polish_17s2.jpg (950 KB, 1632x2448)
950 KB JPG
>>
>>110272353
>>
>>110272506
Now make her fat, as she should be.
>>
i wish girls were real
>>
>>110272644
Witch Azu is the real Azu and she hates fatties!
>>
>>110272676
Excuses!
I would make picture myself, but i'm not running illustrious.
>>
File: sidetie.jpg (1 MB, 1472x2176)
1 MB JPG
Actual canon undies.
>>
File: ComfyUI_36602_.png (3.02 MB, 1920x1920)
3.02 MB PNG
what's wrong with your left nipple, bro?
>>
>>110272783
i'll never understand how a man can become gay, truly.
>>
File: ComfyUI_36646_.png (3.43 MB, 1920x1920)
3.43 MB PNG
>>110272921
same, why gay when both genders have cuties?
>>
File: ComfyUI_36410_.png (2.57 MB, 1728x1920)
2.57 MB PNG
>>110272953
>>
>>110272953
i count "bi" under the same category as gay.
nothing personal but man those pictures are nasty to look at
>>
File: ComfyUI_36560_.png (3.71 MB, 1920x1920)
3.71 MB PNG
>>110272988
yes, I know that men have negative reactions to pictures of two men kissing when shown by some researcher, look at the female stuff I genned tho
>>
>>110272783
It was okay before I saw the cock bulge. Can you remove that?
>>
File: ComfyUI_36011_.png (1.73 MB, 1408x1728)
1.73 MB PNG
>>110273003
a trap with a bulge removed is gayer than a trap, don't you know? straight up troon bottom surgery fantasy there
anyways have a genderbent male, you freak
>>
>>110273022
My eyes are bleeding
>>
>>110273001
yes, straight men find the image of two men kissing to be absolutely abhorrent.
i want to be clear that this is not hate-posting, i really don't care what you're into. i'm into other stuff that would disgust most people as well, it's just that there's other boards for that kind of stuff. i don't post my fetishes here because it doesn't belong here.
>>
File: Krea2_turbo_01249_.png (3.52 MB, 1536x2048)
3.52 MB PNG
>>110273062
furry pony loli shota goes to b or trash anyways, no rule against homos

>>110273031
have some krea2
>>
>>110266186
How did you guys learn to prompt well? I just suck, whether it's tags or natural language I suck.
For example, I stole this from somewhere else, but I would have never come up with something as descriptive;
>A simple, chibi style, anime character drawing with thick clean outlines and no complex gradients or heavy textures. The image is minimalist and uses a limited, color palette against a jungle lagoon background. The lighting is plain and directional-only, with strong shadows.
>>
>>110273096
gotta picture the image you want in your mind then just describe it in autistic detail. like if what you get differs from what you're imagining, HOW does it differ? descibe how it differs. repeat until it doesn't differ anymore
>>
>>
File: ComfyUI_36679_.png (2.96 MB, 1728x1920)
2.96 MB PNG
>>110273096
seems a bit sloppy, just call it simplistic anime drawing of a chibi deformed character, chibi and the deformed tag is quite powerful
>>
File: ComfyUI_36635_.png (2.58 MB, 1728x1920)
2.58 MB PNG
>>110273124
playing with best bro's balls
>>
>>110273132
Uncensored catbox?
>>
>>110273095
actually furry, loli and shota don't lead to bans here as long as they aren't blatantly pornographic. ponyshit i haven't seen posted in some time so that might be just a absolute no-no either way.
you are right that it's not technically against the rules to post non-pornographic homosexual stuff here, though i doubt many appreciate it.
>>
File: malfunction.jpg (831 KB, 2176x1472)
831 KB JPG
>>110273096
>>A simple, chibi style, anime character drawing with thick clean outlines and no complex gradients or heavy textures. The image is minimalist and uses a limited, color palette against a jungle lagoon background. The lighting is plain and directional-only, with strong shadows.
I could never prompt this way, what a pain.
>>
>>110273186
How do you prompt?

>>110273124
>Just call it simplistic anime drawing of a chibi deformed character
Doesn't produce the same style that I like

Comparsion;
https://files.catbox.moe/t5z72b.png
https://files.catbox.moe/3fgijo.png
vs
https://files.catbox.moe/6t3dl5.png
>>
>>110273141
fag
>>
>>110273152
Can you shut the fuck up please? retarded pol schizoid, let him post what he wants
>>
>>110273186
That's what llm's are for.
>>
>>110273223
Mostly tags and fragments of natural language sentences where tags can't do it, for example like in this image's metadata.
https://d.uguu.se/nBqtnime.png

>>110273233
Perhaps, I don't like it when prompts are supposed to look like that though.
>>
>>110273232
i don't visit /pol/ at all. i rememeber when it was made at a containment board and i've never visited it.
>>
any news on danbooru? i know it's dead but does any anon in their boorucord know when it's coming back?
>>
>>110272775
Icarus posting again, huh?
>>
File: ComfyUI_36697_.png (3.55 MB, 1920x1920)
3.55 MB PNG
>>110273141
anima just gens it with a fendi logo, there's no uncensored version

>>110273152
>aren't blatantly pornographic
obviously, I post mother and sho occasionally
>majority rule
bit of a silly thing to say to a "minority" e.g a faggot
>>
>>110273396
There are much more dangerous things being posted in this thread and those are apparently fine too.
>>
File: ComfyUI_36712_.png (2.5 MB, 1728x1920)
2.5 MB PNG
note to self, do not delete the pooool table tags
>>
File: ComfyUI_36724_.png (2.8 MB, 1728x1920)
2.8 MB PNG
>>110273520
>don't be shocked by the width of his hips
>check out my new prompt, the prompt of this gen
>don't be shocked, by the dong of my trap
>check out my new prompt, the prompt of gem

>listen to the sound of my fans
>you can check it on catbox, its the prompt of choice, yeah
>don't be shocked by the size of my prompt
>its the new prompt, the prompt of choice, yeah
>>
File: pomf.jpg (333 KB, 832x1216)
333 KB JPG
>>
File: sellout.jpg (2.3 MB, 4023x1671)
2.3 MB JPG
Goodbye image gen, at least until I get my M5 Ultra. Someone else is now enjoying my rig.
>>
File: ComfyUI_36561_.png (3.66 MB, 1920x1920)
3.66 MB PNG
>>110273800
shame, why did you move from x86? or just needed something portable?
>>
>>110273845
if someones moving to a macbook its most likely for local LLMs since those work well on them
>>
>>110273800
48gbvram gpu anon? Based! Enjoy your new life!
>>
File: ComfyUI_36700_.png (2.76 MB, 1728x1920)
2.76 MB PNG
>>110273886
makes sense, saw people running kimi k3 whatever on a 512gb storage mac mini, the model could not fit properly
>>
>>110273845
>>110273886
Exactly. Also, comfyui stuff now works on M5. I wanted to get to the 256GB RAM tier so I can have a big LLM like qwen-flash and still have room for a big image model too, and keep it on the same machine.
>>
/adt/ (sage) attention!:
Interesting: https://civitai.red/models/3042006/krea-2-anime-tuned
Because this: An anime full fine-tune of Krea 2 (all 12.8B DiT parameters) on ~196k curated anime illustrations.
THIS ->196k curated anime illustrations. <- SIHT
>>
>>110270848
>Cui_ did it.
Am I supposed to know who that is
>>
>>110274235
>Prompting: capgen
Hmmm.. Might try it. Not sure if I should get base or turbo.
>>
>>110274235
>>110274352
>Only general and sensitive ratings
>The training data is almost all single-girl illustrations. Boys, multi-character scenes and non-anime styles are weak.
Nevermind.
>>
>>110274235
>fp8
So far I've only used int8_convrot checkpoints for Krea. Would this run the same?
>>
>>110274314
>>110268174
how to spot him:
2:3 vertical ratio
mostrly mature females
retroish aesthetic
>>
>>110274235
>your image
imagine training on 200k anime images and still not being able to break out of the krea turboslop look
grim
>>
>>110274452
>krea turboslop look
Can you give examples of this?
>>
Thanks for your insterest ´nons
>>110274452
>capgen turns a few Danbooru tags into a full caption in this style. It needs no LLM, runs on CPU, and takes milliseconds per caption. Ways to use it:
Yes, it is a shame. People spend a lot of money training these models, then cut corners on the small things that make the model worse to use.
>>110274409
I am not completely sure, but ""I think"" int8 convrot is faster.
>>110274359
Yes, the old SDXL style of composition still shows up in a lot of models people train, we never outgrew 1girl
>>110274352
> 200k anime images and still not being able to break out of the krea turboslop look grim
You are right, I noticed it in almost all the images too. Still, I thought it was worth sharing this model as a general anime model on /g/, because of how it ranks among the new-gen anime finetunes by dataset size:
This model is in first place in the Krea 2 line with 196000
Then MOSSRAEL with 68,000 images.
Then SlivaniMix MOSS with 23,000.
And then MOSSAD with 3,685.
>>
File: kishipost 13.png (1.23 MB, 832x1216)
1.23 MB PNG
>>
File: ComfyUI_36376_.png (2.37 MB, 1536x1920)
2.37 MB PNG
>>110274532
its tsunade sunday tho
>>
>>110272975
Cute 'beni, is this new?
>>
>>110274521
Krea gets the shapes right but colors and shading in anime stuff feel a little off. I can’t say what it is, but you can tell.
>>
>>110274705
>>110274521
https://civitai.red/user/uncle_krupi/images
>>
>>110274532
Nice one, Kishinon.
>>
>>110274705
>you can tell
Tell what? You haven't said anything solid.

>>110274716
There's no one style here at all. These are also just quite good.
>>
>>110274716
Can you point out a particular image?
>>
>>110274788
>>110274839
These are not intelligent people you're responding to
>>
I think it was fun while it lasted guys but I'm unironically dropping this "hobby" until I can either afford nai or MOSS anime releases.
I wish everyone good genning and have fun
>>
>>110274788
>>110274839
https://civitai.red/images/143613812
https://civitai.red/images/134539610
https://civitai.red/images/147046214
https://civitai.red/images/147046137
https://civitai.red/images/147047048
https://civitai.red/images/144284663
https://civitai.red/images/145690373
https://civitai.red/images/144874256
https://civitai.red/images/144507210
https://civitai.red/images/145693022
Also, this isn't anad hominem post
>Krea gets the shapes right
>but colors and shading in anime stuff feel a little off. I can’t say what it is, but you can tell.
>>
>this isn't an ad hominem post
>redirects to 10 link using the same person as example
>>
>>110274871
These don't have the same coloring or shape-sense at all. There's nothing that really unites these except they're by the same user.
>>
File: kishipost 14.png (1023 KB, 1024x1024)
1023 KB PNG
>>110274761
>>
>>110274901
look just use NAI okay, local sucks
>>
>>110274901
The unifying aspect is that they are all slop.
>>
>>110274901
/lsg/ - local slander general
>>
>>110274969
Well yes, they are AI made, but it seems like anon was lying about the shape and color stuff.
>>
>>110274871
Have you ever heard of the Streisand Effect?
>>
>>110274865
Who are you, come on only a few avatarfags posted today
>>
>>110274934
Dunno if I asked before but have you ever genned any of the girls peeing? Like, I dunno, squatting down behind a bush during the chuunin exams and taking a leak.
>>
>>110274901
tl;dr he's someone who posts here and the anti-local trolls all hate him because he BTFOs their narrative about local being bad
>>
>>110275099
Don't think you did, And I wouldn't gen that, not my thing at all.
>>
>>110275221
All good, I'll jot it down so I know not to ask in the future. How about yuri stuff? Tsunade and Shizune making out for example.
>>
>>110275158
>BTFOs their narrative about local being bad
>looks at said gens
>it's all shit

Anon....
>>
"long skirt + black thighs" are fighting for attention skirt ends up shorter than it should be or cut on sides or in any other ways diffusion tries to make sure it shows that zettai ryouiki on image while diffusing into final result
wtf I'm supposed to do? prompt black pantyhose instead?
>>
>>110275235
You're obligated to think that or abandon your false narrative.
>>
File: anima_aestheticV11_00826_.png (1.82 MB, 1344x1728)
1.82 MB PNG
>>
>>110274865
Next OP with pic related
>>
>>110275382
Why do you want to show everyone that the model you pay monthly can't do text?
>>
>>110275430
NAI can do text, just not quite as well as Krea2
>>
File: ComfyUI_31516_.png (916 KB, 1344x1156)
916 KB PNG
>>110275315
long skirt, thighhighs
if does not work, add 1.5 to 2.1 weight to long skirt tag
also you want kneehighs perhaps?
>>
File: 261108130517_output-0001.png (1.92 MB, 1037x1516)
1.92 MB PNG
>>
File: Shizo.png (1.15 MB, 1664x2432)
1.15 MB PNG
>>110275430
Relax, man.
>>
>>110275569
kek, shoulda went that route for my schizo sleuth gens
>>
File: ComfyUI_36798_.png (1.83 MB, 1408x1728)
1.83 MB PNG
>>
>>110275984
I want to pinch those nips.
>>
File: ComfyUI_36831_.png (2 MB, 1920x1152)
2 MB PNG
>>110276121
what about pinching chinks?
>>
File: chibicap.mp4 (3.39 MB, 1172x728)
3.39 MB
3.39 MB MP4
Here's a video of chibicap doing face selection for character grounding. Pretty much it's a complete functional application end-to-end. I'd like to find people willing to help process shows with it, but I need to think about how to set up the dataset storage - probably do it as a huggingface space with a check-in/out system for shows so people don't waste time creating duplicates. Other issue is I very much do not want to go back to using discord.
Pretty much any CUDA-capable card with a CU >=7 is fine, but you'll need to set up an openrouter account and fund it with credits for the captioning. It even works on CPU-only, but face-detection and grounding are super-slow that way.
Ah fuck, I should probably just do it all myself.
>>
File: bocc.jpg (294 KB, 1152x1728)
294 KB JPG
>>
>>
>>110276469
This gen reminded me of Kishimoto's great intuition re:their headbands, expressing individuality with the way characters wear them.
>>
>>110276398
What a plappable Bocc.

>>110276245
Neat.

>Ah fuck, I should probably just do it all myself.
Easier today than ever at least, thanks to LLMs.
>>
>>110276245
Interesting. I feel more of that can be automated.
>>
Moss is not very good at anime, huh? Also super fucking slow
>>
>>
>>110276804

Beautiful
>>
File: b5edxt.jpg (141 KB, 500x719)
141 KB JPG
>>110276245
Cool anon, which base model were you thinking of when you wrote this? ;^)
>>110276729
>Pic rel
>>
>>110276469
Holy Hinata hips
>>
>>110276208
>A chink is fine, too.
>>
Even in the best case where we get a Moss finetune and Noob DiT finetune, neither can be trained on 16GB of VRAM or leas in case of Loras.
>>
fresh
>>110277037
>>110277037
>>110277037
>>
>>110276923
Moss is indeed a trashcan full of used diapers.
>>
>>110271335
>resubbed
>B A S E D
You can tell you're in the presence of a shill when they celebrate each new customer. Bizarre stuff.
>>
>>110276611
The only part not easily automated is character grounding, meaning the use of reference-based face detection to pre-tag all the images with who is in each image. Pre-tagging the images allows you to tell the vision model doing the captioning that a character is definitely in an image. This improves quality and accuracy a lot. You could probably cut that step down to manually selecting just one face and providing a description, and letting the LLM use the images with the shortest distance from that as the other references. I've tried having the LLM write what distinguishes a character, but so far it's done a shitty job of it.
I feel like it's better to keep the human element and maintain a quality dataset, rather than rush it and have it end up with embarrassing mistakes and be a civit.ai-tier piece of shit.
>>
File: Krea2_turbo_00197_.png (2.24 MB, 1448x1448)
2.24 MB PNG
>>110276923
I'm not really sure. I'm curious to see what actual finetuners think in terms of what modern model can actually stand up to a full finetune without falling apart. I'm just focusing on building a canon dataset, I don't even have a big boy GPU anymore, I sold 'em,



[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.