[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: 1760131896522603.jpg (1.93 MB, 2738x2683)
1.93 MB JPG
Previous: >>109157236
NEW: https://huggingface.co/circlestone-labs/Anima
Anima1.0 is out

>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

>How to generate anime:
image : https://rentry.org/comfyui_guide_1girl
video: https://rentry.org/wan22ldgguide

>Style Explorers
Anima: https://anima.mooshieblob.com/
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 (embed)
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 (embed) >>106520607 (Dead)

>/adt/ logo:
https://files.catbox.moe/pakfgt.png (embed)
>>
>>109225172
wrong board >>>/lgbt/
>>
File: 1777800219297491.jpg (752 KB, 1248x1824)
752 KB JPG
>>
File: 1782507878503150.jpg (195 KB, 768x1024)
195 KB JPG
this is my wife
>>
File: ComfyUI_07234_.jpg (327 KB, 1152x1536)
327 KB JPG
>>109225329
this is a horse
>>
File: ui-smile.png (1.64 MB, 1248x1824)
1.64 MB PNG
>>
Highlights yay!
Also I feel its time for native American indians
>>
>>109225411
She should lay off the carrots
>>
>>109225586
the carrots in america have different nutritional value
>>109225562
did you use heart_in_eye for this or is that a whim of the model?
>>
>>109225172
Shit collage, why no _cui?
>>
>>109225172
1girl, standing garbage
>>
File: 1783457819556360.jpg (968 KB, 1600x2048)
968 KB JPG
the only anon doing anything interesting with 1girl standing and he's not in the collage, this is a reflection of this shit general and the shit people who post here.
>>
criticism of 1girl standing is always so basic, like what do you think 90% of real artists draw?
>>
I never get picked either but im meh about it
I just wanna look at pictures
>>
>>109226381
I'm i'm loooooning
>>
>>109226246
What's being discussed here is the systematic discrimination against _cui, who recurrently comes here to help, interacts with anons, even the ones faking problems just to keep this dead general alive.

I support _cui tho and not putting him in the collage is really greedy of you narcissistic troons
>>
File: Krea2_turbo_00311_.png (1.73 MB, 1928x1088)
1.73 MB PNG
Asking krea2 to gen front-back views so they can be made into two-sided acrylic keychains. it's close enough on about 1 in 4.
>>
Oh my god the collage is back, we can be frens again
>>
>>109226805
Cool
>>
>>109226570
I like it when people post gen

A lot of fellow anon here make pretty picture, not just the people you have a bitter vendetta against

I like the collage its a lot of pictures together, I already know I am the best poster on Reddit so idgaf that much about ‘making the cut’
>>
>>109225172
>nipples not even protudring
trash
>>
>>109226246
Real artists *paint* compositions like these.
>>
How do I find artist tags?
>>
>>109227090
uhh use danbooru?
>>
File: ComfyUI_00536_.png (674 KB, 1024x1024)
674 KB PNG
>>
File: y3OB2H7IRCvP.webm (957 KB, 960x1216)
957 KB
957 KB WEBM
>>
>>109227104
there is no way that's the best way
>>
>>109227090
https://anima.mooshieblob.com/ (Anima)
https://tagexplorer.github.io/ (SDXL)
>>
>>109227552
based
>>
File: file.png (66 KB, 185x289)
66 KB PNG
>>109227552
????
>>
File: 0.png (985 KB, 1344x1728)
985 KB PNG
>>109224672
why would I do that? I like that it's AI, I think the tech is cool and I like making nice pictures.
Besides, there's a bunch of little things that give it away.
>>
File: ComfyUI_temp_zzhev_00005_.png (2.6 MB, 1936x1088)
2.6 MB PNG
ID4 mirrors the images better, but makes shitty Mikus, and it took almost 2 minutes to gen on a 4090D 48GB.
>>
File: NA1.jpg (2.76 MB, 1648x2208)
2.76 MB JPG
https://youtu.be/DvfX5hbhBnw
>>
File: ComfyUI_temp_ssfpc_00011_.png (1.58 MB, 1024x1368)
1.58 MB PNG
>>
File: demon.jpg (404 KB, 1152x1536)
404 KB JPG
>>
Anima-Turbo and Anima-Aesthetic
https://civitai.com/models/2458426/anima?modelVersionId=3108589
https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/diffusion_models
>>
>>109229752
poggers
>>
File: neat.png (1.98 MB, 1536x768)
1.98 MB PNG
>>109229752
>>
>>109229752
cool, any comparisons?
>>
>>109230075
Here you go bro
>>>/h/8910234
>>>/h/8910250
>>>/h/8910261
>>
>>109227924
how about genning a desired front and using controlnet as a stencil mask for back
>>
File: 1777575253001474.jpg (1.66 MB, 3456x1344)
1.66 MB JPG
>>109229752
>>109230075
Turbo seems a lot worse than base, but I can see it being useful if you're trying to goon

Left is turbo (cfg 1, 12 steps)
Right is base (cfg 5, 30 steps)
The built-in NAG that's in ComfyUI didn't seem to do much for turbo
>>
>>109230433
negative prompt doesn't work with turbo. you can use this https://github.com/pamparamm/ComfyUI-ppm if you want negative prompt with turbo
>>
>>109229988
That's very cute
>>
>>109230509
ur cute!
>>
>>109230492
Thanks, didn't know about those nodes
>>
>>109230375
Oh hgg/, what a solid thread. Thanks for sharing, anon, not even sure how I ended up here.
>>
File: Y8UOIxAPv41V8i9lsutK.mp4 (212 KB, 460x344)
212 KB
212 KB MP4
>>109230375
>>
File: ComfyUI_ID4_00047_.png (1.8 MB, 1936x1088)
1.8 MB PNG
>>109230381
I was able to improve it by adjusting the prompt to include more Migu details.
>>
File: ComfyUI_ID4_00080_.png (1.07 MB, 1936x1088)
1.07 MB PNG
My other favorite Miku (from I want to End this Love Game)
>>
File: cui_miolenia_0001.jpg (767 KB, 2048x1600)
767 KB JPG
>>109225812
That's not true, lots of good gens here.
>>
nahida /(genshin impact/), (chibi:1.2), (adorable:1.1), cute girl, little girl, small female, loli, aged down, younger, toddler, (plump:1.1), (chubby:1.1), (short female:1.1), (short legs:1.1), shiny skin, black flip flops, (perfect feet:1.2), (chibi feet:1.2), (loli feet:1.2), (small feet:1.3), feet, toes, toenail polish, lime green nails, blush, happy, head tilt, one eye closed, open smile, green apron, standing, arms behind back, pigeon-toed
>>
>>109231092
haha is tag spam a phase we all went through?
>>
with prompt scheduling, how do i alternate between 3 choices every 0.05? is it [a|b|c:0.05]?
>>
File: orig_00035_.png (1.21 MB, 1024x1440)
1.21 MB PNG
trying some more of the red.
>>
>>109230433
>>109229752
Kinda sad watching the Anima copers run test grids on these 2 new models like they won't be total nothingburgers without an updated dataset.
Dumping this much time and money into this 2 models when CivitAI is already drowning in shit mixes (animayume, one obsession, waiANIMA, hassaku anima) and turbo versions (RDBT), only makes sense if he's building his own in house lineup so buzz stays inside Circlestone Labs instead of bleeding out to every checkpoint maker on the periphery which is a reasonable economic call.
>>
File: orig_00044_.png (1.41 MB, 1024x1440)
1.41 MB PNG
>>109231225
>>
anyone installed uv for forge neo? any tips?
>>
hello, first time lora baker here, does anyone know the difference between cosine and constant?
>>
this is fine as long as it keeps the thread alive, right?
>>
>>109231916
Now that the collage is back, the thread will be kept alive by the power of anons chasing the spotlight!
>>
chat, is that post with the "interesting" random background with melted flowers image serious or not? i can't tell what's bait anymore.
>>
>>109231225
i like this one
>>
File: ComfyUI_24419_.png (3.39 MB, 1248x1824)
3.39 MB PNG
>>109231485
Never used constant. I usually use cosine (with Prodigy) and it's never steered me astray. As to the actual mathematical details I'm afraid that's for the nerds to know
>>
baited
>>
>>109231225
blue > red
>>
>>109231485
Constant keeps the learning rate at a constant number
Cosine decreases the learning rate after each step https://docs.pytorch.org/docs/2.13/generated/torch.optim.lr_scheduler.CosineAnnealingLR.html

As for which is better, I just use constant and my loras all turn out fine
>>
File: ComfyUI_20281_.jpg (192 KB, 1152x1344)
192 KB JPG
>>
whar artist or tag makes those blushes that are a bunch of red lines?
>>
Can I fix anatomy by adding steps?
>>
>>109231053
Cum eater cuck
>>
>>109233104
@grok verify the authenticity of this post.
>>
>>109225172
dry dick haver conference art collage
>>
>>109233179
your foreskin status?
>>
>>109233186
I actually have a mild inflammation, thanks for asking
>>
If I ask for looking at viewer, how does the AI know where I'm sitting?
>>
File: ComfyUI_temp_puvoa_00109_.png (2.91 MB, 1280x1536)
2.91 MB PNG
Oh no!!! You were just minding your own business at the beach and suddenly now all of a sudden recently see this! What do?
>>
Lolis are old news unc
>>
Lolis are kek XD tier
>>
>>109233845
Turn her into my own saliva-producing slave. Just make her drool 24/7. Looks like she already started before I said anything, this is gonna be a good one.
>>
>>109233984
Kekarooooo fpbp
>>
can whoever made this sortable be most faves?
https://anima.mooshieblob.com/
>>
>>109234006
create a github issue retard https://github.com/Mooshieblob1/mooshie-anima-styles
>>
Bump
>>
>>109225172
Not my general
>>
>>109233845
she has a name you know
>>
....
You were powerless when faced with the promise of rock CDs.
There was nothing you could do against the power of Rolling Rock.
Your mind goes blank at the threshold of your house.
How did this happen?
She is Get Pregnant.
Bad end.
(christ it took a long time to get that proompt, its way past bedtime)
>>
>>109232069
>This is what Reifags tell themselves to sleep at night
>>
>>109231225
>>109231287
is this a lora? can i have it
>>
>>109234513
>>109224389
>>
>>109234513
Stop jumping to the conclusion that everything is a lora. Your base model can do SO much more than you think, you just have to learn how to play with it. Open the danbooru wiki and start searching things.
>>
>>109234260
is this a lora
>>
File: CUNFY_00599_.jpg (2.33 MB, 1664x2432)
2.33 MB JPG
>>
Does 1girl mean exactly one girl, or just that the main subject is a girl?
>>
>>109234854
>when you have swimming class in fourth period but you’re a goal-orientated girl who prepares for her day so you sit in swimwear for 3 hours during Japanese history
Hawt
>>
>>109235448
You cannot add any human other than the one-girl
I don’t make the rules I just bother you with them
>>
>>109235448
1girl means there is a single girl in the image. there could be other subjects which are also tagged like 2boy, 1other etc
>>
>>109235470
and isnt' solo redundant if I'm already using 1girl?
>>
>>109235518
If you forget solo, the AI invite friends to the gen, and you don't want that
>>
>>109235518
no, surprisingly. images with different panels or something like a character sheet still count as 1girl if its the same girl. solo is useful for stopping this or multiple views in the negative prompt
>>
>>109235518
Where you always this argumentative anon? Xd
>>
Should I separate tags with commas, spaces, or both?
>>
>>109235518
Generally speaking, you need to do everything in your power to tardwrangle the robot.
>>
>1girl
>1
Can local AI count objects?
>>
>>109236022
no, but humans can, and that means that the training data includes it.
>>
Spent a little over 2 hours on this, would've gone way faster if I generated one full image and lazily cropped it for the panels. For the text on the hands, I wrote it manually first before reprocessing. Didn't bother to get every costume detail correct.
The character design is annoyingly complex without nice easy images around.
>>
>>
>>109236052
is this memento guy
>>
>>109236069
the what
>>
>>109236069
Is the foreskin guy
>>
>>109236072
The guy from the Christopher Nolan film "Memento".

>>109236105
Aren't there a bunch of those?
>>
File: m0Qk8.gif (1015 KB, 249x189)
1015 KB GIF
>>109236123
Had to go find it. It's from "Strangers with Candy" Comedy Central show (1999-2001), idk the episode. I haven't seen this gif in forever and it was in my mind when I made the comic. I didn't even remember what the guy looked like or exactly what he was reaching for.
>>
Do older checkpoints understand newer Danbooru tags?
>>
>>109236687
does old databases know new data?
>>
>>109225411
Make the skirt shorter
Also catbox?
>>
>>109234841
no, just anima aesthetic. that one might have a little rendering detail lora idr.
>>
>>
>>
>>
>>
>>
Can negative prompts accidentally remove things I wanted?
>>
>>109237610
are you this guy?
>>109236687
>>
>>109235448
>>109235518
you can literally just open the danbooru wiki and READ
>>109237610
ok that's it
no more stupid questions
only anime girls from now on
>>
need more bratty hebes
>>
>>109237610
if you're using the turbo lora (or turbo anima), it ignores the standard negative prompt
>>
File: 223021CUI_00001_.png (1.47 MB, 1152x1536)
1.47 MB PNG
>>
>>109237769
Looks like you got lost catjack >>>/g/ldg/
>>
>>109237734
the fuck is a "hebes"?
>>
>>
Can I mix realistic and anime tags, or does that confuse the model?
>>
>
>>
File: CUNFY_00604_.jpg (2.52 MB, 1664x2432)
2.52 MB JPG
>>
>>109237912
small females aged roughly around the beginning of puberty
>>
>>109238175
hebe lora doko?
>>
>>109237758
Not really, that's just an effect of CFG being at 1. You can put it to 1.5 or 2 for turbo, which is slower but you still only need about 12 steps for a decent image and get more stability.

As a side note I'm surprised how much the anima turbo checkpoint sucks. It feels undercooked and the results I get with base + turbo lora are way better at CFG 1 than with the checkpoint. It almost needs you to use 2 if you don't want missing fingers and vague details all over the place and even then I don't think it's as good as base+lora.
>>
>>109238427
>I'm surprised how much the anima turbo checkpoint sucks
This isn't the first time I've heard this, and I don't get it. It's the exact opposite for me. The turbo 0.2 lora isn't terrible, but it's like it's overcooked: too bold, saturated, melty noisy details, puts steam clouds / sweatdrops everywhere etc... Turbo checkpoint just feels like a fairly neutral model which is good.
>>
File: ComfyUI_00040_.png (1.36 MB, 832x1216)
1.36 MB PNG
Came back from using NovelAI, since I thought using ComfyUI was gonna be very complicated. Turns out, it is actually easier than I thought. Just need to figure out how to tune things more and learn how to use LORAs more properly.
It's like a mini-game.
>>
>>109238476
Turbo 0.1 is probably the best for CFG 1 imo, it lacks a lot of detail but it's really nice and stable for traditional flat color clean line art type stuff. Turbo 0.2 is good at about 0.8 lora strength and CFG 1.5 imo. Gets a rid of a lot of noise and brings more stability than leaving CFG at 1. I also like to mix in a bit of rdbt lora (0.37 at 0.5 str is nice) with turbo 0.2.
Maybe I still need to figure out how to get most out of the new turbo checkpoint but so far it feels like I generally get a result that is worse in detail than 0.2 and worse in stability than 0.1.
>>
>>109238520
i also like the turbo loras and don't like the turbo checkpoint. i hope he continues working on the turbo lora, it's also nice that you can use it with other anima based checkpoints
>>
>>109238174
she need to wear a police uniform and look at me with disgust (lust)
>>
File: 00261-1513505297.png (2.65 MB, 1792x2304)
2.65 MB PNG
>>
File: file.jpg (266 KB, 1024x1024)
266 KB JPG
>>
File: 00265-420326219.png (2.67 MB, 1792x2304)
2.67 MB PNG
>>
>>109238479
Return to NAI anon
>>
Haven't used WAI since v14. Is v17 strictly superior?
>>
>>109235518
Image generation models don't literally "count" the way a computer program does. A tag like 1girl is a learned concept from training data that biases the model toward generating a single female character. It's a strong instruction, but not an absolute rule. That's why you can occasionally get extra people, partial figures, reflections, or background characters even when you specify 1girl. The model is predicting pixels based on patterns, not verifying that the final image contains exactly one person.
>>
>>109238092
Yes, in a sense, but not the human sense of being confused. If the prompt contains conflicting conditioning signals the model produces a less coherent result.
>>
>>109239139
It's probably the better recommendation for most users, but whether it's better depends on what you're optimizing for. Newer releases typically incorporate better training, cleaner data, and improvements in anatomy and prompt following.
>>
>>109238533
I came across dummy9996/anima-turbo-lora-extract on huggingface. It seems to actually work which surprised me considering it's like 15 MB while the official 0.1 and 0.2 are 10 times that.
>>
>>109239139
There isn't really a best, it depends on what you want your outputs to look like as different versions have different amounts of different bias. Just gonna have to test and see. Personally I actually only really liked the early ones that were close to base Illustrious but just a little more clean/stable instead of having too much aesthetical correction to the point where it's hard to replicate some simpler flat 2d styles.
>>
>>109239188
>better training
lmao wai is 100% other people's loras and checkpoints merged into a merge of illustrious and noob

the waitard has never trained anything. even wai anima is literally just 50% animayume 50% anima base (he got found out because he used comfy to merge it and didn't realize it puts the workflow in the checkpoint metadata)
>>
>>109239188
>better training
>cleaner data
don't talk shit if you don't know
>>
>>109238092
Yes, you can mix them, whether it actually works depends on your model. Anime models toss in a couple realism tags like photorealistic backgrounds or realistic lighting and you get that nice semi real anime look but if you push it too far into realism shading gets weird.
That is because some tags are just enemies like anime vs photorealistic, cel shading vs real skin texture, 2d illustration vs RAW photos and lineart vs photography. If you feed the model both sides and it tries to please everyone probably the result will be the classic plastic skin or the classic shitmix weird textures that make no sense.
>>
>>109239247
Thank you, are hatsune miku and kasane teto tag enemies to?
>>
>>109239283
Not necessarily, Hatsune Miku and Kasane Teto aren't tag enemies in the same way anime and photorealistic are becauae they're just tags for different characters. If you use both tags without further description, most anime models will assume you want both characters present so you will often get two girls.
For example:
Hatsune Miku, twin tails, teal hair = clearly Miku.
Kasane Teto, drill twintails, pink hair = clearly Teto.
Hatsune Miku, Kasane Teto = usually interpreted as both characters together.
>>
>>109239188
>>109239228

Thanks I'm pretty much just looking for a model to keep learning Comfy with, Anima really can't do some things so it's pointless to learn those functions with it
>>
>>109239247
>>109239313
Thank you for your help kind anon, what luck i had finding this great thread.
>>
...and that's how /adt/ lived happily ever after
>>
>>109232652
As primitive as this gen is, what's funny is that it's still more complex than that artist's actual output because at least there's a background. Ironic.
>>
any idea which artist this is? https://i.4cdn.org/g/1783655879055488.png
>>
File: catbox_cxi9az.png (1.79 MB, 1664x1152)
1.79 MB PNG
I've been playing around with Anima for days and I'm figuring out how to make good-looking shit like in picrel (made by some anon from non-blue board)
.t someone who's been using a NAI/Illustrious shitmix
>>
>>109233077
nonny why dont you try some of these things? this one in particular is very simple to test. lock the seed and re-gen while only changing the step count by 1 in either direction. the short answer is yes but the nuanced answer you need to see to really learn.
>>
>>109239432
>garbled text
aw
>>
>>109239432
>good-looking shit like in picrel
???
There are melted details and broken lines everywhere. What about this is supposed to be good?
>>
>>109239524
>>109239545
I should have posted my gen sorry
>>
File: happy-meek-sleep.png (1.03 MB, 1536x1536)
1.03 MB PNG
>>
https://files.catbox.moe/1ow9ls.png
>>
https://files.catbox.moe/l5gygp.png
>>
>>109240129
where is his cock?
>>
>>109237769
The expression is very nice in this
>>
>>109239432
you already have the prompt so go learn from it?
>>
>>109239424
it's a lora.
>>
File: red.jpg (983 KB, 1248x1536)
983 KB JPG
cool
>>
>>109239240
>>109239245
Look at this guy, self proclaimed local sommelier of every dead on arrival shitmix, can recite the whole family tree of garbage checkpoints nobody asked about, embarrassing knoweldge desu.
>>
>>109240353
proof?
>>
>>109240613
Meanwhile I pay 25 a month and couldn't name a single one of their hundred failed checkpoints if you paid me. Not my problem, that's what I'm paying for.
>>
>>109240078
>>
>meanwhile I pay 25 a month to suck cocks
we know
>>
File: 1770767538314767.jpg (730 KB, 1344x1728)
730 KB JPG
>>109237734
>>
>>109240752
If we both died tomorrow the difference between us is that I spent this whole time generating high quality gens with the fastest, best anime model, living kino every session while you spent it memorizing the origin story of every DOA shitmerge.
All that NetaYume and WAI trivia you acumulated, all those hours you spent into it, bought you nothing, you die tomorrow and that's it, you never had fun, you never experienced kino, you never anime.
>>
File: machine_red.png (200 B, 40x40)
200 B PNG
>>109231225
>>
AI is so cool
it can generate pictures of my wife
>>
>>109240862
Show us a pic of your wife
>>
>>109240878
no
>>
If I don't trust custom nodes, can I still detail faces at all, or am I stuck?
>>
File: 570149669664659.png (1.68 MB, 1024x1536)
1.68 MB PNG
>>
>>109239153
>occasionally
>>>
>>
If I use the same seed on a different checkpoint, shouldn't I get the same picture?
>>
>>109241588
most of the time no. if you're using a finetune from another checkpoint, the images might look closer (similar pose, same angle etc)
>>
>>109241665
Thanks,
>>
File: ???.jpg (529 KB, 1344x1344)
529 KB JPG
>>
>>109241718
Just because hes black
>>
https://civitai.red/models/2767267/anima-x
thuoghts on x pred, dear /g/ anime general?
>>
>>109241972
So X-pred is Anima but for Twitter images specifically?
>>
>>109241972
pointless
>>
>>109241972
Is this the uncensored version? I heard X-rated models predict differently.
>>
>>109242155
yes. /hgg/ also has a bibicipred version if you want to try it
>>
>>109242164
Can I put the model anywhere on my drive?
>>
>>109242179
you will have to ask the author for box
>>
>>109240353
Does the LoRA go in the diffusion_models folder too?
>>
>>109242199
No in the /lora/ folder
>>
>>109241718
Fix the horse's leg nigga
>>
>>109242451
that's not its leg
>>
>>109241718
By the way she has twice the heat and tingling between her legs while riding a (male) horse compared to your typical 1girl. I'm an expert, trust me.
>>
>>109240729
ty it was chilly
>>
File: cui_yuffie_0082.jpg (934 KB, 1600x2048)
934 KB JPG
>>109239139
The WAIs are all very similar if not outright identical in style from what I can tell, 17 will just have more up-to-date knowledge than 14, so yes I guess it's "strictly" better, but that's about the only occasion I can call one model "strictly" better than another since it's up to taste otherwise. There are models where I definitely prefer older versions to new ones, e.g. ramthrust
>>
>>109243051
Masamune, my beloved..
>>
File: nahida-2.png (1.6 MB, 1248x1824)
1.6 MB PNG
>>
>>109244259
can you make her ride a horse
>>
>>109244282
yes
>>
Can I use SDXL prompts on Pony without changing anything?
>>
>>109244514
pony is SDXL...
>>
>>109244521
Wait, so what's the difference between SDXL and Pony then?
>>
>>109244540
Pony it's a SDXL fine tune. SDXL is the general purpose base model and Pony was trained further for stylized anime character art and works better with booru tag style prompts.
>>
>>109244556
Thanks, so SDXL is the engine and Pony is the car?
>>
File: 1762019190875982.jpg (113 KB, 832x1216)
113 KB JPG
>>109225172
Never used any Krea models before. My recent experience with realism models are the flux Klein models. How good are they and how many the steps are usually needed to make a good output?
>>
>>109244674
For RAW arround 30-36 steps is enough for most stuff, 50 if you want max quality. Turbo 8 steps with CFG off , a lot of anons use the Turbo lora on RAW around 12-20 steps since it's a nice middle ground I'd say it has a more cinematic look than FLUX and responds better to natural language prompts.
>>
File: 1765555213955989.png (788 KB, 832x888)
788 KB PNG
>>109225172
I installed ComfyUi with Anima not long ago and I'm very surprised by the results, even though I have no idea of what I'm doing. I need to learn how to proompt better.
>>
>>109244866
he caught her before she fell?
>>
>>109244866
If you've got stupid questions, don't be shy about throwing them out here, we are basically the specialists in answering stupid questions.
>>
>>109244878
Yeah, he's such a kind and caring brother.
>>
>>109244572
Pretty much, SDXL is the base platform, Pony is an SDXL fine tune built on top of it. Think of SDXL as the stock car and Pony as the same car modified for a specific job.
>>
>>
>>109244866
is this some kind of fake kemono friend?
>>
File: CUNFY_00605_.jpg (2.2 MB, 1664x2432)
2.2 MB JPG
>>
>>
>>109245141
You never post the aftermath, boring...
>>
>>109244945
>>109245141
>the writer's barely disguised fetish
>>
>>109245149
That would be pretty messy
>>
>>109244282
>>
>>109245468
anima won big lee
>>
>>109245468
wow my girlies are so cute
>>
>>109245551
I don't think so. Anima is stiff as hell, and those characters are mainly what it was trained on, they all belong to the same generic animeesque gacha bait style.
>>109245468
Could you please use characters from two different franchises instead of gacha slop characters? For example, Mileena from Mortal Kombat and Cnoc Na Riabh swimsuit foreigner from Fate? Also, if possible, have each one cosplay as the other, so we can see whether any non specific clothing details bleed into the other outfit, that would be great.
>>
i shall not. you may take your win.
>>
>>109245715
>Could you please use characters from two different franchises instead of gacha slop characters? For example, Mileena from Mortal Kombat and Cnoc Na Riabh swimsuit foreigner from Fate? Also, if possible, have each one cosplay as the other, so we can see whether any non specific clothing details bleed into the other outfit, that would be great.
NAI can't do this.
>>
>>109245777
based
>>
it is quite adamant that the boots be "similar"
>>
Is there a point where adding more tags actually makes things worse?
>>
>>109245986
There's a point where adding more tags does nothing because it's past the context limit.
>>
>>109246024
And can I overfit my prompts?
>>
>>109246038
yes. sometimes adding more details confuses the model
>>
>>109245777
>>109245468
Can anima gen bottomless Nahida butt
>>
>>109246411
yes
>>
>>109246411
what kind of retarded question is this
>>
>>109245468
exactly what I had in mind, thanks
>>
>>109245815
asymmetrical footwear
>>
File: nahida-laugh-melty.png (1.31 MB, 1248x1824)
1.31 MB PNG
>>
Can I put my Loras inside the VAE folder since both affect how it looks?
>>
>>109247357
try it and let us know
>>
>>109247365
I restarted Comfy. Am I supposed to refresh the models separately?
>>
>>109247428
Did you select the VAE first faggot?
>>
File: CUNFY_00606_.jpg (1.82 MB, 1664x2432)
1.82 MB JPG
>>
>>109247357
no, a lora isn't a vae
the loader only looks for loras in the lora folder and vaes in the vae folder.
>>
File: ougi.png (1.03 MB, 1248x1824)
1.03 MB PNG
>>
If CLIP never sees the generated image, how does CFG know whether the image followed the prompt?
>>
File: 1782982502685677.png (396 KB, 976x549)
396 KB PNG
i still don't like anima
>>
>>109225172
I don't think it's a good idea to make a collage during such a low point for /adt/. Anyone with a bit of sense can tell there are only 3 genners itt and in the collage.
>>
>>109248389
There's no other way around it bro, I wish I was 15 years younger so I could skip this whole stagnant period and pick the hobby back up again once everything was two or three times as advanced. It hurts looking back 6 months since preview 1 dropped and seeing that base is already out and it's absolutely nothing new, and now the dev has put out 3 more variants that don't offer anything either.
>>
>>109248444
There are only max 5 people at any thread at any given time.
>>
>>109248535
Well, but we have two fewer than the other threads.
>>
>>109248535
it's kind of cringe when a general has been running for a year but only 3 people participate in the collage
>>
>>109248444
>there are only two other than me
wao
>>
File: 00022-2956208410.png (791 KB, 832x1216)
791 KB PNG
>>109248699
woa even
>>
File: meguin memian.jpg (303 KB, 1152x1536)
303 KB JPG
>>109248535
/g/ has been declining since 2019
>>
>>109248512
it's hard to notice, but aesthetic is a little better on average. just barely.
>>
>>109248444
everything i make is nsfw sorry m8
>>
>>109248759
are those her..*pukes*
>>
File: slamkeyboard.gif (17 KB, 128x128)
17 KB GIF
does anyone have a custom workflow to use within krita with anima? i tried making my own using their provided example but it fails with the error "image encoder vae default for sd1.5 not found" and i have to restart comfy for krita to be able to connect to it again (maybe disconnecting in the settings after restarting it would be enough)

i tried loading the model and vae etc as i do in regular workflows but it didn't work as well
>>
>>109250527
Here you go: https://litter.catbox.moe/4bwijgtpjcktfvtg.json
>>
>>109251329
thanks. i actually managed to get it to work but the one you linked is better.
>>
>>109248759
tentacles >>> saggy naturals
>>
File: cui_U1196_thurandaut_0001.jpg (1.4 MB, 2048x1600)
1.4 MB JPG
>>
>>109248512
i still get better gens on illustrious
>>
File: 1762025356731751.jpg (2.12 MB, 2048x3072)
2.12 MB JPG
Trying to slop mix my way into getting at least a balance of artist and medium style accuracy but it's rough since anima really doesn't want to blend styles without converging into bland medians. How many weeks until the perfect model comes along?
>>
>>109237912
He really want some bad Jewish girls
>>
>>109252032
Truthclear winter
>>
>>109252032
yep. and they finish in a fraction of the time, to boot. grim stuff.
>>
>>109251944
cringe
>>
File: 00066-4200447769.png (471 KB, 832x1216)
471 KB PNG
>>109253007
it's cui
>>
>>109253024
I do not understand this faggot's point of view. That is not anime but some random woman knight from a forgotten video game, like Monkey Island, standing there stiff as a statue, and he's calling it anime?? I don't get it and I refuse to ever get it, jow do you even make that connection. Every single day I pray his GPU just dies on him, or ComfyUI throws a fit, or his drivers finally give out.
>>
File: kotfeet.jpg (292 KB, 1368x896)
292 KB JPG
>>
>>109240910
What reason do you have to trust the base nodes?
>>
File: nv4.png (41 KB, 777x851)
41 KB PNG
How's this for Anima lora training? I'm getting results that aren't bad but the Illustrious/NoobAI loras I made using the same dataset and tags more closely resemble the characters themselves, minuscule details but it's there. Dataset is around 35-40 images per character made with bigger corpo models (Nano Banana, GPT Image, etc)
>>
>>109253982
might need more steps
I had good results with 2500 steps
>>
File: 1687176916416756.gif (776 KB, 320x348)
776 KB GIF
Anima Aesthetic actually looks worse than Anima Unaesthetic sometimes.
>>
>>109244945
>>109245141
Kawaii Idolmaster AIgold
>>
>>109240795
i like this one
>>
>>109254352
it's Comfy Org. they're to blame for this.
>>
>>109253255
cvte
>>
>>109253255
nekobox for cutie?
>>
File: 1757225239595380.jpg (821 KB, 1072x1440)
821 KB JPG
>>
>>109255364
she gens more like this if you don't bother to get the loras
https://litter.catbox.moe/24kylp7u93x8cdug.png
>>
>>109255655
>>109253255
bad proportions of the lower body. the head should fit 1 and 1/2 times into the femur and tibia, a bit less if we're talking about underaged characters which are more big headed, the torso is fucking long too.
>>
>>109255655 (Me)
awww i didnt notice until just now. in the ':3' branch set cfg to 3.7 to band-aid fix the extra finger.
better seed even: 1076770998085548
>>
>>109255481
this img would actually be beautiful without the 6 yo pedobait
>>
yeah for me it's butiful white backgrounds. 1girl just gets in the way of porcelain perfection.
>>
>>109255952
but that's the best part
>>
>>109256002
proofs?
>>
>>109256008
here bro >>109255481
>>
>>109256015
the five fingers on one hand, with four fingers on the other, and four toes is distracting.
>>
>>109256021
sorry, i'll do better next time...
>>
>>109253024
is she 18+?
>>
>>109255481
my daughterwife waiting for me to take her out for a walk on a sunday morning
>>
>>109256032
ok >:3
>>
>>109256043
are you?
>>
Why do you want to sage the thread?
>>
>>109256095
I have a gf
>>
>>109256255
You don't gain anything by bumping limit this thread, you faggot
>>
>>109251944
Pretty
>>
>>109253255
Id buy the expensive kibbels for her
>>
>number are red now
>>
File: 1781950308405708.png (108 KB, 284x332)
108 KB PNG
I'll bake
>>
>>109256648
>>109256648
>>
>>109254352
it looks better (just barely) usually, but makes certain artists and loras look kind of grotesque in my experience.



[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.