[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

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


Janitor applications are now open. Apply here!


[Advertise on 4chan]


File: adtOP.png (1.83 MB, 1050x1050)
1.83 MB PNG
Previous: >>108859807

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: (free) https://anima.mooshieblob.com/ (grift): https://thetacursed.github.io/Anima-Style-Explorer/
Illustrious: https://tagexplorer.github.io
Neta Lumina: https://neta-lumina-style.tz03.xyz/

>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://catbox.moe|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:
https://gist.github.com/catboxanon/ca46eb79ce55e3216aecab49d5c7a3fb
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

>keywords(test):
ai, anime, diffusion, image, gen, anima, illustrious, noob, pony, local, nano banana, sdxl, klein, flux, grok, novelai, nai, sota, loli, chibi, moe, anime coloring
>>
Nice keywords
>>
File: Nanoda.jpg (490 KB, 1536x1536)
490 KB JPG
>>
File: hiro.png (1.68 MB, 1248x1824)
1.68 MB PNG
good op
>>
>>108884168
now show us her ribs
>>
File: hiro-rib.png (1.41 MB, 1248x1824)
1.41 MB PNG
>>108884178
is this too /e/ for /g/?
>>
File: ComfyUI_13754_.png (1.08 MB, 1024x1024)
1.08 MB PNG
>>
File: image.jpg (182 KB, 850x1202)
182 KB JPG
>>
File: sniperrin.png (39 KB, 812x556)
39 KB PNG
>>108871466
nice thanks
>>
>>108884679
This ain't AI bro.
>>
File: CUNFY_00434_.png (3.59 MB, 1664x2432)
3.59 MB PNG
>>
File: 7257211231.png (2.57 MB, 1664x2432)
2.57 MB PNG
>>
I have a conundrum.

How do you tag/prompt something that isn't itself visible, but it's existence has effect on the character?

Example: Say character is wearing a ponytail. But because the composition is a mug-shot/front facing portrait, the ponytail itself isn't visible, but the way the hair are gathered to the back should be visible.
>>
>>108886027
then the ponytail tag will adsorb other visible things like hair color, bang, face, etc.
>>
>>108886027
There are several ways I use to do stuff like that. One is to simply use a lower weight. Another is to use prompt scheduling to turn on "ponytail" in later steps. And finally another one is to use a known character that looks as close as possible to the character you want. For instance, if she's blond, you can prompt for Samus at a low weight. And combining these can also be effective. But, depending on the prompt and rest of the image, sometimes it just won't work, and you need to just accept it, or inpaint.
>>
File: random test.png (311 KB, 512x512)
311 KB PNG
>>108886027
I guess "straight-on" either "short ponytail" (more likely to show up on top of head) or "short hair, ponytail" (more likely to show up at neck), then either inpaint top or neck areas.
>>
Hi, I'm new here, i want to make fake anime stuff, video game stuff, even lewd stuff

what's the best thing to use?

I tried chatgpt but it keeps telling me to to fuck off with anything lewd
>>
>>108885047
cute butt!
>>
>>108886574
Do you have an NVIDIA GPU?
>>
File: Serena_00021_.png (991 KB, 1216x832)
991 KB PNG
>>
>>108886719
amd 6700xt
>>
File: 01SerenaAmour.png (894 KB, 832x1216)
894 KB PNG
>>
File: 1750115609638482.jpg (674 KB, 2105x1489)
674 KB JPG
>>108884019

What do people use of IMAGE EDIT or INPAINTING nowdays ? Dont tell me its still SDXL
>>
File: 1779489092303112edit.png (1.78 MB, 1248x1824)
1.78 MB PNG
>>108884168
>>
>>108887059
swarmui installer supports amd/rocm ootb
>>
File: choker.png (972 KB, 1152x896)
972 KB PNG
>>
>>108884019
https://animadex.net/
character and style explorer for anima
>>
>>108887648
https://animadex.net/?mode=artists&artist=banned_artist
What does this mean? Who is banning them? Why?
>>
File: urushihara score.jpg (59 KB, 313x523)
59 KB JPG
>>108887648
>Urushihara Satoshi
>Score 17%
Thats fucking wrong!
>>
>>108887648
Can someone explain to me what score is supposed to be? I was under the impression that it was supposed to reflect quality with stuff like lower resolution/high artifact images getting a low score. But based on this website, it looks like low score just means more diversity with artstyle, whereas high score looks like over-rendered slop.

If I'm aiming for the anime screencap aesthetic, should I put low scores in my positives?
>>
>>108887766
>should I put low scores in my positives
what are you saying? score on this website is not the same thing as the score tags in anima
>>
File: top scores.jpg (487 KB, 1435x1185)
487 KB JPG
>>108887766
If you sort by score, the top scored one show clear pattern.

IMO it is judging realism of the picture, as the hair all look similar to how real hair look like. Along with detailed shading.
And why high-quality, but stylized artist like >>108887753 has such low score.

If I were to guess, someone took a realism vision model and asked to score aeshetics of stylized anime pictures.
>>
>>108887785
>>108887766
Oh, yeah Score on that site is not related to the score_ tags used in generators.
>>
>>108887785
I wonder if we could get image score statistics aggregated by artist. Would be much more useful.
>>
>>108887733
It's Danbooru's tag for artists who submitted a takedown notice that bans their work from being reposted on Danbooru.
>>
https://files.catbox.moe/g218hk.png
Japs will call this "safe for work" though
>>
>>108888586
Then when the panties are fully pulled down they'll put a giant ass mosaic on top, so now you see even less: can't see legs, can't even see the navel, half the image is covered in gigantic squares. NOW it's nsfw!
>>
>>108887648
>banned artist
GAY
>>
>>108888507
That sounds absolutely worthless
Maybe they should source their tags elsewhere
>>
>>108888698
>>108888707
Idk what animadex is doing with banned_artist. I know belko and star741 are some and you can search for them just fine. You need a mod/admin account to view "banned" posts on Danbooru (which are fully visible on Gelbooru lmao). Gelbooru only has 18 posts tagged banned_artist that they forgot to remove.
Without such an account on Danbooru the numerical count for banned_artist just shows up as 0 (hardcoded?).
>>
https://files.catbox.moe/19mdu5.png
Risky
>>
>>108888903
No, thats already explicit.
>>
>>108888907
wan 2.2 or ltx 2.3. wan is better for nsfw
>>
>>108888909
I know
(I'm just reposting past slops that I posted on /e/ AI slop thread)
>>
>>108888863
The fact that loli artists don't show tells me everything I need to know
At least the webui plugin doesn't have this problem and I've also created my own version of such a gallery for myself with the older XL models and it still holds up for this.

I might do it again with Anima but genning another 55k images locally is going to be even worse with this model and summer is also starting, pretty bad timing that
>>
>>108888903
Risky you say?
https://files.catbox.moe/gnqbtr.png
>>
>>108889155
Can you do Shantae or Rottytop?
>>
File: 02921-2469687314.png (2.61 MB, 1248x1824)
2.61 MB PNG
>>108889210
Doing something else right now but have an old gen.
>>
File: ComfyUI_00103_.png (2.9 MB, 1440x2160)
2.9 MB PNG
Anima can count to 9. Anima cannot draw a shoelace knot correctly. Anima seems to produce really crusty noise if you use any "_ focus" tags.
>>
>>108889264
Oh and I forgot to mention it will draw 4eared freaks even with extra ears:3 in the neg
>>
For some reason having "MARRY AND REPRODUCE" in you prompt seems to completely fuck Anima up.
>>
>>108889292
>it will draw 4eared freaks even with extra ears:3 in the neg
Yet more proof anime models should exclude blue archive when training
>>
>>108889293
Turns out "AND" is special syntax, further turns out you can't escape it.
>>
>>108889316
>my statistical model is predicting that human heads have ears on the sides
If you don't want them, train a LoRA, that's literally the use case for something like this
>>
>>108889292
negpip helps with this
>>
4 ears > 2 ears.
Yiff in hell, furfags:3
>>
>>108889431
don't troll me bro i'm gonna cry
>>
File: 84014_c810c8d5413a4d0f.png (1.36 MB, 1248x1824)
1.36 MB PNG
>>108887118
damn no gloves or nothing
>>
>>108889372
That shouldn't be necessary when trained on gelbooru. It's not necessary on RouWei which is objectively a much worse model overall. Besides LoRAs for negative concepts don't work and just fuck up outputs.
>>
migu
>>
>>108889546
What tags get you the you've never had sex before face?
>>
>>108889872
aged down
>>
>>108889872
I think it's just Barbara. All I prompted for her face was "smile".
>>
https://files.catbox.moe/3b5fsl.png
test
>>
>>108890192
The visible cameltoe looks weird with this kind of skirt.
>>
>>108890318
A niche fetish that doesn't seem to work well with pleated skirt or something
>>
>>108887648
It's too bad these never have an option to see how artist tags work with male characters. I usually wind up curious what stuff looks like when an artist rarely depicts something.
>>
>>108887059
forge neo rocm install works fine for my 6900xt
the old forge amd fork with zluda worked even better (but doesnt support flux, anima or similar newer models)

training loras will be a PITA though, probably better to save the money for power and rent nvidias
>>
>>108887100
anima, illu is fine for 90% usecases too i guess
>>
810pose
>>
114514
>>
>>108890497
dataset is heavily biased towards female characters anyways
getting males is always a gamble, loras help
>>
>>108890587
Shame anima still has no Character Lora i want
>>
Anima with turbo lora is pretty nice quality for what it is but I wish someone made a no sweat lora since I can't put sweat in negatives at CFG1.
>>
>>108890692
Use negpip https://github.com/pamparamm/ComfyUI-ppm https://github.com/Haoming02/sd-forge-negpip
>>
>>108890714
Oh, neat. I will try.
>>
File: ComfyUI_22431_.png (2.64 MB, 1248x1824)
2.64 MB PNG
>>108890603
Do Strike Witches pls
>>
>>108890725
The model's desire to put sweatdrops wherever there is bare skin is insanely strong but after putting it all the way to (sweaty, sweatdrop:-3.0) for images where I don't want any and it seems to be working. Hooray.
>>
>>108890633
That has kind of always been the case but meta tags genuinely do help. At least on Noob I would put in "(manly:0.9), cleft chin" if the bias was too girlish since it forces the chin to be more square.
>>
>>108890654
Who?
>>
File: 00000-3884046234.png (396 KB, 832x1216)
396 KB PNG
noob here sorry. anyone know why gens with noobai vpred look like shit on forge neo? they're all super saturated and low detail and stuff. i'm doing what the huggingface page says. (I know about anima it works nice but want to try this for some reason.)
masterpiece, best quality, newest, absurdres, highres, safe, 1girl, sakaki (azumanga daioh), azumanga daioh, watercolor (medium), painting (medium), azumanga daioh's school uniform, white background, full body, contrapposto
Negative prompt: nsfw, worst quality, old, early, low quality, lowres, signature, username, logo, bad hands, mutated hands, mammal, anthro, furry, ambiguous form, feral, semi-anthro
Steps: 30, Sampler: Euler a, Schedule type: Automatic, CFG scale: 4.5, Seed: 3884046234, Size: 832x1216, Model: NoobAI-XL-Vpred-v1.0, Model hash: ea349eeae8, RNG: CPU, Emphasis: Original, Noise Schedule: Zero Terminal SNR, Version: neo-2.24
>>
File: ComfyUI_13902_.png (1.16 MB, 1024x1024)
1.16 MB PNG
>>
File: 05686-1924218308.png (2.36 MB, 2487x1200)
2.36 MB PNG
>>108890633
I think anima is pretty good actually. I tried some "@artist, a man is sitting at a round cafe table outdoors and smoking a cigar, upper body" using artists that draw cute/chibi girls and got stuff like this instead of girly boys.
>>
>>
How would you gen a cover page sort of overlay or background to your prompt. Inpaint or just an image editor?

Ex. comic kairakuten covers or Murata style
>>
Anyone here with an AMD card has tried this?
https://github.com/patientx-cfz/comfyui-rocm

It says: it is using AMD's official ROCm and PyTorch packages from their TheRock repo. At the moment, it supports every gpu from RDNA1 (5000 series) to the latest RDNA4 (9000 series) *** some 6000 mobile gpus excluded ***. This is different from Comfyui-Zluda , it doesn't use zluda or other 3rd party apps to launch comfyui , it is also partially portable. It auto-detects your GPU (from the supported ones) and installs the necessary ROCm and PyTorch packages plus triton , sage-attention , flash-attention and bitsandbytes comfyui.***

I'm going to try it now and hope I'm not dissapointed but at the same time I only have a 6650 XT. Can Anima or Illustrious fit?
>>
>>108892217
wont touch comfy, my forge lib is already like 500gb
>>
>>108892004
Just use an image editor if you don't want it to be unreadable garbage. The only thing that gets close (in English) is chatgpt, but it has a very distinct and annoying way of putting text on an image, and probably wouldn't even try if the image is too saucy.
>>
>>108891331
Oversaturation is vpred "overpredicting" latent values (color/brightness) at early steps when CFG is medium to high. You can combat that either by setting cfg way lower (1~2) for early steps then going back to 4.5 when composition steps are done (I don't know which plugins for forge neo do that), or just use shitmixes from /h/ OP, they are really not bad:
>NOOBAI SHITMIXES
102d custom: https://civitai.com/models/1201815?modelVersionId=1491533
291h: https://civitai.com/models/1301670/291h
Also I recommend cfg++ samplers (euler at ~1 cfg, euler A at ~1.5 cfg), those can often solve oversaturation without dropping cfg at early steps even for base vpred.
>>
>>108891331
Oh yeah forgot about this
>and low detail and stuff
Looks about right for watercolor \(medium\), and you will need to upscale your gens with noob, 1MP is never going to be enough with 4-channel VAE unless maybe it's face close-up.
>>
Messy Forge vibe adaption of https://github.com/An1X3R/Anima-Artist-Mixer
https://files.catbox.moe/36u5zq.py
Also added something for specifying artists per layer based on prompt order to maybe steer who affects what.
>>
>>
>>108892859
Cool it works, or at least it's doing something (can turn it on and it will change the generated image and not create mustard gas). Have to do some testing to see if it actually improves mixing.



[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.