/adt/ - Anime Diffusion Thread #4

Welcome to the Anime Diffusion Thread. This is a thread for discussion relating to leveraging generative AI technologies for anime-related content. Images, videos and sound all welcome.
Cloud-based anime generation is also welcome.

Please keep all non-anime gens on >>>/ldg/ >>>/sdg/
Generating cute girls!

Previous thread: >>106087463

FAQ
https://2ch-ai.gitgud.site/wiki/nai

Installation
• NVidia: https://rentry.co/2ch_nai_guide
• AMD: https://2ch-ai.gitgud.site/wiki/nai/install/amd
• Clouds: https://2ch-ai.gitgud.site/wiki/nai/install/clouds

Additional information
• Interfaces: https://2ch-ai.gitgud.site/wiki/nai/interfaces
• Models: https://2ch-ai.gitgud.site/wiki/nai/models
• Prompts: https://2ch-ai.gitgud.site/wiki/nai/prompts
• Upscale: https://2ch-ai.gitgud.site/wiki/nai/upscale
• LoRA: https://2ch-ai.gitgud.site/wiki/nai/lora
• ControlNet: https://2ch-ai.gitgud.site/wiki/nai/controlnet

>Generating Images
https://rentry.org/comfyui_guide_1girl
https://tagexplorer.github.io/

>Generating Videos
Guide: https://rentry.org/wan21kjguide
https://github.com/Wan-Video/Wan2.1

Other
• CivitAI: https://civitai.com
• Link catalog from 4chan users: https://rentry.co/sdg-link

ANCHOR

snowflake general

>>106131482
>Op here, making some Lora and Checkpoint Anchor.

NewAnimeCore is an anime LoRA designed to generate your anime character in a more 3D-like anime fantasy and magic style.

Quick Overview
NewAnimeCore - FLUX [trained on anime images featuring characters]

NewAnimeCore - ILL [high magical energy]

Features & Settings: ILL

Recommended LoRA Strength: [0.7-0.8]

Trigger: NewAnimeCore

Settings: CFG, STEPS, and SAMPLER are defined by the checkpoint and your workflow

Style: 3d like Anime

TIP: use magic words, armor, runes and use pose tags like side view or crouch

Note: Sample images are upscaled + HI-res fix

Ideas for u: silluette, weapons, vehicles, animals,

Features & Settings: FLUX
Recommended LoRA Strength: [0.7-0.8]

Trigger: NewAnimeCore, Anime

Settings: CFG, STEPS, and SAMPLER are defined by the checkpoint and your workflow

Style: 3d like Anime

TIP: Describe the image at the beginning as an anime image

>>106131627
Forgot link
https://civitai.com/models/1628887?modelVersionId=1843612

>>106131627
>anime general
>posts non anime slop

Hello!

>>106131482

weight between [0.8,1], best at 1

trigger words : retro_neo_noir, retro, retro artstyle,

https://civitai.com/models/1836542/retro-neo-noir-illustriousxl

>>106131482
>The shower goon mad me retarded, forgot image

weight between [0.8,1], best at 1

trigger words : retro_neo_noir, retro, retro artstyle,

https://civitai.com/models/1836542/retro-neo-noir-illustriousxl

>>106131699
ok

>>106131699
nice, i like the bottom right one

>>106131482
Introducing IsaBebo Style!
How to Use:
If you're unsure how to prompt this style, check out my showcase images where I keep all the metadata for your reference.

Main Trigger Word:
la3en_sty

https://civitai.com/models/1836701/isabebo-style-or-ayanna-ai

>>106131482
Girls Frontline 2 3d Screenshots style

https://civitai.com/models/1808992/girls-frontline-2-3d-screenshots-style

>>106131482
>Now some checkpoints from today

EQ-Adapted Noobai11 Unet
Using B3 SDXL EQ VAE from here: https://huggingface.co/Anzhc/MS-LC-EQ-D-VR_VAE

This is an EQ adapted unet, that learned to denoise clean latent representations, as opposed to noised default SDXL VAE representation.

This halves loss in training. Yep.

image/png

In my tests it also improves convergence, and makes model more trainable, or so it seems at the very least.
Additionally, this alleviates some of the issues of Noob11 base, like overly bright images.


https://civitai.com/models/1835161/noob11eps-eq

>>106131482
Use this for Anime

masterpiece, best quality, absurdres, highres, very aesthetic, high quality, detailed, insanely detailed, beautiful, very awa, anime screencap
Realistic

(realistic, photorealistic, realistic shading:1.21), (masterpiece, best quality, amazing quality:1.10), very aesthetic, ultra detailed, highly detailed, intricate details, absurdres, (volumetric lighting, HDR:1.10)
Negative

lowres, bad quality, worst quality,bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn,blurry, transparent background,tears, censored, (simple background:1.6), artist name, signature, watermark, holding

https://civitai.com/models/1098213/dvine

>>106131482 (You)
Use this for Anime.

masterpiece, best quality, absurdres, highres, very aesthetic, high quality, detailed, insanely detailed, beautiful, very awa, anime screencap
Realistic

(realistic, photorealistic, realistic shading:1.21), (masterpiece, best quality, amazing quality:1.10), very aesthetic, ultra detailed, highly detailed, intricate details, absurdres, (volumetric lighting, HDR:1.10)
Negative

lowres, bad quality, worst quality,bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn,blurry, transparent background,tears, censored, (simple background:1.6), artist name, signature, watermark, holding

https://civitai.com/models/1098213/dvine

is someone going to make an actual bake or did the schizo destroy your morale? ;p

>>106131482
Ikaholic is perfect for:

Slice-of-life scenes

Comedy and energetic action

Cute & playful character illustrations

Anime-style projects

Whether you are a longtime Ika Musume (Squid Girl) fan or just looking for a fun and lively anime character model, Ikaholic has you covered!

--prompt--
(explicit|general), masterpiece, best quality
--negative--
lowres, worst quality, bad quality, bad anatomy, bad hands, sketch, signature, watermark, logo, border
Recommend settings

Sampling method & Schedule type: Euler a & automatic

Sampling steps: 25~40

CFG Scale: 4~7

VAE: SDXL VAE or xlVAEC_c9

https://civitai.com/models/1833162/ikaholic

>>106131482
>>106131482 (You)
Ikaholic is perfect for:

Slice-of-life scenes

Comedy and energetic action

Cute & playful character illustrations

Anime-style projects

Whether you are a longtime Ika Musume (Squid Girl) fan or just looking for a fun and lively anime character model, Ikaholic has you covered!

--prompt--
(explicit|general), masterpiece, best quality
--negative--
lowres, worst quality, bad quality, bad anatomy, bad hands, sketch, signature, watermark, logo, border
Recommend settings

Sampling method & Schedule type: Euler a & automatic

Sampling steps: 25~40

CFG Scale: 4~7

VAE: SDXL VAE or xlVAEC_c9

https://civitai.com/models/1833162/ikaholic

>>106131482
>Aaaaand last checkpoint for today, yes, someone look interesting and some others look sloppy,

The Oil Light model generates images that mimic a light oil painting with digital enhancements. Its outputs feature soft lighting, detailed textures, and smooth transitions between light and shadow, creating an atmosphere of depth and realism. The model excels at generating portraits and full-body figures, highlighting expressive facial features, soft fabric folds, and complex poses. It is perfect for creating beautiful portraits where elements of traditional oil painting are fused with clear digital rendering, resulting in a refined and polished look.

For the face, it is necessary to use ADetailer

Watch video example - https://civitai.com/images/70459219

For full-length works or long-range plans. The examples use Img2img - Inpaint

Samplers:

Recommended:

Sampler: Euler a

Sampling Steps: 25-45

CFG Scale: 4-7 (For portraits: 4-6)

https://civitai.com/models/1836528/oil-light

>>106131838
schizo

>>106131857
What's your personal motivation to have a dedicated "anime" general instead of using one of the other existing generals?

Any good fart gens?

>>106132270
the loli brap poster is long gone anon

>It starts with
1girl, I don't know why
It doesn't even matter how hard you tag
Keep that in mind, I designed this workflowflow to get what I want in due time

>all I know

VRAM is a valuable thing
Watch it get used as the progress bar swings
Watch it count down to the end of the gen, the GPU ticks steps away

>>106132437
>It's so slop,
Didn't check for bad hands below
Watch the gen go right out the window
Tryna hold on, d-didn't even know
I wasted it all just to watch the waifu go

I kept every lora inside
And even though I tried, it all fell apart
What it meant to me will eventually be a memory of a time when...

>>106132461
I GENNED SO HARD AND GOT SO FAR
BUT IN THE END, IT DOESN’T EVEN MATTER
I HAD TO PROMPT TO LOSE IT AAAAALL
BUT IN THE END, IT DOESN’T EVEN MAAAAATTEEEER

>>106131627
Gonna use this for my negs thanks

>>106132494
I've put my trust in sketch
Pushed my denoise as far as it can go
For all this, there's only one thing you should know
I've put my trust in ControlNet
Pushed the steps as far as they can go
For all this, there's only one thing you should knoooooooow

>>106132537
I GENNED SO HARD AND GOT SO FAAAAAR
BUT IN THE END, IT DOESN'T EVEN MATTER
I HAD TO INPAINT TO LOSE IT ALLLLL
BUT IN THE END, IT DOESN'T EVEN MAAAATTER

>>106132099
/ldg/ is a good thread and I like it. Now with WAN and other cutting edge tech, it's become more oriented toward video and advanced techniques rather than simple SDXL generations.

"A space where anons and creators can discuss or share things like models, LoRAs, or inspire others to create without needing to join Discord servers or lose anonymity by posting on CivitAI."

Or more short:

"An anonymous general for discussing and sharing anime image gens and tools"

What do you think of this idea, Anon?
Yes, I know it will be slow. Yes, maybe only a littlebit of people will be interested. Maybe a thread will last a week, maybe the jannies will delete it before then.

But I want to try creating it, at least until my energy runs out.

>>106131482

MY WORKFLOW
For those who love the V-pred look but hate the mangled hands from Noob's V-pred
Been messing around because I'm jealous of the stuff people make with V-prediction models, but I'm a noob and my results always have garbage hands.

This is my desperate fix.
Models:
>NoobAI-XL V-Pred-1.0-Version: https://civitai.com/models/833294?modelVersionId=1190596

>One obsession 1.5_NoobAi: https://civitai.com/models/1318945?modelVersionId=1972020

>LoRA:
NAI vpred fix: https://civitai.com/models/99619?modelVersionId=1965505

>>106132858
The basic idea is to use a two-model workflow. This lets you borrow the advantages of V-prediction (better color gamut, lighting, composition) while using a second pass (Hires. fix or img2img) to fix the awful hands that the Noob V-pred model tends to generate.
However, this method comes with a few things to watch out for.
The Noob V-pred model is super sensitive. Don't go too crazy with your prompt weights and CFG scale.
If your colors are getting too saturated, adjust the Hires. fix CFG scale. I recommend a range of 0.5-0.7. A lower value will give you more saturation and bring it closer to the original V-pred output.
For the Hires. fix denoising strength, you'll need to experiment since it depends on your second-pass model. I found that 0.45 works well with one15. (I tested a few different models for the second pass, and one15 was the most stable one for me).
Because you're blending two models, your usual artist tag strings might not work as expected. You'll probably need to tweak the weights. Here are a couple of my personal artist blends that work well with this Noob V-pred + one15 setup.

>>106131825
its the real bake now post more animegirls

>>106132871
Personal artist mix 1

hiten \(hitenkei\),(quasarcake:0.85),wanke,(maccha \(mochancc\):0.8),(mochizuki kei:0.4),(ciloranko:0.7),(wlop:0.8),(nekojira:0.7),

>>106132913
Personak artist mix 2
quasarcake,hiten \(hitenkei\),mochizuki kei,(ciloranko:0.9),wanke,(ask \(askzy\):0.8),(wlop:0.5),(nekojira:0.9),

Remember to tip the artists you use in your prompts!

>>106132955
Nah, let the checkpointmakers do that.

>>106132892
https://files.catbox.moe/1kmh4z.png

>>106131439 (OP)
>golde gate over sunfloers
Wait a minute, wasn't that one from the Landscape Diffusion General Thread?

Pos:
{masterpiece, best quality, 0.9::artist:leonmandala::, 1.05::artist:mellow_kaede::,0.85::artist:mitsu_(mitsu_art)::, 0.95::artist:ergouzi_echo::, 1.25::artist:leon_decarabia::,:0.75:: artist: ginhaha::,year 2024,1girl, zombie girl ,gray skin,pink hair, messy hair, oversized shirt, nude,suture mark,fangs ,Cemetery,disgust for garlic,garlic in front, very aesthetic, masterpiece,}

Neg:
{nsfw, lowres, artistic error, film grain, scan artifacts, worst quality, bad quality, jpeg artifacts, very displeasing, chromatic aberration, dithering, halftone, screentone, multiple views, logo, too many watermarks, negative space, blank page, lowres, worst quality, low quality, normal quality, jpeg artifacts, compression artifacts, blurry, watermark, signature, username, text, logo, artist name, censored,bad anatomy, deformed, malformed, mutated, ugly, disfigured,extra limb, merged body parts, long neck, long body, (bad hands:1.3), (poorly drawn hands:1.3), (deformed hands:1.3), (malformed hands:1.3), (mutated hands:1.3), (distorted hands:1.2), (bad fingers:1.3), (poorly drawn fingers:1.3), (deformed fingers:1.3), (malformed fingers:1.3), (mutated fingers:1.3), (distorted fingers:1.2), extra fingers, missing fingers, fewer digits, extra digit, fused fingers, too many fingers, (poorly drawn face:1.2), (deformed face:1.2), (malformed face:1.2), (mutated face:1.2), multiple eyebrow, poorly drawn feet, deformed feet, malformed feet, multiple people, crowd, group, extra characters, extra person, simple background, plain background, white background, empty background, black and white, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, signature, watermark, username, blurry, artist name, no hair}

Resolution: 832x1216
Seed: 3750174355
Steps: 28
Sampler: k_euler_ancestral (karras)
Prompt Guidance: 5
Prompt Guidance Rescale: 0.3
Undesired Content Strength: 0
Model: NovelAI 4.5 full

>>106133213
>NovelAI 4.5 full

>>106132858
Nice gen, catbox?

>>106131439 (OP)
>russian idiot desperately wanted to bake
>too stupid to realize nobody wants his russian language garbage in the OP

Can we change the OP pasta to be less trash?>>106132858
>>106132871
Very pretty. I prefer to use advokat's own checkpoints which have the fix baked in. I have found Obsession to be weirdly resistant to artist tags compared to other NAI models. Although as seen in picrel even with that fix a detailer could still help further.

Sar please explain to me how to generate the anime with the bobs and vagene
Sar real women do scare me

Saar please do the needful

>>106135064
Hnnnnnggg nice tamamo

>>106135064
>advokat's own checkpoints
Name?

>>106133213
>millions of tags and weights of artists
>a shakespearean paragraph of negatives
>the most sloppy of images

>>106135819
That particular gen used 2DN-anime but using artist tags to mimic his earlier nijiNoob, which is one of my favorite checkpoints of all time (although I am not the guy gushing about it in the civ comments). https://files.catbox.moe/5ijqy5.png

animagine-xl-4.0-opt
Artists : (da mao banlangen:0.5),(nardack:0.5),(ningen mame:1.2),(hiten \(hitenkei\):0.3),(ask \(askzy\):1.1),(mitsu \(mitsu art\):1.0),

>>106135892
Thanks, I'm working now. Will try it later!

Has there been anything good since the last versions of noob? All i see here are plastic shitmixes thus far.

>106135967
Not artistically, but in terms of prompt adherence, let's assume that "yes" there has been some change, but Noob is still king, and worst of all, it was made on December 22 2024.

>>106135985
Noob as an architecture is still king but the model itself has long been surpassed by retrains using vpred and newer datasets don't pretend there's been zero progress like the dry spell from aug 23-dec 24.

>>106131857
>The Oil Light model generates images that mimic a light oil painting with digital enhancements. Its outputs feature soft lighting, detailed textures, and smooth transitions between light and shadow, creating an atmosphere of depth and realism.

I find it funny how this checkpoint tries to justify its SD1.5 slop aesthetic.

>>106135997
>but the model itself has long been surpassed by retrains using vpred
Which ones?

>>106135997
I'm newfag at this, do you know the story of Noob or how it was created? Its a corpo? Its an autist? A group of artist?

What's the best anime model for gooning?

>>106136512
SAAR

very cool

>>106136512
In my experience, every anime model it's a gooning model itself. Go for the mainmlstream ones like Hassaku or JANKU , Noob itself, EIH or SIH for more hardcore uncensored stuff

>>106136795
First video I've seen that actually kinda looks like anime and not rotoscoped 3D.

>>106136795
KINO

I want to make a Cloverworks style Lora, so I'm going to look for some of their movies or series, the most colorful and beautiful ones and put them.
Do I need to tag the images for the training style Lora?
Pic related is obviously not mine, it's from Google, sorry.

Why is this pathetic general still a thing?

>>106136896
They need a place to goon to cartoon children
Why they wouldn't use literally any other board on 4chan is beyond me

>>106137821
off topic
this is not anime and thus my eyes are literally bleeding

ew what is this, another general? no thanks.

>>106138122
Slop
Thank god this containment general exists

>>106136874
You have to tag everything escept the style!

>>106137821
Nice! Are you the dalee anon? You did the same with the same dalle? Very impresive? Pr did ypu use dalle and then import ethe image for a hires or adetailer?

>>106139693
kek

>>106139693
I don't use any detailers, but sometimes I use NAI to inpaint over mistakes.

What happened with the Gemma text encode they did update or something?
I remember I was testing it gemma vs clip and ironically spatial reasoning of gemma is literal dogshit, prompted "gun in right hand, sword in left" and gemma randomizes it every time and ironically clip followed the directions

>>106131439 (OP)
Can we do unfiltered video yet?

>>106140472
Nyoooo

>>106140701
>six toes
AAAAAAAAAAAIIIIIIIIIIIEEEEEEEEE FIX THAT!

>>106131439 (OP)
Widescreen bias

>>106140701
This whole image is a mess. 0 effort was involved here.

>>106133290
Prompt: fur dataset, nsfw,A gentle and beautiful wife is waiting for you in the living room, the afternoon sun streams through the window, casting long shadows. 1girl, solo, mature female, milf, wife, {gentle smile}, blushing, long messy hair, {{naked apron}}, apron only, nude, large breasts, wide hips, kneeling on the floor in front of the sofa, looking up at viewer, seductive pose, parted lips, in a cozy living room, cinematic lighting, dramatic shadows, sunbeam, from viewer's perspective, shallow depth of field, detailed background.
, very aesthetic, masterpiece, no text
Undesired Content: lowres, artistic error, film grain, scan artifacts, worst quality, bad quality, jpeg artifacts, very displeasing, chromatic aberration, dithering, halftone, screentone, multiple views, logo, too many watermarks, negative space, blank page, worst quality,low quality,artist collaboration, bad anatomy,extra fingers,extra legs, missing legs, missing fingers, mutation, text, watermark, low resolution,furry,anthro, animal ears, cat ears, dog ears, fox ears, tail, cat tail, fox tail, paws, animal on head, kemonomimi, animal girl, monster girl, hybrid, animal features
Character 1 Prompt: girl, 5::best quality, absurdres, very aesthetic, detailed, masterpiece::,1.2::misaka_12003-gou ::, 0.8::dino_(dinoartforame), wanke, liduke ::, year 2025, realistic, 4k, 1.5::photorealistic::, 1.5::photo(medium)::, 2::oily shiny skin ::,
Character 1 UC:
Resolution: 832x1216
Seed: 2777038069
Steps: 28
Sampler: k_euler_ancestral (karras)
Prompt Guidance: 4.9
Prompt Guidance Rescale: 0
Undesired Content Strength: 0

Whats the problem, anonie? This is local and cloud difussion, have a hag and fuck off

>>106140701
JESUS that's even worse

>>106141083
agreed, some people don't even try

Help, I'm trying to learn ComfyUI. I found this on CivitAI. Is there an approved user there I can learn from? This is hell!

>>106141492
Read the rentry guide: https://rentry.org/comfyui_guide_1girl

Contains everything you need to know explained nice and simply.

>>106141492
stop downloading bloated and overly complicated workflows that dont explain anything whats going on. youll never learn that way

>>106142207
that's not ai retard

>>106131482
I'm hijacking this thread. Now, it's /adt/ - algebraic data types.

>>106142347
very based
what should i know about algebraic data types before posting here?

>>106141492
>he comfied
LEL

>>106142314
>>106142347
I sense a new king of slop... like the kind of slop from Goku in /ldg/. That's not SDXL, what is it? Flux?

>>106141492
use the fucking built in txt2img sample workflow the ksampler node is just as easy as using a1111, schizo workflows should only ever be needed for exotic conditioning or bleeding edge models.

>>106141492
Anon, I have the solution for you:
Go to llmarena.ai.
The smartest model currently is Claude Opus4 Thinking, and the most patient assistant is Gemini 2.5pro.
Export that workflow to .json, open it with Notepad.
Ctrl C+CtrlV in the chat with the AI.
Explain to the AI that you are new to Comfy and ask it to explain what each node does and how they relate to each other.
It will take time, but you will end up learning. I recommend doing this kind of thing in your free time, for example on a bus or train.
Meanwhile, when you are at home, use Forge, ReForge, or ForgeClassic.
When you feel like you've learned everything there is to know about Comfy, jump over to Comfy and you can use it without any problems.

>mandatory pic to accompany the thread topic

>>106142556
>>106141492
https://windwp.github.io/comfyui-ghpages/comfyui/

Also, on this website, you can also visualize workflows online anywhere and interact with them without having to run Comfy.

can SOMEONE do the needful pls SAAR

>>106142556
every new gpt image gen that i see convinces me more that gpt is actually the worst

>>106142719
Do you want to talk about something anon?

>>106142732
w-what do you mean

>>106141492
Hohoho, good stuff Anon, that fuels my rage.

this thread sucks

is this AI with the stable diffusion?

>>106140701
Will we ever be able to?

>>106142370
If you’ve worked in any language that supports ADTs, that’s already enough.

>>106142493
It's NAI. Sprite and bg are generated separately to better mimic the VN-like style. Inpainting is used on the base sprite to give it different emotions. UI is shopped on top.

>>106143990
NovelAI or Noob? It's not very pretentious with the color and texture to be Noob

>>106142940
>>106140701
>>106136795
Why are childs so erotic?

https://files.catbox.moe/pf9t22.png

https://files.catbox.moe/p8c095.png

https://files.catbox.moe/mrcni8.png

>>106132892
okie dokie :3

>>106142207
been there haha
keep ya head up
>>106144718
cute!
>>106140077
s a v e d

>>106145137
why the hell can't they keep their mouth SHUT

https://litter.catbox.moe/6ofqte81xeo95wiy.png

This >>106140077
Was made wit this? >>106131699
I hate today anime designs. They are so gay with their big faces, big eyes, and child proportions. I dont like it, but retro anime women are the best.

>>106145260
Cool. Saw that neon eyes style Lora on Discord or Civit, can't remember.

>>106145190
its something he also cant do
BeAHAHAGaH

bored
>>106144601
>>106145260
nobody asked

guys i misclicked and landed in another general where i was forced to see a REALISTIC woman D:
i'm so happy this general here exists...

>>106145284
nah that's imagen-3 mate
>image fagmiliar

>>106144288
NovelAI

https://developer.nvidia.com/cuda-downloads
https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html

UPDATE YOUR CUDA 13.0 TECHNOLO/g/Y

>>106146997
That was a close one.

>>106140701
When can we?

>>106148115
Ew

>>106146997
ok

>>106148115
Kino, are you paying or using it via proxy? I love the color and dynamic. My proxy provider died a month or so and now im without NovelAI :(
If NovelAI could support for example Loras or API nodes to use it as a workflow via Comfy it could be Kino

will nai ever jump in the video gen game or is that shit too computationally expensive to charge a reasonable price for ?

>>106149274
I could see them adding it, if it costs 100 anals per video.

>>106149274
they havn't updates their text model for a while, i think they are just concentrating on image, which is fine,
its best to do what you're good at, though lewd anime video gen is an open market...

>>106149431
prompt?

>>106149445
The girl turns toward the camera while holding the tray with one hand and lifts her skirt up with her other hand revealing her white panties.

I've posted this before in ldg but seeing how it is my proudest (and dare I say finest) work to date, I'm willing to grant you all the privilege of seeing it again.

>>106148129
it's ass and doesn't work with llms. do not downgrade

>>106148129
why should i?

>>106149756
because cold stone said so

>>106149072
Paying. I'm so tired of sdxl that I'm ready to pay to escape it. (Chroma and Lumina are still too underbaked to use them)

>>106150057
I'm basically paying so I can gen at work

have this as well

>>106149515
Thank you. Can you make her fart?

>>106135875
this is so cool, do you have more?

Can NovelAI generate this kind of aesthic?
https://files.catbox.moe/vc773g.png

https://files.catbox.moe/rygnxe.png

>>106152513
>those workflows
kys

I inherited an extra 19-inch monitor. I currently have a 24-inch monitor. How can I use this new monitor for genning?

>>106152513
>ultra slop just smear my shit up with vaseline nigga
yeah i think it could that

>>106152513
>just fuck my shit up jeetmix
I assume so, if you overload it with bad quality tags in the positives

>>106151953
Style, or the character?
The style is just Akihiko Yoshida with some medium tags.

>>106136874
Based idea, a Cloverworks LoRA would be kino.
Yes, you have to tag. The trick is you tag everything except the style. So tag stuff like 1girl, blue eyes, school uniform, city background, etc. The model learns the "style" from what's left over.
Post it here when you're done, I wanna see if it's slop or not.

I love hags so much its unreal

What's the most advanced way of prompting two distinct subjects interacting together nowadays? I've heard about comfy couple, is forge couple any different? It doesn't tend to produce coherent results all the time still.

>>106152862
The two methods are shit, there are Loras, regional prompting, but you will have a hard time. The best is to already choose an existing image and ControlNet it. Create it from scratch with SDXL its really hard

>>106153032
Even chroma shits itself with two subjects

>>106152648 >>106152694 >>106152777
Schizo prompting and workflows is the way of doing things. I'm using a SeaArt japan gooner shizo prompting and settings in a 2024 NTRmix checkpoint and im having the best results of my life

>>106153072
Chroma is losted with this Qwen and WAN 2.2 things, besides realistic NSFW is no point in that model

>>106153101
Neither of those have chroma's knowledge of art styles

>>106153112
for now, WAN 2.2 has two weeks old and Qwen has 1 day old

>>106153072
it can't even draw a cat and a rat in the same picture without shitting itself

>>106153126
ok chang

>>106150920
she just did

>>106152513
Something like that?

>>106154006
nice!

>>106154020
I'm cheating a bit here, because I'm using vibe transfer (NAI's IP adapter). But it should be possible to get close enough just with text.

>>106153854
cute

>>106154154
That's looks nice.

anyone tried qwen for animeme?

Any opinion on the AYS scheduler? Just tried it today and it felt solid.

>>106154403
link?

>>106154638
https://research.nvidia.com/labs/toronto-ai/AlignYourSteps/

I'm using it from ComfyUI's KSampler node from InspirePack (or the Efficiency custom nodes).

>>106131482
>OP here! Great! Thanks for taking part and for SHARING workflows and artist concepts! To keep the discussion going and focused on art, I'll keep sharing checkpoints!

>ANIME CHECKPOINTS NEWS 4-5 AUG

Illustrious FLAT 2D from HaDeS v2.0.1

Version 2.0.1 brings a significant leap forward in stability, accuracy, and visual output. Built on the refined Illustrious from HaDeS foundation, this update focuses on delivering cleaner, sharper, and more expressive 2D generations while keeping the signature flat aesthetic intact.

What’s New in v2.0.1

Enhanced Image Quality – Sharper details, smoother shading, and overall improved visual consistency across all outputs.

Eye Rendering Upgrade – Eyes are now more vivid, aligned, and anatomically accurate, greatly enhancing character expressiveness.

Improved Poses, Gestures & Object Interaction – Characters move, emote, and handle objects more naturally with fewer distortions.

Reduced Background Artifacts – Simple and plain backgrounds are now much cleaner, minimizing unwanted noise and glitches.

New Poses Added – Expanded variety of dynamic and natural poses for more creative flexibility.

Optimized Prompt Responsiveness – Better adherence to complex prompts with more stable results.

Generation Details

Resolution: 1152x1536 native

Sampler: Euler A — consistent, stable, fast

Steps: 25+ recommended

CFG: 5–7 for optimal balance

Prompting: No keywords required — just describe what you want.
Of course, using the classic Illustrious XL base keywords and negative ones will significantly enhance the image quality.

This version is designed for artists and creators who need flat-styled 2D illustrations with high reliability and smooth generation flow. Whether for character portraits, scenes, or stylized compositions, v2.0.1 provides more control and polished results than ever before.


https://civitai.com/models/1691607?modelVersionId=2078598

>>106131482
Flat Justice is a 2d/vector trace (danbooru wiki) anime checkpoint based on noobai v-pred checkpoints and some nice loras.

After using Hexbloom - Ionsyx Revised by Arctenox I started enjoying the 2d style and since I've been generating images for a long time but never made any checkpoints or lora's I figured why not give it a try and make my own.
Recommended settings

Positive prompt:

masterpiece, best quality, very aesthetic, absurdres, newest,

Negative prompt:

worst quality, low quality, bad quality, bad, watermark, bad hands, extra digits, artistic error, old, worst aesthetic, loli, censored, halo,

Sampler/Scheduler:
DPM++ 2M SGM Uniform
or
Euler a Beta

Steps: 30 (40 is even better)
CFG: 4-5~

https://civitai.com/models/1724753

>>106131482
>MysteryStew

Merge of 4 deliberately chosen checkpoints, just intentionally not remembered.

>>106131482
Anime style Comic style

A fresh and innovative atmosphere, like Showa-era anime revived in the Reiwa era.

https://civitai.com/models/1835634/rustoriousil

>>106131482

A model merged for my own LORA.

I used WAI-NSFW-illustrious, Hassaku XL (Illustrious), and TanemoMix and mixed them with my own model.

https://civitai.com/models/1837272/hawawamix-il

>>106131482
Merge & Training Details

The backbone of any of our mixes are built upon numerous LoRAs, many of which were trained in-house by Ktiseos Nyx. We're always experimenting and creating! Due to the wonderfully rapid pace of Ktiseos Nyx's LoRA development, we confess that sometimes the exact lineage of merged checkpoints becomes a delightful mystery. We believe in training as much as we can ourselves, but the spirit of community and shared creation is vital!

Custom Lora Moosh: does not include the new ringo style, but does have a lot of our illustrious loras.

AnthromixXL

ArthemyComics

HomoSimilie NAI XL EPS

Comic Fictional? (I forgot the name of it)

Ultra Comix (Ours)

NoNoSquareMix

Recommended Settings

To get the best out of ANY of our models, try these settings:

Samplers: Euler or Euler Ancestral

Scheduler: Karras

Negative Prompt: Generally not needed! The model is designed to perform well without one, but feel free to experiment based on your preferences.

HiRes Fix:

Steps: 15

Denoise: 0.3

Upscale: 2x

Upscalers: Rex, 4x_Remacri, or your personal favorite.

ADetailer: Use sparingly. Due to potential VAE differences, it might occasionally produce grey squares.

Embeddings: We've used some of our own custom embeddings to showcase character types in our examples, but no stylistic embeddings were necessary to achieve the core look.

LoRA Compatibility: Chances are good that most LoRAs will work with this. However, be aware that while they will function, the result might not perfectly align with your expectations if the LoRA heavily influences style. It's a "What You See Is What You Get" situation with the base model's strong aesthetic – it'll likely produce a good image, just perhaps not the one you had in mind with a very specific LoRA

https://civitai.com/models/1837227/isekai-mix-illusxl

>>106131482
This checkpoint was created by merging anime-style models with realistic-style models in an attempt to achieve better lighting and detail rendering. The fusion seeks to combine the best of both worlds—anime expressiveness and realism detail fidelity.


### Recommended Positive Tags:

- masterpiece, best quality

### Note:

- Tags like realistic, photorealistic still output more anime-like results than truly photorealistic ones.

## Recommended Generation Settings

- CFG Scale: 3.5

- Steps: 28

- Sampler: Euler a

- Clip Skip: 2

https://civitai.com/models/1839777/pxsynth-10

>>106131482
Prompt

masterpiece, best quality, very aesthetic, 1girl

Negative prompt

worst quality, bad quality

License
https://civitai.com/models/1522819/lemonsaltmix

https://civitai.com/models/1522819/lemonsaltmix

>>106131482
Lora for netalumina!


触发词:<style>An image in the style of Blue Archive CG</style>
Trigger Word: <style>An image in the style of Blue Archive CG</style>

It is recommended to use English tag form prompt words

Using English tags for prompts is recommended.

In some cases, v1 may not perform better than v0.5.

v1 is not always an improvement over v0.5

https://civitai.com/models/1841399/blue-archive-cg-stylecgnetalumina2

>>106154774
it's supposed to be used with a special Lora and a scheduler. it's like a better lcm

>>106131482
>This checkpoint is my favorite. It's still in "early access mode", but I recommend you to be notified when early access ends. I really like it, the prompts adherence, the consistency, the creativity, everything!


Prefect illustrious XL 3.0

https://civitai.com/models/1224788/prefect-illustrious-xl

>>106149329
Isn’t it crazy expensive and impossible to compete vs the big boys even though they refuse to fill the nsfw niche?

>>106131439 (OP)
Geez another AI art thread split?

>>106149274
>too computationally expensive
This one, they already said they can't do video shit with how expensive this is, and the price wouldn't be reasonable at all, like most good video gen
iirc they said v4 to v5 will be a bigger gap than v3 to v4, so they are focusing on that

>>106155833
It won’t stop until every board has the same set of ten ai threads.

>>106155955
or they capitulate to that catgirl schizo and add an ai board

;3

>>106155833
more flies with honey than vinegar anon;3

>>106155833
Here micro stuff, artist tags and vibes, /ldg/big picture, models, workflows and all that.

>>106156083
>>106156096
this is /aco/ shit, not anime

>>106156143
post more

>>106156748
Clownpiece specifically?

>>106156869
got more of the mecha ones?

is there any place to test Qwen online SFW anime?

>>106156962
can't use the spaces on huggingface?

>>106156950

>>106156973
didn't know, will try! thanks!

>>106156975
thanks man i like those

>>106156973
oh its generating! great! thanks nonie! I can use it forever? its free?
I LOVE IT!!!! Beter image composition than the corpo models!

>>106157021
>its free
i'd expect there to be a token limit

Now that the dust has settled, how do we feel about Qwen image? Don't reply if you don't have at least 16gb of vram because you aren't physically capable of giving an objective opinion.

>>106157038
AAAAIIIIIE one gen per day!
>>106157044
I have 12VRAM its over for me?

>>106157056

>>106157021
>I can use it forever? its free?
it's free though not sure how long you can use it consistently for. i've only used spaces a couple of times to test something so i don't know much about them myself.

>>106157009
>>106156975
thanks for posting more man. i liked those and the girls with the guns/swords.

>>106157069
KEK

ive only got 4gigs of vram...

Anyone else have vram rainy day fund just to make sure they never become a vramlet?

I have a second 3090 on standby and $4k in cash saved in a brown envelope marked "vramlet fund" should both of those fail. I could not live as a vramlet.

>>106157106
you got great use of it tho imo i like your gens bro.

>>106157056
throw me some prompts and pick a res from here. i can gen some as i'm watching a movie and don't really have anything to gen atm.
"1:1": (1328, 1328),
"16:9": (1664, 928),
"9:16": (928, 1664),
"4:3": (1472, 1104),
"3:4": (1104, 1472),
"3:2": (1584, 1056),
"2:3": (1056, 1584),

Bootleg walter white

Euler/Simple? Nono. Deis/Beta!

The anime and stylized outputs from Qwen looks 10 times better than the realistic outputs which look grainy and low detail. The anime outputs are sharp and sleek as fuck. Makes me think the character LoRAs are going to be amazing.

>>106157021
>>106157176
oh thanks anon but im going out soon so i dont really need you to do it for me, thank you! :D

Ok I love Qwen for image composition, I will add it to my schizo workflow. At least it gives more dynamism to the image.

>>106157269
Tomorrow I will try with 0.54 control weight and lineart maybe there dummy SDXL will recongize the rose and the backround doesnt repeat the sword pattern!

This one was other that I did with FLUX Kontext, and inbetween Google 2.5pro for image taging this things I wont never do it with SDXL alone

these mechs are quite silly

>>106157307
lol good stuff

'morning

so many young girls...where are the hags?

>>106157021
I’m going to need a tard friendly guide

>>106157668
Google:
Qwen image model online hugging face
Enter the first result
At the right side of the page select one of the deplyment options aviable ZERO something its the one i choosed

>>106157269
>>106157802
It's Crusade time? I think it's Crusade time

for the next bake can you please make a proper collage or just not do that at all?
horrible cropping ;c

>>106157849
Nice gen. The style reminds me of Qwen, but the annoying sword patterns remind me of SDXL.

>>106157620
Morning, nonie :3

>>106158651
go on...

>>106157650
have some. I really like the neo noir dataset

>>106159473
How did you perform this sorcery

>>106157106
beahagahahah

>>106159973
you need to get your eyes checked because that is shit

>>106159973
wan2.2 inside comfyUI

>>106159473
>>106160046
what is a good workflow? Also do you have any prompt tips?

>>106160074
For 2.2 I keep the prompt fairly simple, certainly much simpler than 2.1. Just describe what you want to see and work from there.
I always use I2V, never T2V.
Never use loras because wan loras are rubbish. The 2.2 base model is capable of generating 90% of what I want.

>>106158203
>>106157522
>>106157307

lol, love the touhou mecha

Which model did you use for those touhou characters?

I'm still using noobai for my Touhou gens, but the 'booru prompt' is a bit unnerving.

I'm trying qwen, btw , it's really impressive,prompt adherence is one of the best i've ever seen, But has really few base characters ,Like Hatsune Miku. Great letdown.

>>106157973
OP here.
When should I make a new post?
I now have Replies: 235/Images: 125/page: 1 (or 7, I think there's a bug)
I made the grid and letters in MSpaint intentionaly for KINO.
Also I like that it has a title in the OP's image because that way people browsing the catalog will find out about this new thread.

>>106160700
Sorry 265 replies

>>106160700
Are you still scared of women?
Also normally you bake when the thread is on page 10, not earlier

>>106160700
fuck off with your shitty russian garbage in the OP, nobody is reading that trash

>>106161022
Anime website,If there are AI threads, then logically, there should also be an anime AI general on an anime website.
>References:
https://desuarchive.org/_/search/boards/g.desu.meta/text/anime%20website/
>>106161022
Okay
>>106161053
I understand but I'm not Russian either.
Which link should I use? I have an extension that automatically translates my website, so I don't have any problems, but I understand a native english speaker POV, who was led to believe that the world revolves around him.

>>106161022
you are still hung up on people abandoning your thread eh schizo?

Qwen image is on LLMarena for free!

https://lmarena.ai/?mode=direct&chat-modality=image

->Direct Chat, ->Qwen Image

I am currently parsing 26tb ndjson files from hdds to get image links kek

>>106161677
?

Why do NoobAI model always crank up the saturation in the output, it's so annoying.

>>106162003
There are some stabilizers on Civit, did you try it? Also I think is part of the "Noob Style"

>>106162070
Thanks anon i'll look into it

>>106161677
>>106161730
It sounds like he's brute force/rainbow table-ing URLs to get image links from somewhere.

>>106160664
I'm using cocoa illustrious 7.0 which is a merge of illustrious, noob and some other stuff that I dont remember since it was taken off civitai

>>106157269
mmm

Mmmm apparently there is no dataset that a leg is not visible because the person is riding an animal and the leg remain hidden...

>>106163792
could try straddling in your prompt instead of riding or sitting

>>106164176
Thank you, yes, It could be! I would have to try it. One thing I've noticed is the bleeding of color patterns or shapes patterns in images, how they repeat, which doesn't happen with Qwen or any non SDXL model. Even so, I respect the boy because it is from 2023.

>>106164475
for me its symmetrical backgrounds like the character is standing in a tunnel. I actually put symmetry in my negative prompt which seems to help a bit

>>106164527
Hum, d-do you have a p-picture of her tiny doll feet?

guys what kind taags are u using with illustration models
e621 or danbooru

>>106165172
nice

>>106165013
I use danbooru, e621 are a bit more specific of the checkpoint

>>106164995

>>106131439 (OP)

>>106160700
if you NEED HELP with the collage
i can help you (sometimes, maybe this time)

however,
i would advise to spend a few extra minutes baking instead of rushing things <3

>>106167867
canva
or photocollage . com

cfg slightly too high ;c

>>106166264
Hnnnnnnnnnng

This thread might just get me back into making gens again.

>>106168187
good.

>>106168187
Thanks, feel free to share, this space is made for that.
>>106167867
I asked gpt. Now I understand better. Now that we've surpassed 300 replies, I'll wait until page 9 or 10. The collage in MSPaint is an artistic preference of mine. I want to make some anime foxgirls in MS Paint that also hold a banner or something like that...

>>106168395
you can bake at any time anon
but take care with other peoples gens
they spent time creating and dont want their heads chopped off haha<3

>>106168431
Hmm, good point, I wasn't considering that perspective. Thanks.

>>106168572
im gonna try to animate this ;3

>>106168806
>>106168572
;c failgen

>>106168970
I mean .. it's better than I would've expected.

Mm, baking slowly, it's hard not to cut back on the gens, sorry :(

>>106169330
take ur time sweaty ;3
>>106169283
i'll retry tomorrow

Imposible to bake :(
Im on my phone its a toaster, everything slow, ads everywhere :(
https://files.catbox.moe/ilzshv.png
https://files.catbox.moe/vz5a01.jpg

>>106169763
For those who want to bake, please put thread #5. :(

>>106169789
which one are you wanting to bake with?

New thread:

>>106171902
>>106171902
>>106171902