This model added flat anime elements to "Diving-Illustrious Anime v11".
NSFW is OK.
I hope you enjoy creating it.
（Sales are prohibited. Please use for personal enjoyment）
If you like, please try the another anime model as well.
https://civarchive.com/models/1170176/diving-illustrious-anime
Recommend Settings
Clip Skip：1 or 2
VAE：not necessary
Sampling method：Euler a ( or DPM++ 2M or DPM++ SDE )
Schedule type：Karras
Size：896 × 1152 (You can set it to your liking here)
Step：25
CFG : 4〜7
ADetailer：face_yolov8n.pt
Recommend Prompt
(anime coloring, anime screencap:1.5)v1.0
Added flat anime elements to "Diving-Illustrious Anime v11".
v2.0
Add contrast.
v3.0
More contrast.
v4.0
It has been re-adjusted based on v2.0. The generated texture is similar, but the finer details have been improved, such as the beautiful depiction of hands.
v5.0 and later are experimental models. Please use them according to your preferences.
v5.0
This model is adds more flat animation elements.
v6.3
adjusted it to have a flatter texture.
v7.0
This model has been updated in v6.3 with the addition of "line drawing" and "maturity" elements.
v8.0
This model is version 7.0 with added elements of hentai anime. The contrast has been reduced, the lines are thinner, and some eye abnormalities have been slightly improved.
Description
FAQ
Comments (14)
This model is seriously awesome
It seems highly dependent on the number of steps. I think fine-tuning will be necessary.
This has quickly become my new favorite anime-style checkpoint.
style has been changed，i dont like it
i love the style, but i can't avoid the mature aspect that trigger me a bit :(
From a flat style, the model has turned into another shiny 2.5 model, like all the others, and in general, it’s too much of a maturity in my opinion
Easily solved by using (anime coloring) and (anime screencap) in the prompt. You can add other 2d prompts as well.
@VunKorb
That's because you don't know how to use tags.
Just (anime coloring:1.4) and be done with it.
with all due respect. thats not how it works. In a 2d model you can technically make 3D art but it will not look as good as a model made in a 3d model. Vice versa with 2d. There is no such model that can do both 2D,2.5D and 3D equally good. To have more versatility you have to let go quality. I noticed people saying things like this everywhere and it is just very counter productive, you might be fine with slightly less quality if it means more versatility but thats not the case for everyone.
I assure you, I know how tags work. After experimenting with different tag weights, I discovered a simple principle: different models behave differently, but in most cases, the more you increase the tag weight, the worse the photo quality becomes, and the colors become terrible. In the case of this model, the previous version immediately produced flatter colors, while the new version has a kind of 2.5D style, like most others, such as WAI-Illustrious.
Overall, I've found that you can achieve balance using certain types of embeddings. For example, I used "lazyneg" and achieved a certain balance of quality and style.
It's just straight up bs. What’s the point of the model if you can’t even use it out of the box without cranking the hell out of token weights like anime_coloring? Literally any model will give you a flat style if you just jam in (anime_coloring:1.4).
So far the only ones that actually pump out a legit flat base style are Rin’s like his Flanime. It’s a bummer that even though the style is spot on it’s kind of mid on its own. But once you slap a lora on it the whole thing just hits different
V2:Color is a bit too muted. Otherwise my favorite.
V4: Background tend to have a bit too much noise. Some issues with Toes and Fingers.
V5: Still some bg issues. Toe and fingers are fixed.Causes face shape to change fsr.And makes a lot of mistakes in detailed clothes or character LORA. Does this have some mature version mixed in?
Overall none of these are full flat still in comparison to some other models i tried. even with more emphasis on anime coloring and 3d at negatives.I like V2 the most overall, would like to see an even more flat anime style focused version of V2.
Mind sharing what models have you used that are flatter than this please?
I've got the same problem as you, I struggle to find reliable and well made purely 2d flat models. all models seem to be 2.5d these days.
for V5 Prompt adherence is the best I have seen upto about 380 tokens it doesn't drop tokens yeh some will become somewhat muted but there still there and I would suggest using an embedding like lazyneg, and asking an AI to list you a bunch of artists with your preferred style that are listed on danbooru just cycle through a few of them and this model becomes great, personally I use "(miv4t), so-bin," do need to go in and add "shiny skin, shiny hair," into the negative prompt.
All that turned this from me just being impressed with the prompt adherence to my new first pick model for anime generations.
my full negative boiler plate is "embedding:lazyneg, (bad proportions, bad definition, misshapen, smudges, forced perspective, shiny skin, shiny hair, bones) smooth skin, excessively shiny, bad symmetry, bad texture, bad colour,
long legs, long arms, picture_\(object\), painting_\(object\), "
if you're not rendering the base image at more than 1MP you can throw out a lot of the negative and mostly rely on the embedding.
and the positive concat, but bare in mind that I use Detail Demon and CFG Zero Star and my workflow treats the concat differently to the main prompt so I can have a relatively long positive boiler plate
"(masterpiece, flawless detail, best quality, absurd quality, absurdres), sharp focus.
(miv4t), so-bin, painting_\(medium\), hyper stylized.
perspective, rule of thirds, detailed shadows, atmospheric, cinematic, dramatic, bokeh, depth of field,"