This is an experimental merge of anime models which use a style I'm a fan of. As I find more models, I will try to add them into newer versions. As of v2.10, this merge uses 64GB of models.
I've used ClearVAE as the baked in VAE just as something sorta in between to replace the messed up VAE that results from the merge. I've included 2 comparisons in the previews so you can decide which one is best for you.
I'm not sure how exactly to describe the style, but it's sort of a flat, semi-2D anime style that I'm interested in. mistoonAnime v2.0 is a good example of it.
Below are the settings for the preview images. You can use them as a guideline for how to use the model if you want, or just use the model however you normally do. I'm just providing this for extra details on the process. Some images used slightly different settings, but they all hover around the same settings.
Preview images use prompts chosen with Ranbooru.
Initial Generation Settings:
512x768.
EulerA sampler.
Exponential Sigma Algorithm (It's recommended to use over Karras now, at least for DPM++ 2M).
50 to 100 Steps (50 is the lowest I got for most models, but 32 might be okay).
3.5 CFG.
Clip Skip 1 (Not sure if 2 is better. I haven't really tested it).
Hiresfix Settings:
1.5x to bring it to 768x1152.
EulerA sampler.
Force Hires
50 to 100 Steps.
0.55 Denoise.
ESRGAN 4x NMKD YandereNeoXL upscaler.
Description
AniMerge_v2-10 = PhotoMerge_v1-2 + (
(
(
(
mistoonAnime_v20Merging
+ flat2DAnimerge_v30
+ toonyouJP_alpha1
+ forgottenmixAnimation_v10
+ lively_v10
+ Nullstyle_v20
+ entity404
+ helloflatart_V12a
+ lembo2D_v10
+ pikasNewGeneration_v20
+ aamAnyloraAnimeMixAnime_v1
+ aiomonstergirls_v3
+ ghibli-diffusion-v1
+ sudachi_v10
+ helloflatanime_2dV12d
+ yden_v30
)
/ 16
)
- v1-5-pruned
)
* 0.85
)