CuteSweetAnime版本O_CSA_MIX12_LF (ID: 473021)

Summary :

Those models are merges of nice anime models that also could be merged model.
It provide a nice and cute anime style.
The style is rather stable with small variation a seed to an other.

Can do creative art, help with chatGPT prompts.

Technical information :

Sampler : Euler a (fast, nice results and basic one)

Steps : 20-35 provide good results

CFG : 7-8 is good

Clip skip 2

I guess that short prompt are better, could be confused by very long one, you can try it if you want.

For negative prompt, could use my setting or another one, should be fine.

Special tips :

To get a very nice workflow, i suggest doing this :

1) Generate an image you like, with not big defect (512x768 here)

NB : You can also use HiresFix to have a better quality before up-scaling (here set the ratio x1.0 and choose R-ESRGAN 4x+ Anime6B, it's x2.0 by default). Notice that it will most of the time make lines thicker

2) Upscale it with img2img x1.5 (batch count 3-5), denoising strength 0.4-0.5

3) Select the best one.

4) Upscale it with extras --> R-ESRGAN 4x+ Anime6B --> x2

5) Results : High Quality image with medium VRAM

Bonus : You can enhance more the quality by re-img2img your first upscale image with around 0.20-0.30 denoising strength and x1.0 before sending to extras.

描述:

MIX 12 + some loras fused in it => global improvement on the aesthetics and style.

Gives nice shiny body when specify those keywords : oily skin, shiny body

Can draw a large range of things and good stability with loras.

训练词语: