The model is trained on the same dataset as the Pony version, but I believe this one might perform better.
base model：https://civarchive.com/models/647664/sanaexl-anime-v10
If you use this model, you should pair it with the DTG extension to better optimize its performance.
https://github.com/KohakuBlueleaf/z-a1111-sd-webui-dtg
Usage：
Steps: 28
Sampler: DPM++ 2M SDE Heun
Schedule type: Exponential
CFG scale: 5
Hires.fix:
Hires upscale: 1.8, Hires steps: 12, Hires upscaler: Lanczos
Denoising strength: 0.5
Description
-
FAQ
Comments (1)
Maybe it's just not for me, but this took way too much fiddling to notice any effect, and it wasn't a strong enough effect to use it often. It's a nice concept though!