Kasumi Arimura 有村架純: v1.0: ArimuraK4sum1-640-704.safetensors

Trained with 20 640*704 images on Vivid.

Maybe larger (640*704) training images are not better than normal (512*512 or 512*640)... At least I tested for hours and outputs are not satifying enough.

Maybe set your resolution around 640*704 before (or without) Hires.fix would help get better results.