A LoRA model trained on images of lingerie from Affix Parapara.
Description
Trained on 66 768x768 images at 100 steps per image
FAQ
Comments (3)
this alters the face of the original character too much
Hmmm. that's odd considering the original data set had the vast majority of model's faces cropped out. It may be my choice of training checkpoint. What characters/lora's are you having conflicts with? I'm thinking about experimenting with a V2 at some point.
imakphotoJun 25, 2023
Having the same problem. Using with wyvermix checkpoint. No other LoRAs. Even at 0.5 the faces end up terribly distorted. At 0.3 it seems to mostly clear up, but I can still see it's affecting the usual beautiful faces that wyvermix turns out.