Usage

This model was trained on a very small dataset. You might need img2img/impainting techniques to generate acceptable results in certain scenarios. All demo images are not post-processed. Using with AnyLoRA is advised, but you can use any other model (the style might be off a bit).

Activation tag: kanon-mom

Extra tags (recommended): medium hair, dark red hair, purple eyes, bangs, adult face

Ratio: 0.6 - 0.7

Training

Dataset: anime screenshots; 11 close up pictures (repeat 8) + 17 mostly upper body pics (repeat 5)

Unet lr: 3.5e-4, Epochs: 14, Rest: default

Base: AnyLoRA