The goal of this repo is to
- Capturing webtoon character's unique characteristics
- Get varieties of poses, gestures and actions without damaging too many characteristics
For the LoRA inference
- Current LoRA checkpoints are under development. Instruction will be added soon
- For those who want to try out. I recommend Midnight Mixers as the base model.
- 512(width) x 640(height)
- 50 steps / 7 cfg. Step below 40 would yield poor quality
- 0.5~0.6 range LoRA weight.