Notes:
- Doesn't work with Nunchaku(yet)
- 4-8 steps LoRA's and checkpoints doesn't generate good results, so not recommended
- Recommending usage of Fp8_e4m3fn or it's distilled versions
- Recommending 20-40 steps, CFG 2.5, Euler Simple
- Might generate burn lines but fixable with an upscaler
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support