with only need 4 sampling step, this checkpoint can generate a good 2D anime picture, if combined with TensorRT, you can get 64 batch image of 512x512 under 1 minute
This is my first time playing with the LCM Training script. Thanks for LCM Team with complete explanation and paper regarding LCM.
So, i retrain my Everyjourney dataset with LCM training script, and the result is this checkpoint model. So no need to load LCM LoRA to reduce sampling step value.
If you like my works, please consider buy me a Ko-Fi
i am also accept commision for custom trained checkpoint model or LoRA, visit my page at Fiverr
Suggested Parameters:
- LCM Sampler is a must
- start from 4 sampling steps, if still no good, 8 sampling steps is a safe start.
- CFG scale is 1
A1111 webui user must install AnimateDiff extension, to get LCM sampler. ComfyUI is already at default.