Experimental usage of stable-fast.
git clone https://github.com/gameltb/ComfyUI_stable_fast custom_nodes/ComfyUI_stable_fast
It can work with Lora, ControlNet and lcm. SD1.5, SDXL, and SSD-1B are supported.
Run ComfyUI with --disable-cuda-malloc may be possible to optimize the speed further.
Note
- If you are using WSL, please do not install Triton for the time being due to bugs.
- stable fast not work well with accelerate, So this node has no effect when the vram is low. For example: 6G vram card run SDXL.
Note
- stable fast will optimize the speed when generating images using the same model for the second time. if you switch models or Lora frequently, please consider bypass it temporarily.
- stable fast should be directly connected to ksampler, and it is better not to have other nodes between them.