Thanks for open-sourcing this great work!
May I ask what steps are needed to implement this speedup method in the Automatic1111 backend? I am particularly interested in using this method for our NeurIPS 2025 iterative image editing workflow, where more than 100 sequential editing steps is usually required. In each editing step, a single Canny edge ControlNet will be activated for 20% of the denoising steps, on average.
https://github.com/kenantang/spice
https://github.com/kenantang/spice-photoshop-plugin
Thanks for open-sourcing this great work!
May I ask what steps are needed to implement this speedup method in the Automatic1111 backend? I am particularly interested in using this method for our NeurIPS 2025 iterative image editing workflow, where more than 100 sequential editing steps is usually required. In each editing step, a single Canny edge ControlNet will be activated for 20% of the denoising steps, on average.
https://github.com/kenantang/spice
https://github.com/kenantang/spice-photoshop-plugin