Have you seen fal.ai/dynamic where you can perform image to image synthesis (basically editing an existing image with the help of diffusion process) using LCMs to provide a real time UI?