Hi, currently the difussion isn't the bottleneck, it's more the fact that I repeat the same process for each frame and the process invoke multiple part (like depth estimation, text desctiption from vision llm, etc...) some can be done once every two frame or more, some could be run on another pc in same local network, is there any ways to do that currently with comfyui (and maybe asyncdiff?) ? The diffusion process has become the fastest part (with turbo xl on 2 samples), but not the others parts.
Hi, currently the difussion isn't the bottleneck, it's more the fact that I repeat the same process for each frame and the process invoke multiple part (like depth estimation, text desctiption from vision llm, etc...) some can be done once every two frame or more, some could be run on another pc in same local network, is there any ways to do that currently with comfyui (and maybe asyncdiff?) ? The diffusion process has become the fastest part (with turbo xl on 2 samples), but not the others parts.