[fumufumufumu]

Trying to use Wan2.1 to generate AI video or other various LLM or StableDiffusion style stuff is slow compared to other other platforms. I don't know how much of that is because the code is not optimized for M1+ Max (Activity Monitor shows lots of GPU usage) or how much of it is it's just not up to the competition. Friends on 4070 Windows PC are getting results many X faster and 4070 perf iss not even close to 4090

[rcarmo]

You need to run it under MLX, and AFAICT ComfyUI and the like are not really optimized for it (or at least not as optimized as LLM inference).