[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).