H3 fastvideo VSA acceleration becomes much slower after loading a LoRA in ComfyUI

#37
by Ree0123 - opened

Hi,

I am using MiniMax H3 fast video with VSA attention in ComfyUI.

The workflow runs at fast speed without LoRA, but becomes much slower after enabling even a single H3 LoRA. The generation completes successfully and no obvious error is shown.

Does it support regular H3 LoRAs? Is there a recommended way to convert or load a LoRA while keeping VSA acceleration active?

Thank you.

Sign up or log in to comment