How Do I Run MiniMax H3 in ComfyUI?

Overview

Right now, only through an API node — not with local weights. As of early August 2026 MiniMax H3 is API-only: it was announced on July 31, 2026 as open-weight, but the weights have not been released, so there is no local checkpoint to load into ComfyUI. Source note: Verified 2026-08-06 against the MiniMax official H3 blog, MiniMax Video Generation API docs, and Hugging Face MiniMax-H3 model page.

What works today is an API-based workflow. If a ComfyUI node exists for MiniMax H3, or a generic video-API node, it calls the hosted service, sends your prompt and any reference inputs, and polls for the finished clip — the generation runs on MiniMax’s infrastructure, not your GPU. The graph looks familiar: load inputs, set duration (4–15s) and resolution (up to 2K), submit, then save the result.

A true local, weights-based ComfyUI workflow only becomes possible once the open weights ship and a compatible node lands. Keep ComfyUI and any custom nodes updated so you are ready when that happens.

Until then, read “run H3 in ComfyUI” as “call the H3 API from ComfyUI.” Verify current node availability and the model’s release status before building a graph around local weights that do not yet exist.