H3 Studio
Make short videos with their own sound on your own GPU — wire image, words and clip into a pipeline and press Run.
About
A studio for MiniMax H3, the omni-modal model that renders video and synchronized stereo audio in one pass, running locally through ComfyUI. Instead of node fields you build a pipeline on a canvas (or as a tree): a still, a description, a soundscape and music go into a clip, clips go into a reel. Run it, or run from any step, and the history keeps every clip with the prompt that made it, ready to rate and reuse. Templates start the common jobs: an image reel, a face swap onto a reel, a cut-out. An AI writer can fill the prompts from a photo, and a build-with-AI job lays out a pipeline for you. It runs on your own desktop GPU; the web version sits behind an email wall, with its own limits on what a run may cost.