MiniMax H3 Opens the Floodgates: Open-Weights Video Model Generates Stereo Audio and 2K Clips on a Consumer GPU

MiniMax releases H3, its first open-weights omni-modal video model — capable of generating 2K video with native stereo audio from text, images, video, or audio prompts, and it runs on an RTX 3060.

MiniMax has released H3, a 33-billion-parameter omni-modal video generation model under open weights — and ComfyUI shipped day-zero native support. The model collapses five tasks into one: feed it text, images, video, or audio, and it generates up to 15 seconds of 2K video with native stereo sound in a single pass.

Unlike previous generations Hailuo 01 and 02, H3 is the first MiniMax video model released with open weights. It runs locally on consumer hardware — ComfyUI confirms it operates on an RTX 3060. The model handles text-to-video, image-to-video, first-and-last-frame control, and reference-to-video where a reference clip supplies motion while separate inputs define subject and style.

The standout feature is native stereo audio, generated alongside video rather than bolted on afterward. Motion transfer allows a reference video to drive camera moves or performance rhythms while the model fills in new content — making iterative shot editing practical on desktop hardware. The Hugging Face model card shows 1,750 community likes within four days of release.

Sources: ComfyUI Blog, Hugging Face