MiniMax H3

MiniMax H3 is the successor generation in MiniMax's Hailuo video line, so if you searched for Hailuo, you're in the right place. It is an omni-modal model that reads text, images, video clips and audio clips together in one context.

Base H3 targets resolution (768p, 2K, 4K), H3 Max targets prompt adherence and aesthetics at 480p, 768p or 1080p, and H3 Max Turbo targets speed and lower cost. Max also adds first, middle and last frame control that can be combined with references. Turbo costs roughly half as much but gives up some fidelity, and it has no references or middle frames.

Use H3 reference-to-video for identity images, motion clips and audio in a high-resolution shot, or H3 Max reference-to-video when you also want first, middle and last frame anchors. Both cost more than plain generation, and H3 Max reference runs have the widest cost range. Turbo does not accept references.

You can use a prompt, a start image, an end image, and in reference mode up to 9 reference images, 3 reference videos and 3 audio clips, with a limit of 12 files in total. Video and audio clips are 2 to 15 seconds each and 15 seconds combined. Cite them in the prompt as Image 1, Video 1, Audio 1.

It is not built for edits longer than 15 seconds in one pass, exact on-screen text or logos, or guaranteed frame-accurate lip-sync. Faces can turn soft in distant or repeatedly re-fed footage, and conflicting references can cause identity drift. Avoid 4K or full 12-file reference loads while exploring.