Minimax H3

MiniMax H3 (Hailuo 03) is a frontier video model that generates 2K video up to 15 seconds long

MiniMax H3: Cinematic 2K Video from Text, Images, and References

MiniMax H3 (Hailuo 03) is a frontier video model that generates 2K video up to 15 seconds long. Give it a text prompt, a starting image, or a whole set of references: up to 9 images, 3 video clips, and 3 audio clips in a single generation.

The reference mode is the headline. H3 keeps your subjects consistent with their reference images, follows motion from reference clips, and syncs to reference audio. Tag each file right in the prompt (@Image1, @Video1, @Audio1) and tell H3 what role it plays.

What can you do with it?

How to use it

  1. Pick your mode. Omni Mode takes a prompt plus reference media. Frame Mode takes a first frame, an optional last frame, and a motion prompt.
  2. Attach your references. Up to 12 files total: 9 images for subjects and style, 3 video clips for motion, 3 audio clips for sound. Video and audio references should be 2 to 15 seconds each.
  3. Write the prompt. Call out each reference by its tag and say what it contributes. Be concrete about camera movement, light, and pacing.
  4. Set duration and ratio. 5 to 15 seconds, at 2K (2560×1440). Seven aspect ratios from 21:9 to 9:16, or adaptive to match your references.

What you get

One 2K video, 5 to 15 seconds, as mp4. In frame mode the output ratio follows your first frame; in omni mode it follows your aspect ratio pick or adapts to the references.

Tips

All AI models & tools

See pricing and start creating on Mitte