How to use MiniMax H3 for your first AI video

The simplest way to use MiniMax H3 is to design one shot, not an entire film. Choose a single visible outcome, prepare only the references that support it, and write a prompt that another person can review. This keeps the first generation diagnostic. If the result fails, you can identify whether the problem came from identity, action, camera, sound, or an overloaded brief.

Supplied H3 demonstration: a 15-second horizontal text-to-video result.

Step 1: choose the right H3 mode

Use text to video when the shot does not depend on a specific person, product, location, or opening frame. This path begins in first and last frame mode with no image attached.

Add one image when the opening composition matters. Add two when both the start and finish need visual anchors. A first-and-last-frame request is useful for a reveal, a transformation, a UI transition, or any shot where the destination is as important as the opening.

Choose Omni Reference when several media sources have different jobs. The manual allows up to nine images, three videos, and three audio clips, with 12 mixed files at most.

Step 2: give every reference one role

Write the role next to the asset before uploading it. A portrait controls facial identity. A full-body sheet controls proportions and clothing. A product image controls shape and surface details. A video controls action timing or camera movement. An audio clip controls a spoken line or voice tone.

Do not use two references that disagree about the same detail unless the prompt explicitly describes a change. If one character sheet has short hair and another has long hair, the model has no reliable identity target.

Step 3: write the shot in visible order

Begin with the output format and duration. Then name the subject, setting, action order, camera, sound, and ending. A compact structure looks like this:

1. Create a 10-second 16:9 video.
2. Use image 1 for the character's identity and image 2 for wardrobe.
3. The character crosses the room, stops at the desk, and opens the box.
4. The camera makes a slow lateral track and holds on the open box.
5. Use quiet room tone, footsteps, and one soft mechanical click.
6. Keep the face, jacket, box shape, and desk layout unchanged.

The prompt limit is 7,000 characters, but clarity matters more than filling it.

Step 4: select duration and aspect ratio

H3 supports 5 to 15 seconds. Pick the shortest duration that contains the full action. Extra seconds can invite unnecessary movement.