MiniMax H3 Prompt Guide: Camera and Sound

The short answer

A MiniMax H3 prompt works best when it reads like a shot brief, not a list of adjectives. Name the subject, the action, one camera move, the light, and the sound you expect. That five-part structure gives the model enough to generate motion and native audio together, instead of guessing what should happen between the first and last frame.

This guide covers text-to-video prompts, image-to-video prompts, and the mistakes that make a shot drift. Every example below uses the same controls the studio exposes: 5, 10 or 15 seconds, 768P or 2K, and six aspect ratios.

Use a five-part shot brief

Write the prompt in this order. You do not need five sentences, but each part should be present or deliberately left out.

1. Subject — who or what is on screen. A rose-gold chronograph, a concrete atrium, a woman by a window.
2. Action — what changes during the shot. The camera orbits, dust drifts, the subject turns her head.
3. Camera — one move, with a direction and a speed. Slow push-in, low aerial dolly, handheld follow.
4. Light and atmosphere — where the light comes from and what is in the air. Window light from the left, sea haze, neon reflections on wet asphalt.
5. Sound — the ambience, impacts or score that belong to the scene. Rain on metal, a low room tone, a soft synth pulse.

A useful test: read the prompt back and ask whether a camera operator could shoot it. If the answer is no, the model has even less to work with.

Give the camera one job per shot

MiniMax H3 generates a continuous clip, so a prompt that asks for a push-in, a whip pan and a cut in the same five seconds is asking for a contradiction. Pick the move that carries the idea.

• Push-in — product reveals, portraits, a detail that should become the subject.
• Pull-back — endings, reveals of scale, a subject left alone in a space.
• Orbit — objects with more than one interesting side, watches, bottles, sculptures.
• Handheld follow — street scenes, documentary energy, a subject moving through a crowd.
• Static frame — when the action inside the frame is strong enough. Wind in fabric, light moving across a wall, a head turn.

You can add a lens or height cue when it matters: low angle, eye level, wide lens, macro. Keep the cue physical. "Cinematic" is a mood, not a direction.

Describe light as a source, not a mood

"Beautiful lighting" gives the model nothing to position. A source does.

• "Soft morning light from a tall window on the left, curtain moving, gentle falloff on the right."
• "Golden-hour sun low behind the building, long shadows across the floor, dust visible in the beam."
• "Neon signs on both sides of a wet street, coral and cyan reflections, dark sky above."

Direction, hardness and color are the three levers. If you want a specific look, state the source and the contrast. If you want the model to choose, leave one of them open and keep the rest specific.

Put sound cues in the same prompt

MiniMax H3 generates native stereo audio with the picture. You do not need a second audio pass to get ambience, impacts or a music bed. Describe the sound the way you described the light: source first, then character.

• "Rain ticking on a metal awning, distant traffic, no music."
• "Low room tone, a single clock tick, footsteps on concrete."
• "Soft synth pulse under a wide desert wind, no dialogue."

Two practical rules. First, do not ask for a separate voice-over or a licensed song; the model generates sound for the scene, not a post-production mix. Second, listen to the result before you publish. Native audio is part of the output, but it is still your job to check levels and fit.

Adapt the prompt for image-to-video

When you upload a first frame, the model already knows what the subject and light look like. Do not spend the prompt re-describing the image. Spend it on change.

Bad: "A woman by a window, soft light, film grain."

Better: "She turns her head slowly toward the camera, the curtain lifts in the window draft, the light stays soft on her left cheek, quiet room tone."

If you provide both a first and a last frame, describe the journey between them. The model has to arrive at the ending you supplied, so the prompt should explain the motion, not a new scene.

Three copy-ready prompts

These prompts match the examples in the MiniMax H3 showcase. Use them as templates, not as a fixed house style.

Macro product shot

"Macro shot of a rose-gold chronograph on black glass, slow orbiting camera, rain droplets catching coral light, shallow depth of field, soft ticking and distant rain."

Why it works: one subject, one camera move, one light source, and a sound cue that belongs to the surface.

Architecture in motion

"Brutalist concrete atrium at golden hour, camera glides upward through floating dust, long shadows, quiet monumental mood, low room tone with a faint echo."

Why it works: the camera move is the action. The sound describes the space instead of competing with it.

Window-light portrait

"Portrait of a woman by a tall window, soft morning light, curtain moving, subtle head turn, film grain, quiet room tone and a distant bird."

Why it works: the motion is small enough for a short clip, and the light has a clear direction.

Common MiniMax H3 prompt mistakes

• Too many events. A five-second clip cannot hold a beginning, a turn and a reveal. Pick one beat.
• Vague verbs. "Moves beautifully" is not a direction. Say what moves and how.
• Contradictory light. "Hard noon sun" and "soft window light" in the same shot pull the frame apart.
• Re-describing an uploaded image. In image-to-video, describe the change instead.
• Ignoring the aspect ratio. A 21:9 shot and a 9:16 shot need different compositions; write for the frame you chose.
• Prompting a full edit. H3 generates a continuous shot. Cuts, captions and transitions belong in your editor.

FAQ

How long should a MiniMax H3 prompt be?

One to three sentences is enough for a five-second shot. For a fifteen-second shot, add one more sentence about how the action develops. Length is less important than completeness: subject, action, camera, light and sound.

Does MiniMax H3 generate audio from the prompt?

Yes. Audio is generated with the picture, including stereo ambience and effects. Describe the sound source in the prompt, then review the result the same way you would review a recorded track.

Can I reuse a text-to-video prompt for image-to-video?

You can reuse the structure, but not the content. In image-to-video, the first frame already establishes the subject and light. Rewrite the prompt around motion, timing and the ending.

Which settings should I change first?

Start at five seconds and 768P while you are testing the prompt. Move to 2K and a longer duration when the motion and framing are right. The MiniMax H3 pricing page shows the credit cost for each combination before you generate.

Conclusion

A good MiniMax H3 prompt is a compact shot brief: what is on screen, what changes, how the camera moves, where the light comes from, and what the scene sounds like. Build that habit and you will spend less time re-rolling and more time choosing between usable takes.

Ready to test a prompt? Open the MiniMax H3 video generator, start with a five-second shot, and compare the result with the brief you wrote. For a motion-focused workflow, read the MiniMax H3 image-to-video guide.

Sources

• MiniMax H3 launch post — official model announcement, 31 July 2026.
• Hailuo AI MiniMax H3 page — official product page with resolution, duration and audio details.