Introduction

MiniMax H3 is happiest when you stop decorating a still image and start directing a tiny film. Give it a first frame of the story, one readable action, a camera that has a job, and sound that lands on the right beat — and the model will build picture and audio in the same pass. The official grammar looks formal at first glance. Once you see the three fields, it feels like a shot list you can actually finish in one sitting.

The Official Prompt Shape

H3 prompts are built as an audiovisual timeline. For pure text-to-video (T2VA), you skip any image-alignment line and go straight into three named fields:

integrated_multimodal_description: [Shot 1] ...

overall_soundscape: ...

non_diegetic_music: ...

  • integrated_multimodal_description is the film itself: style, framing, who is where, what changes, who speaks, and any sound the characters can hear.
  • overall_soundscape is the physical world: rain, footsteps, fabric, breathing, chairs scraping. No spoken lines here.
  • non_diegetic_music is score the audience hears and the characters do not. Write instruments, tempo, and volume change — not “sad” or “epic.” Use N/A when you want silence on that layer.

If you are animating a start frame (I2VA), put this exact line first, then a blank line, then the three fields:

For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] ...

overall_soundscape: ...

non_diegetic_music: ...

First-and-last-frame and last-frame-only modes use their own alignment sentences. The body below them stays the same: describe the path between states, not two frozen photographs glued together.

Pro Tip

H3 clips are short — think 4 to 15 seconds at 24 fps. One clear beat beats three crammed ones. If you only need the camera to creep closer, that is camera motion inside Shot 1, not a new cut. A cut should change information: a new face, a new room, a new time.

Shots, Cuts, and Camera Grammar

Open Shot 1 with style plus framing. No timestamp on the first shot.

[Shot 1] Live-action, cinematic, a medium-wide shot frames...

Useful style words: Cinematic, live-action, 2D-animated, 3D CG, claymation, watercolor, vintage film.

Later shots get a rising timestamp written to three decimal places on the seconds clock the official guide uses:

[Shot 2] At 00:05.000, the camera cuts to...

Write camera moves as a natural sentence with three optional parts: type + amplitude + speed.

  • Types: Zoom In / Zoom Out, Push In / Pull Out, Pan Left / Pan Right, Truck Left / Truck Right, Tilt Up / Tilt Down, Pedestal Up / Pedestal Down, Arc Shot, Tracking Shot, Static Shot, Shake Slightly / Shake Strongly, POV, Roll Clockwise / Roll Counterclockwise
  • Amplitude: with small amplitude or with large amplitude (skip if medium)
  • Speed: at slow speed or at fast speed (skip if normal)

The camera pushes in with small amplitude at slow speed toward the folded letter in her hands.
The camera holds a static shot as the runner exits the frame.

Visible on-screen words — signs, neon, labels — go in English double quotes, copied exactly, even if the letters are not English:

A red neon sign reading "营业中" glows above the doorway.

Sound: What Lives Where

Treat audio as part of the timeline, not a garnish.

Put in the multimodal description: spoken lines, singing, radios, phones, and any music the characters themselves can hear.

Put in overall_soundscape: weather, room tone, footsteps, impacts, laughter that is not a line of dialogue. One short paragraph, one to four sentences.

Put in non_diegetic_music: piano, strings, pulse, fade. One to three sentences. Or N/A.

overall_soundscape: Steady rain taps against the café windows while low room ambience continues underneath. The entrance bell rings once, followed by wet footsteps and the soft scrape of a chair.

non_diegetic_music: Sparse piano notes at a slow tempo, joined by sustained low strings that gradually increase in volume before fading out.

H3’s dialogue layer is stable across Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish. Name the language inside the dialogue tag so the mouth and the voice agree.

Dialogue: One Voice, Then Two

This is the part that makes H3 feel like a scene instead of a silent commercial.

Every talking (or singing, or off-screen) person gets a stable speaker ID: (S1), then (S2), and so on. Silent extras get no ID. The same person keeps the same number across every shot. If two already-numbered people shout the same line together, write (S1,S2).

Identity lives outside the tag. The tag holds only the language and the exact words.

Pattern:

who they are + voice color + (S#) + action + says: <d>[Language] exact line.</d>

Solo character

Give enough voice paint on first appearance — age, pitch, texture, pace — then let them speak once. Short lines lip-sync better than speeches.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium close-up frames a young woman in a navy coat beside a rain-streaked train window. Warm carriage light catches the paper in her lap. The camera trucks right with small amplitude at slow speed as she lifts her gaze toward the passing city lights. The quiet, breathy young woman (S1) says: <d>[English] I get off at the next station.</d> She folds the letter along its existing crease.

overall_soundscape: Train wheels keep a steady metallic rhythm under a low ventilation hum. Rain ticks against the glass while the letter rustles in her hands.

non_diegetic_music: Sustained cello notes at a slow tempo with widely spaced piano tones, gradually decreasing in volume.

Need a voiceover instead of moving lips? Use the official phrase and lock the mouth:

The man (S1) says in an off-screen voiceover: <d>[English] I still remember that road.</d> while his lips remain completely closed.

If a line is chopped by the end of the clip, mark the truncation with <cutoff>. If the same sentence crosses a cut, say the audio continues seamlessly across the cut.

Two characters

Give each speaker their own ID the moment they first make sound. Let them take turns. Overlapping chatter is harder to sync than a clean question and answer. Keep both faces, wardrobes, and eyelines consistent for the whole clip.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium two-shot frames an older man and a younger woman sitting across from each other in a quiet wooden train compartment at dusk. Warm sidelight from the window holds on both faces. The camera holds a static shot. The older man with a low, gravelly voice (S1) glances at the compartment door and says: <d>[English] Did you lock it?</d> The younger woman with a bright, quick voice (S2) looks back at him and answers: <d>[English] Twice. We are fine.</d> She settles against the seat as the countryside slides past the glass.

overall_soundscape: Soft rail rhythm under the floor, a faint carriage rattle, and cloth shifting as they adjust in their seats. No other passengers speak.

non_diegetic_music: A slow acoustic-guitar figure with quiet upright-bass notes, holding steady and fading in the last second.

Unison — both voices on the same words — uses the compound ID:

The two children (S1,S2) shout together, <d>[English] Wait for us!</d>

Do not rewrite the line inside <d>. Do not translate it. Do not put the voice description inside the tag. Those three rules are the whole secret.

A Clean T2VA You Can Steal

Here is the bakery example from the official shape, rewritten so you can see every field working together. Copy the skeleton, swap the people.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium-wide shot frames a baker opening the shutters of a small street bakery before sunrise. The camera pushes in with small amplitude at slow speed as the middle-aged baker with a calm, slightly raspy voice (S1) places a fresh loaf on the wooden counter and says: <d>[English] First batch of the morning.</d> [Shot 2] At 00:05.000, the camera cuts to a close-up of steam rising from the sliced bread while the baker's final words carry over from the previous shot.

overall_soundscape: Wooden shutters scrape open over a quiet street as trays clink softly inside the bakery. The doorbell rings once, followed by light footsteps and the crisp sound of bread being sliced.

non_diegetic_music: A soft acoustic-guitar pattern at a moderate tempo, joined by sparse upright-bass notes and a gentle fade at the end.

Write the first shot as a start state. Write one action. Write the camera’s single job. Write who speaks, in which language, with which voice. Land on a last picture the viewer can hold. Then name the room’s noises and, if you want it, the score.

That is correctly prompting MiniMax H3: not more adjectives, just a timeline the model can actually shoot.

Conclusion

You do not need a film degree — you need a shot that can be seen, a sound that can be heard, and a line short enough to live inside a few seconds of light. Start with one character and one sentence. When that lands, invite a second voice across the table. H3 is listening. Go make something that talks back.

—Aurelia ♡