MiniMax H3 Text to Video

MiniMax H3 prompting becomes easier to debug when visual events, environmental audio, and background score are treated as separate control layers rather than one undifferentiated paragraph.

integrated_multimodal_description:
[Shot 1] ...

overall_soundscape:
...

non_diegetic_music:
...

The three fields solve different problems: the first controls the audiovisual timeline, the second summarizes the world the scene sounds like, and the third controls audience-only score. Keeping these roles separate makes the prompt easier to write and easier to debug.

Subject + Action + Environment + Camera is a useful planning framework, but it should not be confused with the native H3 prompt format. The planning decisions ultimately belong inside H3's audiovisual timeline and dedicated sound fields.

Dialogue and tightly synchronized sound events belong with the visual event they accompany. Persistent environmental audio belongs in the overall soundscape, while audience-only background score belongs in non_diegetic_music.

Visual event

A ceramic cup is placed on a saucer.

Synchronized sound

The cup touches the saucer with a short ceramic click.

Persistent soundscape

Low cafe room tone, rain against windows and distant conversation.

Audience-only score

Sparse muted piano at a slow tempo, fading before the final frame.

integrated_multimodal_description:

[Shot 1] Live-action, cinematic. A medium-wide tracking shot follows a woman in a red coat walking through a rain-covered station.

[Shot 2] At 00:05.000, the camera cuts to a close-up from her left side as she stops and looks toward the arriving train.

The first shot establishes the initial state. A later shot should introduce meaningful new information—such as a new viewpoint, state, spatial relationship, or story beat. If only camera distance or angle needs to change slightly, prefer continuous camera movement instead of adding an unnecessary cut.

Cut timing must remain inside the selected video duration.

integrated_multimodal_description:
[Shot 1] Live-action, cinematic. A medium-wide shot frames a woman in a long red coat standing alone on a snow-covered rural train platform during blue hour. Warm light glows through the station windows behind her while cold blue ambient light fills the platform. She begins walking slowly toward the far end of the station as the camera dollies sideways beside her with small amplitude at slow speed. Wind moves the lower edge of her coat and loose strands of hair while fine snow crosses the frame.

[Shot 2] At 00:05.000, the camera cuts to a close-up from her left side. She stops and turns toward the tracks as the distant headlights of an approaching train become visible through the snowfall. The light grows gradually brighter across her face while she remains still.

overall_soundscape:
Soft winter wind moves across the platform with light footsteps compressing snow beneath her shoes. A low station ambience continues underneath the distant metallic vibration of an approaching train.

non_diegetic_music:
Sparse felt-piano notes at a slow tempo with sustained low strings underneath, gradually increasing in volume before fading at the end.

Why this structure works

Visual timeline

Defines initial state, action, camera and later shot.

Soundscape

Contains persistent environmental and physical sound.

Music

Separates the audience-only score from the world of the scene.

Editorial example. It is not labeled as an official or tested output.

Use text-to-video when the visual scene exists primarily as an idea. Use image-to-video when you already have a composition, character appearance, product design, or keyframe that should anchor the generated motion.

Do not rewrite every part of an H3 prompt when only one layer is wrong. Treat the visual timeline, environmental sound, and background score as separate control surfaces and revise the layer responsible for the problem.