How to write a MiniMax H3 video prompt

Use integrated_multimodal_description for the visible timeline, actions, shot changes, speakers, dialogue, and synchronized sounds. Use overall_soundscape for ambience, physical sounds, and non-verbal human sounds across the clip. Use non_diegetic_music only for score that the characters cannot hear; write N/A when there is no audience-only music. Keep the field names and their order unchanged.

Begin with [Shot 1] and no timestamp. Introduce later cuts as [Shot 2] At 00:03.500, using strictly increasing times inside the requested 4–15-second duration. A cut should add new information; use camera motion for a smaller viewpoint change. Describe motion naturally as type, meaningful amplitude, and speed—for example, a slow small-amplitude push in—while keeping subject motion separate.

Give every speaking or singing source a stable ID such as (S1) and reuse it across shots. Put only the language tag and exact spoken words inside <d>, as in <d>[English] The train is leaving.</d>. For an off-screen voiceover, say so explicitly and state that the visible character keeps their lips closed. Put text that must appear in the scene in straight double quotation marks and preserve its original language and punctuation.

Keep timed dialogue, singing, radio, and other sounds heard by characters in integrated_multimodal_description. Summarize rain, footsteps, impacts, breathing, and room tone in overall_soundscape without repeating dialogue. Describe non-diegetic music with instruments, tempo, rhythm, and dynamic change—not an abstract mood. Match every action and cut to the selected duration.