MiniMax H3 Prompt Assistant
Write MiniMax H3 prompts for storyboards, ads, performances and edits, with every detail spelled out.
Start from a use case
- Write a performance sceneSmall changes in face and voice, with the line kept exact.Prompt
- Plan a product ad in shotsTimed shots with the picture, the sound and the score written out.Shot plan
- Cast a presenter with a productYour presenter and your product each defined once and kept in every shot.References
- Bring an old photo to lifeMotion that starts exactly on your picture.Frames
- Fix a face that talks when it shouldn'tFind out why the wrong person moved their lips.Fix
What you provide
- Every detail you care aboutThis model does not fill gaps, so say who stands where, what moves and what is heard.
- Who speaks and howThe lines, who says them, and anyone on screen who stays quiet.
- References and their jobsUp to nine pictures, three videos and three sounds, each with what to take from it.
How it works
- Spell out the shotPositions, facing, motion, speed, camera and sound are written out, because nothing gets added for you.
- Split picture and soundPicture and dialogue in one field, the soundscape in a second, the score, or none, in a third.
- Time the cutsLater shots start with their time and a new framing, with speaker IDs kept stable.
- Send itPrompt, shot list, checks and settings, ready to open with MiniMax H3.
One idea, before and after
girl walking in the rain in a neon city at night cinematic
integrated_multimodal_description: [Shot 1] Medium full shot at eye level. A young adult woman with shoulder-length damp dark hair, wearing a dark water-resistant hooded jacket over streetwear, walks forward at an unhurried pace along a glistening, wet city sidewalk at night. Continuous rain falls through beams of vibrant magenta, cyan, and amber neon light cast by towering storefront signs and overhead digital displays. Water ripples in shallow street puddles reflecting glowing neon hues beneath her boots. The camera tracks backward smoothly at a matching speed directly ahead of her, keeping her centered in the frame while maintaining a soft, atmospheric bokeh in the background city street. Around 00:04.000, she tilts her chin upward slightly to glance toward the high-rise lights before returning her calm gaze forward as light mist drifts past. No on-screen text, no subtitles, no captions, no watermark. Coherent anatomy, natural hands. overall_soundscape: Consistent patter of rainfall on wet asphalt and pavement, rhythmic wet footstep splashes in shallow puddles, and a distant low drone of urban night traffic. non_diegetic_music: Atmospheric downtempo synth pads with subtle, melancholic reverberation.
Questions
Why three fields?+
MiniMax H3 reads picture and dialogue, the soundscape and the score separately. An empty score field invites added music, so the assistant writes N/A when you want none.
Why are its prompts so detailed?+
H3 does not expand prompts by itself, so anything left out is up to chance. The assistant writes every position, movement and sound.
How does it handle dialogue?+
Lines stay in their original language and wording; each speaker gets a stable ID, and anyone else on screen is told to keep their lips closed.
How long can clips be?+
5 to 30 seconds from text, 4 to 30 from an opening frame, and 4 to 15 with references.
Do camera commands in brackets work?+
No. H3 understands camera moves written in plain English, which is how the assistant writes them.
What does a run cost?+
A few credits, shown on the button before you run. You never pay more than that price, and nothing is charged if no usable result comes back.