MiniMax H3 Prompt Guide: Structure, Camera & Audio
How to write MiniMax H3 prompts: the official three-field structure, shot cuts, camera motion terms, dialogue tags, soundscape and music, plus image-to-video rules.
MiniMax H3 Prompt Guide
This guide explains how to write a MiniMax H3 prompt, based on the official prompt-writing guides published with the model. If you only want ready-to-use starters, see MiniMax H3 prompt examples instead.
Third-party note: minimaxh3.net is an independent MiniMax H3 topic hub, not the official MiniMax website. Primary sources: the MiniMax-H3 model card, the base prompt-writing guide, the full-reference guide, and the official prompt skills.
Two levels of prompting
MiniMax H3 is a system, not a single model. According to the model card, H3-Context-IR first rewrites free-form input into a structured “Context Intermediate Representation”, and H3-Base generates video and stereo audio from that representation.
That gives you two ways to write:
| You are using | What to write |
|---|---|
| A hosted app or the online studio | Plain natural language is fine. Be specific about subject, action, camera, sound and style. |
| A local H3-Base deployment without Context-IR | Write the structured format below yourself. The model card says Context-IR is critical to output quality and recommends following the prompting guidance if you build your own preprocessing. |
Even when a hosted rewriter does the formatting, the structure below is a useful checklist: a prompt that already answers “what is seen, heard and shot, in what order” leaves less for the rewriter to guess.
The core structure: three fields
For text-to-video (T2VA), the official final prompt has three fields, in this order:
integrated_multimodal_description: [Shot 1] ...
overall_soundscape: ...
non_diegetic_music: ...
- integrated_multimodal_description — the main body. Visuals, subjects, actions, shots, speakers, dialogue, singing and on-screen sound, written along the timeline.
- overall_soundscape — ambience, physical action sounds and non-verbal human sounds across the whole video.
- non_diegetic_music — background music only the audience hears.
Every detail should correspond to something visible or audible. Avoid abstract plot summaries.
Writing the multimodal description
Start Shot 1 with style and composition
Open [Shot 1] with the overall style and the initial framing. The guide lists styles such as Cinematic, live-action, 2D-animated, 3D CG, claymation, watercolor and vintage film.
[Shot 1] Live-action, cinematic, a medium-wide shot frames a baker opening the shutters of a small street bakery before sunrise.
Add cuts with timestamps
The first shot has no timestamp. Each later shot is numbered in sequence and starts with a strictly increasing cut time inside the video duration:
[Shot 2] At 00:05.000, the camera cuts to a close-up of steam rising from the sliced bread.
A cut should introduce new information: subject, space, state, viewpoint or time. If you only need a slightly closer or different angle, use camera motion instead of a cut. Cross-dissolve, fade or wipe are for when you explicitly want them.
Describe camera motion: type + amplitude + speed
Write camera motion as a natural sentence, not a pile of tags at the end.
| Motion type | Meaning |
|---|---|
| Zoom In / Zoom Out | Focal length changes, camera stays put |
| Push In / Pull Out | Camera moves forward / backward |
| Pan Left / Pan Right | Camera pivots horizontally in place |
| Truck Left / Truck Right | Camera slides sideways |
| Tilt Up / Tilt Down | Camera pivots vertically in place |
| Pedestal Up / Pedestal Down | Whole camera moves up / down |
| Arc Shot | Camera circles the subject |
| Tracking Shot | Camera follows a moving subject |
| Static Shot | No camera movement |
| Shake Slightly / Shake Strongly | Handheld-style shake |
| POV | Subject’s point of view |
| Roll Clockwise / Counterclockwise | Rotation around the lens axis |
Add with small amplitude / with large amplitude and at slow speed / at fast speed only when they matter. Medium amplitude and normal speed are usually left out.
The camera pushes in with small amplitude at slow speed toward the folded letter in her hands.
The camera pans right with large amplitude at fast speed, revealing the open doorway.
Tag speakers and dialogue
Give every character who speaks or sings a stable ID such as (S1), (S2), kept across shots. Describe the voice the first time (age, gender, pitch, timbre, pace, accent). Put only a language tag and the exact spoken words inside <d>...</d>; keep them verbatim and untranslated.
The young woman with a quiet, breathy voice (S1) says: <d>[English] I get off at the next station.</d>
The two children (S1,S2) shout together, <d>[English] Wait for us!</d>
For voiceover, use says in an off-screen voiceover and state right after the <d> block that the on-screen character’s lips remain closed. Use <scenetrans> when a line continues across a cut, and <cutoff> when speech is cut off by the end of the video.
Quote on-screen text exactly
Signs, labels, subtitles or neon text that appear on screen go in English double quotes, verbatim:
A red neon sign reading "OPEN LATE" glows above the doorway.
Writing the sound fields
overall_soundscape — 1–4 English sentences in one paragraph: wind, rain, traffic, footsteps, fabric, impacts, breathing, laughter. Do not repeat dialogue or on-screen music here. Use N/A only if you want complete silence.
non_diegetic_music — 1–3 sentences about instruments, tempo, rhythm and dynamics. Skip abstract mood words. Music the characters can hear (a radio, a singer in the scene) belongs in the multimodal description instead. Use N/A for no score.
A complete text-to-video example
This is the official T2VA example, reproduced for reference:
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.
Image-to-video and keyframe prompts
When images are involved, the three core fields stay the same, but a one-line alignment instruction comes first, followed by a blank line.
First frame (I2VA) — the image is the frame at 0.00 seconds:
For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.
Describe the image’s style, subject, composition and setting first, then the action that develops from it. Keep identity, clothing, colors and spatial layout consistent.
First and last frame (FL2VA) — Picture 1 opens, Picture 2 closes the video:
How the reference pictures align with the target video — Picture 1 (from Shot 1) aligns with the 0.00-second mark of the target video; Picture 2 (from Shot 1) aligns with the 8.00-second mark of the target video.
Do not describe two static images. Describe the motion path between them: poses changing, objects being handled, the composition evolving. A single shot usually works best.
Last frame (L2VA) — the image is the final frame:
How the reference pictures align with the target video — <Picture 1> (from [Shot 1]) aligns with the 6.00-second mark of the target video.
Invent a plausible earlier state, then let the action converge on the image in the final shot. The timestamp is the video duration with exactly two decimals.
Multi-reference prompts
H3’s omni-reference mode accepts several images, videos and audio clips. Its official format uses six sections (subject_definitions, summary, retention_analysis, detailed_description, overall_soundscape, non_diegetic_music) and labels like <Subject 1>, <Picture 1>, <Video 1>, <Audio 1>. For example:
<Subject 1> is the young woman in <Picture 1>, with long dark hair, a blue cardigan, and a thin silver necklace.
This mode has more rules than fit here; read the full-reference guide directly if you build for it.
Quick checklist
- Style and opening composition stated at the start of Shot 1
- One clear subject action per beat; cuts only when they add new information
- Camera motion written as a sentence, with amplitude/speed only when meaningful
- Speakers tagged
(S1), dialogue verbatim inside<d>with a language tag - On-screen text in double quotes, not translated
- Soundscape and background music described separately
- For image input, the alignment line comes first and the frames are not re-described statically
Using this on minimaxh3.net
The online studio takes a normal text prompt, so you do not need the full structured format. Borrow its logic instead: state style and framing, then action, camera motion and sound, in time order. Pick duration and aspect ratio in the studio controls rather than in the prompt. You can draft a prompt before signing in; final generation usually needs sign-in and credits.