H3 Max Prompt Guide: How to Write MiniMax H3 Video Prompts

A good H3 Max prompt describes one shot in time order: who or what is on screen, what happens, how the camera moves, and what you hear. On this site, plain sentences are enough — every prompt is automatically rewritten into H3's detailed format before generation. When you want precise control over cuts, dialogue or music, use the structure from MiniMax's official prompt guide, explained below.

Looking for ready-made examples instead? See 30 copy-ready H3 Max prompts.

The simple formula

Write one to four sentences in this order:

1. Subject — who or what is on screen, with the details that matter (age, clothing, color, material).
2. Action — what happens from start to end of the clip.
3. Camera — one clear movement, such as a slow push-in or a tracking shot.
4. Style and light — live-action, 3D animation, watercolor; golden hour, neon, overcast.
5. Sound — ambience, action sounds and music. H3 generates audio with the picture, so describe it.

Live-action, cinematic. A cyclist in a yellow rain jacket rides through a narrow cobblestone street at dusk, splashing through puddles. The camera tracks alongside her at slow speed. Wet tires hiss on stone, a church bell rings in the distance, soft piano plays underneath.

How prompts are processed on this site

H3 expects a long, structured description of the clip. MiniMax's hosted service and fal.ai both rewrite short prompts into that kind of description before generation. On h3maxvideo.net, every prompt goes through fal's balanced prompt expansion, which takes about a second.

What this means for you:

• Short, plain prompts work. The rewriter adds detail that is consistent with your text.
• Specific beats vague. The rewriter can only expand what you give it. "A dog runs on a beach" leaves the breed, light, camera and sound to chance.
• Exact words survive best when quoted. Put dialogue and on-screen text in quotation marks so they are not paraphrased.
• Up to 2,000 characters are accepted in our generator.

The official MiniMax H3 prompt structure

MiniMax's video prompt writing guide describes the format H3 expects after rewriting. It has three fields:

A short example in this structure:

integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium shot frames an elderly watchmaker at a cluttered workbench under a single desk lamp. The camera pushes in with small amplitude at slow speed as he lifts a tiny gear with tweezers and fits it into an open pocket watch. The watchmaker with a low, gentle voice (S1) says: <d>[English] Almost there.</d> [Shot 2] At 00:04.000, the camera cuts to a close-up of the watch hands starting to move.

overall_soundscape: A quiet workshop hum with the faint ticking of several clocks. Metal tweezers click softly against brass, followed by a small mechanical snap.

non_diegetic_music: Slow solo cello with long sustained notes, fading out at the end.

You can paste a prompt like this into the generator. The rewriter may keep it or adjust it, so treat it as precise guidance, not a guarantee.

Camera moves: type, amplitude, speed

The official guide describes a camera move with up to three parts: the motion type, an optional amplitude (with small amplitude / with large amplitude) and an optional speed (at slow speed / at fast speed). Leave amplitude and speed out when a normal move is fine.

Write the move as part of a sentence, not as a list of labels at the end:

The camera pans right with large amplitude at fast speed, revealing a crowded night market.

Shots and cuts

• Most 5–10 second clips need one shot. Use camera motion for small changes in distance or angle.
• Cut only for new information — a new place, subject, viewpoint or moment.
• Number the shots and time the cuts. The first shot has no timestamp; later shots start with the cut time: [Shot 2] At 00:03.500, the camera cuts to …. Cut times must increase and stay within the clip length.

Dialogue, voiceover and on-screen text

• Give each speaker a stable ID such as (S1) and (S2), and describe the voice once: age, gender, pitch, speed, accent.
• Wrap the spoken words in <d>[English] …</d>, with the language tag first. The words inside are kept exactly as written; do not translate them.
• Voiceover: write says in an off-screen voiceover and add that the character's lips stay closed.
• On-screen text (signs, labels, subtitles) goes in double quotation marks and is kept verbatim: a neon sign reading "OPEN 24 HOURS".

The barista with a bright, fast voice (S1) says: <d>[English] Oat latte for Sam!</d>

Sound and music

• Sound effects and ambience belong in the description or the soundscape: rain on glass, footsteps on gravel, a door slamming.
• Music the characters can hear (a street musician, a radio) is part of the scene, so describe it in the main description.
• Background score is described by instruments and tempo, not by mood words. Write "slow piano with soft strings" rather than "emotional, inspiring music".
• Silence: say so explicitly if you want no music or no sound.

Prompts for image to video

When you animate an image, the picture already defines the look. Spend your words on motion:

• First frame only: start from what is in the image, then describe the action as it begins, develops and ends. Keep the person, clothing, colors and layout consistent with the photo.
• First and last frame: describe the path between the two images — how the subject moves, how the pose and composition change — and keep it to one continuous shot. Do not describe the two images again.

The woman in the photo turns from the window, picks up the coffee cup from the table and walks toward the camera, smiling. The camera pulls out slowly to keep her in frame. Soft footsteps on wooden floor, a cup clinks on a saucer.

Common mistakes

Test cheaply

Draft a prompt at 480P, 5 seconds with H3 Max Turbo — it costs half the credits of H3 Max. When the motion and sound work, render the final version with H3 Max at 768P or 1080P. See pricing for credit costs.

FAQ

Do I need to use the official structured format? No. On this site every prompt is expanded automatically, so natural sentences work well. The structure helps when you need exact cuts, dialogue or music.

How long can an H3 Max prompt be? Our generator accepts up to 2,000 characters. One to four clear sentences is usually enough for a single shot.

Can I write prompts in Japanese or German? Yes. English prompts tend to give the most predictable results, but dialogue should be written in the language you want it spoken in, with the matching language tag.

Why does my video ignore part of my prompt? Usually the prompt asks for too much in one clip. Cut it down to one shot and one main action, or make the clip longer.

Where can I find example prompts? In our H3 Max prompt library, and in the step-by-step H3 Max tutorial.

Sources: MiniMax H3 video prompt writing guide and the fal.ai H3 Max API documentation. Last updated: October 1, 2026.

h3maxvideo.net is independent and not affiliated with MiniMax or fal.ai.