MiniMax H3 prompt generator

The three-part H3 prompt

• Reference asset instructions: name each upload by type and order (@Image 1, @Video 2, @Audio 1) and say what it is for, such as character, environment, keyframe, voice, style or camera-motion reference. Skip this part without uploads.
• Core concept: one sentence that covers subject, location, action, genre and style, any special camera treatment, camera movement and editing style.
• Shot by shot: a timeline such as “0–3 seconds: wide shot …” or “Shot 1 — …”, each with shot size, what is visible, the camera move, the action, exact dialogue and sound.

The sample scene as an H3 prompt:

A young woman in a red raincoat stops and looks up as the rain starts in a neon-lit Tokyo side street at night, cinematic 35mm film style, soft neon lighting, quiet, nostalgic mood, the camera pushes in.
0–3 seconds: Wide shot, she walks toward the camera under the neon signs.
3–6 seconds: Close-up, raindrops land on her face and she smiles.
The woman says: "It finally came."
Sound: rain on the pavement, distant traffic.
No subtitles or on-screen text.
Non-diegetic music: N/A

Cuts are the default

H3 cuts between shots by default. For one unbroken take, the cookbook says to drop the Shot N structure, keep the action in one paragraph and say “one continuous take”. Tick “One continuous take” in the generator and it does exactly that.

Camera moves in plain words

Describe camera movement explicitly. For an orbit, the cookbook recommends building it from simple moves, truck left + pan right, instead of writing “orbiting camera”; the generator’s orbit option does this. MiniMax’s guides also show short cues such as [pan], [zoom] or [static] placed right after a description, but there is no published list of camera commands for H3.

The older Hailuo 2.3 and 02 models do have a list of 15 bracket commands, including [Truck left], [Pedestal up], [Tilt down] and [Shake], where several commands in one bracket run at the same time. Because that list is documented for the older models, the generator writes H3 moves in plain words.

Exclusions and dialogue

There is no negative_prompt field. State exclusions at the end in plain language; to drop background music, end with “Non-diegetic music: N/A”. Put exact dialogue and any exact on-screen text in quotes. Keep the amount of dialogue in proportion to the shot: the cookbook says many lip-sync problems come from squeezing a long line into a short shot. Prefer concrete, visible descriptions over metaphors.

MiniMax H3 at a glance

For text-to-video the aspect ratio must be set explicitly; with a first-frame image the output follows the image. H3 Max makes 5–15 second clips at 480P or 768P. Every generation includes native stereo audio.

Related tools

• MiniMax H3 price per secondMiniMax H3 price per second: $0.08 at 768P and $0.13 at 2K with native audio; H3 Max is $0.05 at 480P. Compare fal, Replicate, Runware, WaveSpeed, OpenRouter.
• Seedance prompt generatorFree Seedance prompt generator. Builds Seedance 2.5 prompts with timestamps and ByteDance’s sound tags, or Seedance 2.0 prompts in Shot 1 / Shot 2 format.
• Kling prompt generatorFree Kling prompt generator. Writes Kling 3.0 prompts in Kling’s own order and splits them into up to six shots with the official “shot n, m, words;” syntax.
• Wan 3.0 prompt generatorBuild Wan 3.0 prompts in Alibaba’s recommended order: overall description, timed shots, dialogue, sound, style and a negative list. Free, no sign-up.