How to Write Better MiniMax H3 Prompts

Use a structured MiniMax H3 prompt framework to control subject, action, camera, visual style, timing and native audio in a coherent AI video shot.

Try MiniMax H3 View pricing

Quick answer

A useful H3 prompt reads like compact direction for one shot—not a bag of disconnected visual adjectives.

Updated August 1, 2026Independent service · Not affiliated with MiniMax

At a glance

Choose the right starting point

Modes: Text · Image · First & Last Frame · Subject Reference
Image formats: JPEG · PNG · WebP
Resolution: Up to 2K, depending on available generation settings
Camera direction: Static · pan · push-in · tracking · orbit

Use a single primary camera move and one clear action before adding atmosphere, light and sound.

How to choose a generation mode

Match the mode to how much visual control you already have. Text alone is the fastest starting point; images and references trade flexibility for consistency.

Text: written creative brief
Image: starting still to animate
First & Last Frame: known visual endpoints
Subject Reference: recognizable person or character

Core MiniMax H3 prompt structure

Build the brief from the same spine: [subject + action] in [setting]. [camera move], [lighting and mood]. [what must remain consistent]. This gives the model a scene and a way to photograph it.

Lead with nouns and verbs
Keep the shot count low
Name one camera move
Finish with consistency constraints

Text-to-video prompts

Text mode has no visual plate, so define composition and action before style. Mention only setting details that affect the frame.

Who or what is on screen?
What happens during the clip?
Where is the camera?
What light and atmosphere shape the shot?

Image-to-video prompts

The image already defines subject and composition. Prompt motion, atmosphere and camera travel while naming the visual elements that must not change.

Preserve face, product, label or logo
Describe restrained subject and environment motion
Match the source lighting
End with a useful hold when needed

Subject-reference prompts

Use one clear reference and write action, wardrobe and environment separately from identity. Prefer a continuous action when likeness is important.

Use a well-lit unobstructed face
Do not request an identity change
Avoid multiple competing face crops
Keep camera movement predictable

First-and-last-frame prompts

The frames define the start and destination. The prompt should explain how the transformation happens, including camera, pacing and light continuity.

Keep endpoint framing compatible
Name one transformation
Avoid cuts between unrelated spaces
Use lighting language that fits both frames

Common prompt mistakes

Most weak prompts overload a short clip, choose the wrong mode, contradict a reference or use vague camera language.

Replace five scene changes with one readable action
Use frame pairs for controlled before-and-after stories
Animate what is already present in an image
Replace “cinematic” with a concrete camera instruction

A practical iteration loop

Generate a short draft, inspect subject, motion, framing and sound, then change one variable. Saving the original prompt and settings makes improvements traceable.

Draft short
Review the largest failure
Change one instruction or asset
Save the winning version

Common questions

Prompt guide FAQ

What should a MiniMax H3 prompt include?

Include the subject, action, setting, one camera move, lighting and mood. Add a clear consistency instruction for image, frame-pair or subject-reference workflows.

Can I use these prompts as written?

Yes, but results improve when you replace the subject and constraints with details from your own scene and reference assets.

How long should an H3 prompt be?

Long enough to define one coherent shot. Several concrete sentences normally work better than a long list of disconnected adjectives.

How do I keep a face consistent?

Choose Subject Reference, upload one clear face image and separate identity instructions from wardrobe, action and environment.

How do I prompt first and last frames?

Describe the motion connecting the supplied endpoints. Do not ask the model to invent a third unrelated composition.

Can camera movement be written in the prompt?

Yes. Use terms such as static wide, push-in, pan, tracking shot or orbit, preferably one primary move per clip.

Related resources

Copy prompt examples Open the step-by-step guide

Create your first H3 video

Start with text or an image, direct one clear shot, then preview and refine the result in your browser.

Open the generator