MiniMax H3 Prompt Guide: Getting Motion You Can Control

MiniMax H3 generates high-quality 2K video with native stereo audio, but the gap between a generic result and a directed scene almost always comes down to how the motion is described. Vague prompts produce vague movement. The five elements below give a systematic way to write prompts that produce the motion you actually want.

For a full overview of H3's capabilities and pricing, see the MiniMax H3 model page.

Why H3 Needs Explicit Motion Prompting

H3's model architecture gives it strong spatial coherence and cinematic quality out of the box. What it doesn't do automatically is interpret "a woman walking" as a slow deliberate stride versus a rushed hurry. Both are "walking." The difference lives in the prompt.

The sections below cover five motion elements in order of impact. Each has a mini-example. The final section combines them into a complete prompt.

Element 1: Camera Instruction

Camera movement defines the viewer's relationship to the subject. Without it, H3 defaults to a static or gently floating frame.

What to specify: the camera move, its speed, and whether it tracks the subject.

• Static: static wide shot, camera holds position
• Push: slow dolly push toward subject
• Tracking: camera tracks subject from left, maintaining distance
• Orbit: slow arc around subject, 90 degrees, clockwise

Mini-example:

slow dolly push toward a woman reading at a café table, camera holds steady at chest level

Element 2: Subject Motion Specificity

Generic motion verbs leave duration, rhythm, and body mechanics unspecified. H3 responds well to biomechanical detail.

What to specify: speed, rhythm, which body parts move, how they move.

• Weak: a man walking down the street
• Strong: a man walks at a measured pace, arms swinging slightly, looking straight ahead, footsteps deliberate on wet pavement

Mini-example:

a dancer raises both arms slowly overhead, fingertips leading, weight shifting from right foot to left, movement fluid and unhurried

Element 3: Temporal Structure

Temporal structure describes how the action unfolds over the clip. H3 can execute multi-beat sequences when they are written as explicit transitions.

Pattern: begins with [state A], transitions to [action], culminating in [state B]

Mini-example:

begins with subject standing still at the window, transitions to turning and walking toward camera, culminating in a close-up as they reach forward

For clips under 6 seconds, keep to two beats. Longer clips can support three.

Element 4: Speed and Energy Language

Speed affects not just how fast things move but the emotional register of the shot. H3 picks up on energy descriptors.

Vocabulary that works:

• Slow: unhurried, languid, measured, drifting, deliberate
• Neutral: steady, continuous, even-paced
• Fast: rapid, sharp, sudden burst, immediate
• Weighted: with gravity, heavy footfall, controlled descent

Mini-example:

a flag unfurls with a sudden sharp snap, then settles into a slow, languid ripple

Element 5: Physical Forces

Gravity, wind, fluid dynamics, and fabric tension give motion its realism. Naming the force anchors H3's physics simulation.

Mini-example:

a silk dress falls with natural weight and drag, fabric catching air slightly before settling, subsurface scattering visible on skin in soft window light

Combine force description with lighting when the force itself has a visual effect (water, fire, fabric in wind all interact with light).

Complete Combined Prompt

Here is what all five elements look like assembled:

[Camera] Slow dolly push toward subject, camera holds at mid-torso level.[Subject + biomechanics] A woman in a linen jacket turns from a bookshelf, moving at a deliberate pace, shoulders relaxed, eyes tracking downward to the book in her hands.[Temporal] Begins with her back to camera, transitions to a three-quarter turn as she takes a step forward, culminating in a near-profile view as she opens the book.[Speed/energy] Movement unhurried and weighted, no sudden acceleration.[Physical forces] Jacket fabric has natural drape and slight swing as she turns, hair moving minimally in response to body rotation.Soft diffused interior light, 2K, cinematic.

This level of specificity is not always necessary for every shot, but it is the ceiling of what H3 can reliably reproduce from a prompt. Start with elements 1 and 2, add 3 through 5 when the default output doesn't match the intended feel.

Developer Notes

H3 is available through three endpoints on Siray: minimax/minimax-h3-t2v, minimax/minimax-h3-i2v, and minimax/minimax-h3-ref2v. All three share the same prompt interpretation layer, so these motion prompting techniques apply across modes.

The ref2v endpoint accepts reference images to anchor character appearance — useful when maintaining visual consistency across clips that each need different motion specifications.

Rate: $0.26/second for t2v and i2v at 2K. The ref2v rate adjusts based on reference image count (base $0.26/s plus an additional charge per reference image beyond five).

One API key, normalized request format. Swapping from t2v to i2v or ref2v is a one-line change to the endpoint parameter.

Summary

The five motion elements — camera instruction, subject biomechanics, temporal structure, speed language, and physical forces — give a replicable framework for getting consistent directed motion out of MiniMax H3. Use them as a checklist: start sparse and add specificity until the output matches what the shot requires.

Create your free Siray account and start generating with MiniMax H3 today.