Practical workflow · Updated August 5, 2026

MiniMax H3 Reference to Video: Direct Identity, Motion, Camera, and Sound

Reference mode is where Hailuo 03 becomes more than a prompt box. Give every image, video, and audio file one clear job, then describe how those jobs combine in the target shot.

Open Reference to Video

Image references

Use images to anchor identity, wardrobe, product design, environment, color, or visual style. Choose a clean primary identity image before adding supporting angles.

Up to 9 images · JPEG, PNG, or WebP · 30 MB each

Video references

Use video to communicate motion, performance, camera behavior, editing rhythm, or a shot that should be transformed while preserving its timing.

Up to 3 videos · MP4 or MOV · 50 MB each

Audio references

Use audio to guide voice, ambience, effects, music, or rhythm. Explain whether the sound is diegetic, off-screen, or audience-only.

Up to 3 audio files · MP3 or WAV · 15 MB each

The role-first Reference prompt structure

MiniMax and ComfyUI guidance both emphasize explicit reference labels and responsibilities. Start with the target result, assign each source, then describe the timeline and what must remain unchanged.

TARGET
A 10-second cinematic product film with native stereo sound.

REFERENCE ROLES
<Picture 1>: preserve the exact product shape, material, and logo placement.
<Video 1>: use only the slow orbital camera movement and pacing.
<Audio 1>: use the mechanical click as the transition cue.

TIMELINE
0-3s: macro detail, shallow focus, quiet room tone.
3-7s: follow <Video 1> camera motion; product remains unchanged.
7-10s: reveal the full product as <Audio 1> lands.

PRESERVE
Do not redesign the product, add text, or change its proportions.

How to reduce identity and detail loss

  1. 1Match the target aspect ratio to the dominant reference instead of forcing a crop that removes identity cues.
  2. 2Use one primary identity source. Add extra images only when they provide a missing angle or detail.
  3. 3Separate identity, motion, style, and audio roles. Do not ask every reference to control the entire scene.
  4. 4State what must stay unchanged: face shape, wardrobe, product proportions, typography, or background geometry.
  5. 5Test the shortest useful duration before spending more credits on a longer shot.

Reference workflow checklist

Every uploaded reference is mentioned in the prompt.
The dominant image and target aspect ratio agree.
Identity and motion come from separate, named sources.
Dialogue, ambience, effects, and music have clear roles.
Preservation constraints are explicit and observable.
The final result will be reviewed once with sound on.

Primary references

Official workflow

ComfyUI MiniMax H3 Reference to Video

Reference tags, role assignment, supported input counts, reference sizing, and the dedicated ref2va workflow.

Hosted provider contract

Kie MiniMax H3 inputs and limits

Current file formats, upload limits, durations, ratios, resolution settings, and multimodal request fields.

Need a broader prompt framework? Read the MiniMax H3 prompt guide or inspect the attributed H3 examples.

Create in your browser

Turn your direction into a MiniMax H3 video

Use Text, Image, or Reference to Video with clips up to 2K, native stereo sound, and private creation history.

Create with MiniMax H3 View pricing