MiniMax H3 and Hailuo workflows: multimodal references and motion
Assign distinct roles to image, video and audio references without mixing the contracts of different workflows.
What the H3 announcement says
MiniMax's July 31, 2026 H3 announcement describes unified text, image, video and audio context, native stereo sound, up to 15 seconds and 2K output. Those are vendor statements, not a promise of every specification in every integration.
HUUUIAI has both Hailuo 2.3 and MiniMax H3 workflows. A single-image Hailuo task should not be given an H3 multimodal input contract. Announced plans to release weights also do not prove the weights are already available.
Give each modality a clear role
- Use an image for subject appearance.
- Use a video for movement or camera rhythm.
- Use authorized audio for the intended sound or timing.
- Explain the relationship between those references in words.
A motion-reference example
Use the sneaker in the image as the subject. Borrow only the slow orbiting camera movement from the reference video. Keep the shoe stationary, preserving its logo and sole structure. Use subtle ambience with no vocals.
Check what was actually transferred
Motion references can accidentally introduce their background or objects. Inspect whether only the requested movement was used. Start with a short clip, remove conflicting references one at a time, and confirm live account settings before expanding the task.
Sources and verification
Checked on the dates below. Prompts are editorial examples, not measured performance results. Model availability and parameters in HUUUIAI depend on the current service.