MiniMax Launches Hailuo 3.0 (H3), AI Video Model With Native Synced Audio and Multi-Reference Character Consistency
AI video model releases
MiniMax Launches Hailuo 3.0 (H3), AI Video Model With Native Synced Audio and Multi-Reference Character Consistency
MiniMax H3 (Hailuo 3.0), launched July 31, 2026, generates native 2K video at 24fps with built-in synchronized audio including dialogue, sound effects, and ambient sound, and can produce up to 15 seconds of continuous footage per generation.
Omni-Reference is a new control system that accepts up to 9 reference images, 3 video clips, and 3 audio clips as unified context to keep a character's appearance, motion, and voice consistent across shots.
Audio is generated in the same pass as video rather than added afterward, which turns the output into a near-editable first cut instead of a silent visual asset, cutting post-production work for short films, serialized content, and brand campaigns.
The model is the third generation in MiniMax's Hailuo video family and is positioned to compete on two long-standing weak points of AI video generators: character consistency and audio-visual sync.
cho.sh