MiniMax quietly released MiniMax H3, also called Hailuo 3.0, on July 31, marking the third generation of its Hailuo video line. The company describes the model as “omni-modal,” meaning it reads text, images, video, and audio together in one pass. Its specifications include native 2K (2560×1440), 24fps, 4 to 15 seconds per clip, stretchable to about 30 with MiniMax’s Extend tool, native stereo audio generated alongside the video, and six aspect ratio options from 21:9 down to 9:16. The model’s Omni-Reference feature allows users to feed up to nine images, three video clips, and three audio clips in the same request, treated as one combined context, which helps maintain character or product consistency. The standout feature is editing: MiniMax H3 can take existing footage and change a specific part—such as swapping an object, changing the background, adjusting lighting, altering movement, or replacing dialogue with the mouth re-forming around new words—while leaving the rest of the shot unchanged. The article notes that reference-heavy generations require trial and error, and results vary based on source material cleanliness.