MiniMax H3 is an omnimodal generative model that understands text, images, video, and audio together to create videos with stereo sound. Up to 2K resolution, 4–15 seconds.