MiniMax H3 one model, every reference

MiniMax H3, also known as Hailuo 03, is a multimodal AI video model that creates video and native stereo audio from text, frames, or mixed image, video, and audio references.

Text mode creates a video from a prompt. Frames mode animates a required first frame and an optional last frame. References mode combines up to 9 images, 3 videos, and 3 audio files.

You can generate clips from 4 to 15 seconds at 768P or 2K. Text and reference modes offer multiple aspect ratios; frame mode follows the uploaded image composition.

768P costs 32 credits per second and 2K costs 51 credits per second. Reference-video seconds are added to the generated duration. The first 5 reference images are free; each additional image costs 16 credits. Audio references are free.

Yes. Upload a first frame to define the opening composition and optionally add a last frame to guide the final composition and transition.

Give every asset a clear role: identify which image controls identity or design, which video supplies movement or camera behavior, and which audio guides voice, music, or rhythm.

Yes. MiniMax H3 generates native stereo audio with the video. Describe dialogue, ambience, music, and sound effects directly in your prompt; uploaded audio references do not add credit cost.