Create videos with MiniMax H3
MiniMax H3 creates short videos from a written scene, opening and closing frames, or a set of reference assets. Choose 768p or 2K and a duration of 4–15 whole seconds. Every result includes a generated soundtrack.
Describe the motion and composition
Specify the subject, setting, action and camera movement. Text and reference modes support 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. Prompts can contain up to 7,000 characters.
Choose frames or reference assets
Use a first frame alone or add a last frame to guide the ending. The video follows the first frame’s shape; frame inputs cannot be combined with references. Reference mode accepts up to 9 images, 3 videos and 3 audio clips, with 12 assets in total. Video and audio clips must each last 2–15 seconds, with a 15-second total per media kind. Audio needs at least one image or video.
Refer to assets by name
Type @ in the prompt to select an uploaded reference. Image, video and audio references have separate numbering: @Image1, @Video1 and @Audio1. Explain which subject, movement or beat each reference supplies. Reference-video seconds are charged along with output seconds; check the credit estimate before generating. Seed, draft and silent output are unavailable. Review faces, details, motion and sound before publishing.
Frequently asked questions
Can I generate a silent video?
MiniMax H3 always generates a soundtrack. Remove the audio later in an editor if you need a silent result.
Can I combine first and last frames with references?
No. Choose frame inputs or reference mode for each generation.