Upload a source clip as a multimodal video reference, then describe the new subject, setting, style, and sound you want. MiniMax H3 uses the reference as creative direction for motion and shot rhythm while generating a new video; it does not perform a frame-for-frame restyle.
Add up to 12 references: 9 images, 3 videos, and 3 audio clips. Each video or audio clip must be 2β15 seconds; videos and audio each have a separate 15-second total limit. Refer to them as Image 1, Video 1, and Audio 1 in your prompt.
Preview video - Generate your own video above
This H3 Max parkour example shows the kind of action arc, tracking movement, and pacing that can make a useful source reference. Upload your own clip above and tell MiniMax H3 what to retain as direction and what to reinvent.
The current workflow treats your video as a multimodal reference. A precise prompt separates the motion language to borrow from the visual details to replace.
Use a clip with clear subject movement, camera travel, and timing so the intended action pattern is easy to identify.
Name the camera angle, speed, rhythm, and action you want carried into the newly generated scene.
MiniMax H3 creates a new video from your instructions. It is not a pixel-level filter, rotoscope, or guaranteed one-to-one motion transfer.
Use one focused source clip and a prompt that distinguishes reference motion from the new creative treatment.
Choose a short clip with readable action and camera movement. The multimodal reference form accepts video alongside optional image or audio references.
Call the upload Video 1. State which movement, framing, tempo, or transitions to follow, then describe the new subject, environment, style, and sound.
Set the duration, aspect ratio, and resolution, review the credit estimate, and generate. Compare the result with the source's motion intent rather than expecting identical frames.
Reference-driven generation is most useful when the source communicates a motion concept and the prompt supplies a clearly different visual treatment.
Carry the feel of a push-in, orbit, tracking shot, or handheld move into a new environment.
Use a blocking clip to explore a new character, location, and cinematic treatment before production.
Reference the timing of a reveal or turntable shot while generating a fresh product scene.
Reimagine the broad action and rhythm in a different medium while describing the desired look explicitly.
Develop alternate settings or subjects around a motion pattern that already works for a short-form clip.
Combine a motion clip with optional image or audio references when each asset has a distinct role in the prompt.
Understand what the current reference workflow can do, how to prompt it, and where generated results may differ from the source.
Not currently. This page uses MiniMax H3's multimodal reference mode, which accepts source videos as creative context. It is the closest available workflow for guiding new generation with motion, camera, and pacing from a clip.
No. The source guides the generation, but exact poses, timing, camera paths, and frames are not guaranteed. Describe the motion traits that matter most and expect a newly generated interpretation.
Refer to the upload as Video 1, identify the action and camera rhythm to follow, and separately describe the new subject, scene, style, lighting, and sound. Avoid conflicting directions.
Yes. Multimodal reference mode can combine video with image and audio references within the form limits. Assign each asset a clear purpose so the prompt does not leave their roles ambiguous.
No. It generates a new clip and does not edit the source timeline or apply a deterministic filter to every frame. Identity, choreography, composition, and duration can change.
Choose a short, visually readable clip with one dominant action, stable subject visibility, and intentional camera movement. Complex cuts, occlusion, and several competing actions make the reference harder to interpret.
Upload a motion reference, describe what to preserve and reinvent, and generate a new MiniMax H3 video.
