Video to video AI for a different visual direction

Video-to-video generation starts with footage rather than a blank prompt. In Genjutsu AI, MiniMax H3 uses your reference video as visual and motion context, then generates a new clip from your instructions. You can explore another setting, character, or style without filming a new reference performance.

How to use this workflow

1. Upload a 2–15 second MP4 or MOV reference, up to 50 MB. Use H.264 or H.265 footage with readable movement and a standard frame rate.
2. Describe the new scene and explicitly identify what Video 1 should guide. An optional reference image can anchor the intended look.
3. Choose a 5, 10, or 15 second output and either 480P or 768P. Portrait and landscape output are available independently of the input framing.
4. Review the regenerated clip before using it. Keep your original footage: generated output is a separate interpretation.

A prompt to start with

Follow the slow camera orbit and subject movement in Video 1. Rebuild the scene as a hand-painted fantasy garden, with warm morning light and a soft watercolor texture. Keep one continuous shot. Sound: gentle wind and birds.

What to expect

This workflow regenerates a clip. It does not edit a masked area of the original file, keep every untouched pixel, or guarantee the source soundtrack. Audio is generated by the model.

Updated October 6, 2026 · Genjutsu AI