GPT IMAGE 2 FRAME · MINIMAX H3 RE-PERFORMANCE · ORIGINAL AUDIO

Swap the performers, keep the performance.

Redraw the first frame with your new characters, then H3 re-performs the footage clip by clip: every clip gets the exact source motion, continues from the previous frame, is checked by a vision model, and the untouched original soundtrack goes back on top.

Need frame-exact motion? Try the Exact Motion lane Want the most faithful performance with face-free characters? Try Reference Video Studio

EXAMPLE OUTPUT

Source rappers → Elon Musk + Optimus

SAMPLE SWAPREADY

Same moves, same track1280×720 · 24 fps

01 / SOURCE

Source video

SOURCE PREVIEWLOADED

REFERENCE FRAMEEXTRACTED

MP4 · WebM · MOV · 5–60 sFrame extracted automatically

Or paste a public video URLFrame image URL

02 / FRAME

Redraw the first frame with GPT Image 2

Character swap promptSign in to generateOr paste an image URL

BEFORESOURCE FRAME

AFTERREADY

03 / VIDEO

Re-perform the video with MiniMax H3

Re-performance prompt

Experimental: redraw one frame per detected shot so close-ups and angles follow each cut (+$0.30 per extra shot; the set can drift between shots)Also give H3 the song as an audio reference (experimental lip-sync guidance)

Estimate appears when signed in768P · $0.20/s

SWAPPED VIDEOWAITING

No video yet. Finish steps 1–2, then render.

Build this workflow into your product

One JSON request returns an async job that ends in a re-performed music video with swapped characters on the untouched original soundtrack.

API reference