prunaai/p-video-replace

p-video-replace swaps the person in a video with one from a reference image, keeping motion, timing, camera, and scene exactly as they were. 3.58s per 1s of video generated.

1.2K runs
Playground API Examples README

p-video-replace

p-video-replace takes a source video and 1-3 reference images, and produces a new video where the person on camera is swapped to match the reference, while motion, timing, camera, and scene stay exactly as they were in the source.

It’s optimized for speed and cost: about 3.58 seconds of generation per 1 second of output video. That’s 3.7x faster than Wan 2.2 Animate Replace and 17x faster than Luma Ray3 Modify. It’s well suited to high-volume creative workflows like UGC ad variations, content localization, avatar and mascot insertion, and game cinematics.

How it works

Give the model two things:

  1. A source video (video) — an MP4 that defines the motion, timing, camera, and scene.
  2. Reference image(s) (images) — 1 to 3 images of the identity (or identities) you want in the video.

The model returns an MP4 of the same scene with the original person replaced by the reference. Source audio is muxed back in by default.

Inputs to know about

  • instruction_prompt — free-form text that gives extra direction on how to place the people from the reference images into the scene. Leave blank for default behavior.
  • resolution — 720p (~1 MP) or 1080p (~2 MP). Aspect ratio is preserved from the source video.
  • target_fps — original, 24, or 48. Default keeps the source video’s FPS.
  • save_audio — when on, the source audio is muxed into the output.
  • ignore_audio — when on, the model doesn’t condition on the source audio during generation. If save_audio is also on, the source audio is still muxed in at the end.
  • turbo — faster generation for slightly lower quality.
  • seed — set for reproducible results.

Tips for good results

  • Clear, front-facing reference images. Heavy angles or occlusion hurt identity preservation.
  • Strong, well-lit motion in the source video. Subtle or occluded movement is harder to transfer cleanly.
  • Pick the right resolution. 720p is half the price of 1080p and works well for most cases. Use 1080p when you need extra detail.
  • Match aspect ratios. The output uses the source video’s aspect ratio, so frame the source the way you want the result framed.

Pricing

Billing is per second of output video, based on resolution:

  • 720p: $0.03 per second
  • 1080p: $0.06 per second

A 10-second clip at 720p costs $0.30.

Launch discount

From Thursday, June 4 at 16:00 CEST through Sunday, June 8 at 23:59 CEST, p-video-replace is 70% off:

  • 720p: $0.009 per second
  • 1080p: $0.018 per second

What you can build

  • UGC ad variations — keep the shot, swap the talent across dozens of variants.
  • Content localization — recast the same scene with different talent for different markets.
  • Avatar and mascot insertion — drop a brand character into existing footage.
  • Game cinematics — swap actors into pre-shot performance capture.
  • Creator workflows — reuse one strong performance with different identities on camera.

Try it on the Replicate playground.

Model created
Model updated