prunaai/p-video-2-pro-replace

Pruna's person-swap model. Put the person from a picture into a source video, keep the camera, motion, and soundtrack, from 1 to 120 seconds and 480p to 4K.

462 runs
Playground API Examples README

p-video-2-pro-replace

Replace the person in a video with the person from a picture using Pruna’s p-video-2-pro-replace.

Give it a source video and one or two photos. It returns an MP4 at 24 fps where the new person is in the scene. The camera, framing, background, other people, timing, and motion stay as they were, and the source soundtrack is kept so the new person’s lips follow the original audio. Videos run from 1 to 120 seconds, at 480p, 768p, 1080p, 2k, or 4k. The output is as long as the source and keeps its aspect ratio.

prunaai/p-video-replace is a faster, cheaper 720p or 1080p swap. Use p-video-2-pro-replace when you need longer takes, up to 4k, or two people in one call.

How it works

The model takes:

  • A video that sets the motion, camera, scene, and audio.
  • One or two images of the person or people to put in.
  • An optional instruction_prompt that says who replaces whom.
  • A resolution, plus audio and safety settings.

It returns the same scene with the new person in it.

Give one picture to replace the main person. Give two pictures and an instruction to say who is who. With no instruction, the two pictures are matched left to right.

Inputs

  • video (required): source video. MP4, MOV, or WebM, 1 to 120 seconds, up to 8K. Longer videos are cut to 120 seconds. Aspect ratio must be between 1:4 and 4:1.
  • images (required): 1 or 2 pictures. One person, or two people with one picture each. Extra pictures are ignored.
  • instruction_prompt: who replaces whom, or other placement instructions. Leave empty for one person, or to match two pictures left to right.
  • resolution: 480p, 768p (the default), 1080p, 2k, or 4k. Aspect ratio is kept. 1080p, 2k, and 4k generate at 768p and then upscale.
  • save_audio: keep the source soundtrack in the output (default on).
  • ignore_audio: generate without listening to the source audio, so lips do not follow it. If save_audio is on, the soundtrack is still added to the output (default off).
  • enable_safety_filter: check your inputs against the content policy (default on). When it is off, the model still blocks sexual content involving minors and self-harm.
  • seed: set a seed to get the same result again.

Tips

  • Use a clear, well-lit photo that fits the voice on the source. The voice stays the source voice.
  • For one person on screen, leave instruction_prompt empty.
  • For two people, name left and right, or appearance. For a group, name the one person to change and what stays.
  • Use 480p for drafts, then rerun in 768p. Use 1080p, 2k, or 4k when you need a larger frame.
  • Keep save_audio on for dialogue and songs. Set ignore_audio when you only need the motion and do not want the lips to follow the source.
  • When the new person looks very similar to the person in the video, the model hides the source face so the new identity comes through. Expressions then come from the body pose and the audio.

Pricing

You pay per second of output video, based on resolution, counted in whole seconds. The output is as long as the source. The number of pictures, the instruction, and the audio settings do not change the price. 1080p, 2k, and 4k generate at 768p and upscale, so they take longer. A 10-second clip takes about 1.5 to 2 minutes at 768p, about 2.5 minutes at 1080p, and about 4 minutes at 4k.

Resources

Docs: https://docs.api.pruna.ai/guides/models/p-video-2-pro-replace

Model created
Model updated