Pruna P-Video 2 Pro Replace

Overview

P-Video 2 Pro Replace by Pruna edits an existing video by replacing the on-camera person with a new identity from photos. You provide a source clip plus 1–2 reference images, and the model returns the same shot with the new person inserted. It keeps the original camera movement, scene layout, and subject motion, so you don’t need to reshoot. This is useful when you want many variants of the same performance while preserving timing and audio.

What you can build

• UGC ad variants that keep the same take but swap the talent
• Localization versions that recast the same scene for different regions
• Creator or avatar inserts in existing b-roll while keeping camera motion intact
• Two-person recasts for dialogue clips when you have one photo per person
• Character swaps for short-form content while preserving the original pacing

Inputs

• One source video file in MP4, MOV, or WebM format. It must be 1–120 seconds long. If it’s longer, the model uses the first 120 seconds. The input can be up to 8K. The aspect ratio must be between 1:4 and 4:1.
• One or two reference images in JPG, JPEG, PNG, or WebP format. Use one image per person you want to replace. With two images, the model typically matches people from left to right unless you say otherwise.
• An optional instruction sentence that says who replaces whom, plus any placement rules. This helps when there are multiple people on screen.
• An output resolution choice: 480p, 768p, 1080p, 2K, or 4K. The output keeps the source video’s aspect ratio.
• An optional toggle to keep the original soundtrack in the exported result.
• An optional toggle to ignore the source audio during generation. This can reduce audio-driven mouth movement. You can still keep the original soundtrack in the final export.
• An optional safety filter toggle that checks inputs against the provider’s content policy.
• An optional reproducibility number (integer) to help you re-run the same setup with similar results.

Outputs

• One MP4 video.
• The output length matches the source clip length, up to 120 seconds.
• Frame rate is 24 fps.
• The shot keeps the original camera motion and scene timing, while replacing the person’s identity using the provided photo(s).
• If you keep audio, the output MP4 includes the original soundtrack from the source video.
• For 1080p, 2K, and 4K selections, the result may be produced by upscaling from a lower internal resolution.

Recommended settings

• Use 480p when you want quick iterations and smaller files.
• Use 768p as a strong default when you want more detail without relying on upscale-only tiers.
• Use 1080p, 2K, or 4K when you need larger delivery formats, knowing these can be upscaled.
• Keep the original soundtrack on when timing, dialogue, or ambience must stay unchanged.
• Set a reproducibility number when you need consistent A/B tests across prompt tweaks.

Limitations

• The model only accepts source clips between 1 and 120 seconds. Longer clips are trimmed to the first 120 seconds.
• You can replace one or two people based on one or two reference photos.
• Identity matching with two photos is typically left-to-right unless your instruction says otherwise.
• Heavy motion blur, low light, strong occlusions, or tiny faces can reduce identity fidelity.
• Low-quality inputs can cause artifacts. This includes highly compressed video, noisy footage, or inconsistent lighting.
• Extreme aspect ratios outside 1:4 to 4:1 aren’t supported.

Safety & compliance

• Only use footage and reference photos you have rights to use.
• Get clear consent when replacing a real person’s identity in a real video.
• Safety filtering may block disallowed content based on the provider’s policy.
• If safety filtering is turned off, some categories may still be blocked by always-on enforcement.