P Video Replace takes a video you already have and a reference photo of the person you want to appear in it, then outputs a new clip where that person is placed into every frame. The motion, timing, camera work, and background stay exactly as they were in the original. This solves a real production problem: you no longer need to reshoot an entire clip just to change who appears on screen. The model runs at roughly 3.58 seconds per second of output video, so a 10-second clip is ready in under a minute. You can supply up to three reference images to help the model match the subject's face from different angles, improving consistency across the whole clip. Resolution goes up to 1080p, and the original audio track can be preserved or stripped depending on what your project needs. P Video Replace fits naturally into a post-production workflow where the footage is already shot. Provide the source video, upload a clear photo of the replacement person, pick your resolution and frame rate, and let the model handle the rest. If the result needs a nudge, a short instruction prompt lets you describe exactly how the replacement person should appear in the scene.
P Video Replace does one specific thing: it takes a source video and swaps the person in it with someone from a reference photo, while leaving the original motion, camera angle, timing, and scene completely untouched. If you have footage where the subject needs to change without reshooting, this model handles that in a single pass. Picasso IA runs it directly in your browser with no software to install. Upload your clip, add a reference photo of the person you want in the scene, and get back a finished video where the new face and body move exactly as the original did.
Do I need programming skills or technical knowledge to use this? No, just open P Video Replace on Picasso IA, adjust the settings you want, and hit generate.
How long does it take to get results? The model processes roughly 3.5 seconds of compute time for every second of finished video. A 10-second clip typically comes back in around 35 seconds at 720p. Turbo mode cuts that down further if you are iterating quickly on a draft.
What output formats are supported? Results are delivered as .mp4 files. If your source video had audio, it is preserved in the output by default. You can also choose to ignore or strip audio before the final file is saved.
Can I use more than one reference photo? Yes, you can upload one to three identity reference images. Providing multiple photos of the same person from different angles tends to produce more consistent results across cuts and camera movements.
Can I adjust the output quality? You have two resolution options: 720p (roughly 1 megapixel) and 1080p (roughly 2 megapixels), with aspect ratio preserved from the source. Turbo mode is available when you need a faster preview at a slight quality trade-off.
Where can I use the finished videos? The output files are yours to use in any project, including client work, social content, internal presentations, or creative production. No watermarks are added to the final video.
What happens if the swap does not look right on the first try? Try adding a short instruction prompt to guide placement, or supply an additional reference image that better matches the angle or lighting of the source footage. You can also run the same inputs again with a different seed for a variation.
Everything this model can do for you
Every frame keeps its original camera angle, timing, and background with no distortion.
Feed multiple photos of the replacement person to improve face consistency across the whole clip.
Generate at up to 1080p resolution with the source aspect ratio preserved.
Keep the original soundtrack in the final video or strip it based on your production needs.
Cut processing time with a speed setting suited for quick drafts and fast iteration.
Choose between the original FPS, 24, or 48 to match your delivery target.
Direct how the replacement person is positioned in the scene with a short text description.
Pin the random seed to reproduce the same result reliably across multiple runs.
