Pruna P Video Replace: AI Video Character Replacement (Image-to-Video)
What is Pruna P Video Replace?
Pruna P Video Replace is a video-to-video character replacement model that swaps the on-screen person in an existing clip with an identity from one or more reference images. You provide a source video and one to three portrait references, and the model composites the new person into the shot while keeping the original motion, camera movement, lighting, background, and audio unchanged. Only the character changes.
Unlike a simple face swap, P Video Replace reads full-body posture and spatial placement across every frame, then renders the reference identity into that position for the whole clip. Built by Pruna AI on the LTX-2 architecture, it is tuned for speed and cost efficiency, making large-batch character-replacement workflows practical.
Key Features
- •Full-body identity replacement, not just face swap: posture, scale, and movement are preserved.
- •Accepts 1 to 3 identity reference images for stronger likeness across angles.
- •Keeps source motion, camera, lighting, background, and audio intact.
- •Output at 720p or 1080p, with target FPS of original, 24, or 48.
- •Optional instruction prompt to name who to replace in multi-person shots.
- •Turbo mode for faster drafts and a seed for reproducible results.
Best Use Cases
P Video Replace shines when you already have footage that works and simply need a different person in it. Common workflows include UGC ad variations, content localization, avatar or brand-mascot insertion, and recasting iconic film or meme shots. In our testing, a single clean portrait reference dropped a completely new identity into a talking-head clip while the sweater, room, lighting, and head motion stayed perfectly consistent. Because one source video is reusable, teams can generate many casts from the same plate.
Prompt Tips and Output Quality
Use a clean, well-lit, front-facing portrait as your reference; let the source video supply wardrobe and scene. For multi-person shots or held objects, use the instruction prompt to name the target by position ("the man on the left") and list everything to preserve. Draft in 720p, finalize in 1080p. Expect some identity drift on whip-pans or back-of-head framing, where the model has less of the face to anchor on; add angle references to stabilize.
FAQs
Is this a face swap? No. It replaces the full character, preserving body posture, motion, and spatial position, not just the face.
How many reference images can I use? One to three identity images. Add more angles when the subject turns or moves significantly.
Does it keep the original audio? Yes. Save audio is on by default, so the source soundtrack stays with the output.
What resolutions are supported? 720p and 1080p, with frame rates of original, 24, or 48.
Can I replace only one person in a crowded scene? Yes. Use the instruction prompt to name the target by position and list other people or objects to preserve.
Is output reproducible? Yes. Fix the seed to reproduce a run or vary it to explore alternatives.