Pruna P Video Edit — Instruction-Based AI Video Editing (Video-to-Video)
What is Pruna P Video Edit?
Pruna P Video Edit is an instruction-based video-to-video model from Pruna AI. You upload a source clip, describe the change in plain language, and get back an edited video that keeps the original motion, framing, and scene structure. Unlike text-to-video generation, it never re-invents your shot — it edits the footage you already have. You can guide the result with up to four optional reference images, so the model matches a specific garment, product, character, or environment instead of guessing. Source videos can be up to 15 seconds long, and the source audio track is preserved by default. Built on Pruna's performance stack, the model is tuned for the Pareto front of speed, cost, and quality, with a draft mode for fast previews before you commit to a full-quality render.
Key Features
- •Instruction-driven edits: change an attribute, swap an object, or replace the environment from a single text prompt.
- •Motion and structure preservation: the original camera work, timing, and subject movement stay intact.
- •Up to four reference images: guide identity, wardrobe, products, or style precisely.
- •Draft mode: render a faster, cheaper low-quality preview, then switch off for the final edit.
- •Prompt upsampling: automatically expands your instruction into a richer edit prompt, or disable it for literal control.
- •Audio passthrough: keep the source soundtrack, or turn it off for a silent clip.
- •Reproducible seeds: fix a seed to iterate on the same result.
Best Use Cases
Pruna P Video Edit fits production workflows where reshooting is expensive. Swap a creator's outfit or a product in-hand across UGC ad variations without filming again. Change wardrobe, replace a background or sky, restyle a scene, or alter an object's color and material while the performance stays untouched. Marketing and MCN teams use it to localize content and spin one recording into many variants across formats. Because motion and audio survive the edit, it slots cleanly into pipelines alongside Pruna's other P-Video models — generate a base clip, then edit it here. The confirmed example — replacing a woman's white t-shirt with a red leather biker jacket from a reference image while keeping her pose and studio background — shows the model's core strength: targeted, reference-guided change with everything else preserved.
Prompt Tips and Output Quality
Write one explicit instruction and name both what to change and what to keep, for example "replace the sky with a sunset, keep the camera motion." Avoid pronouns; reference subjects specifically ("the woman in the white shirt"). Add reference images when you need an exact garment, product, or identity match. Use draft mode to validate the edit fast, then rerun full quality with the same seed for a clean final render. For literal wording control, disable prompt upsampling.
FAQs
How is P Video Edit different from P-Video-Replace? P-Video-Replace focuses on swapping characters or people using identity references. P Video Edit is broader instruction-based editing — outfits, objects, backgrounds, styles, and attributes — from a text prompt.
What is the maximum input video length? Source videos can be up to 15 seconds.
Do I need reference images? No. Images are optional; add up to four when you want the edit to match a specific garment, product, character, or scene.
Does it keep the original audio? Yes. The source audio track is saved by default, and you can turn it off.
What does draft mode do? It renders a faster, lower-quality preview at a cheaper per-second rate, ideal for iterating before the final edit.
Can I reproduce a result? Yes. Set a fixed seed between -1 and 2147483647; use -1 for a fresh result each run.