Keep the take you already shot and change the one thing that is wrong in it: the product on the table, the jacket, the sky behind the actor. Describe the swap, attach a reference photo if the look must match, and the motion, camera and audio carry on as filmed.
A clip is almost right and one thing in it is wrong: the mug carries the old design, the jacket is last season's, the sky behind the actor is flat grey. Reshooting means the location, the light and the take all over again. Replacing the object with AI changes that one thing and lets the rest of the shot run as filmed. Here is how it works on Picasso IA, which models do it, and where it stops being reliable.
Most clips that need a fix do not need a new shoot. They need one element changed. A product video where the packaging was updated after filming, a lifestyle clip where the outfit clashes with the brand colors, a travel shot with a dull sky, a kitchen scene with a prop you do not have the rights to show. In a traditional editor each of those means tracking the object frame by frame and painting a replacement in by hand, which is why people usually give up and reshoot or live with the flaw.
Video editing models work differently. You give the model the clip and describe the change, often with a reference photo of what should appear instead, and it rewrites the affected pixels across the whole clip while keeping the camera move and the timing. Picasso IA lists 502 models, 28 of them in the video editing category, and they all run in the toolkit, so you can try a swap on a short clip and see whether it holds before you invest in anything longer.
These three edits sound alike and use different tools, so it pays to name the one you want before you open anything.
Replacing puts something new where something old was: a different bottle on the same table, a red jacket where the grey one was, a sunset where the overcast sky was. Removing makes the object disappear and fills the gap with background, which is a separate job covered in remove objects from videos. Restyling changes the look of the whole picture, such as turning live footage into a painted or animated style, and that is the territory of turning videos into anime rather than a targeted swap.
One more distinction matters on Picasso IA. Two models carry the word replace in their name, P Video Replace and Wan 2.2 Animate Replace, and both swap a person or character, not a prop. For a product, a garment or a sky you want a prompt-driven editor instead. If you are swapping a person, use only your own likeness or a fictional character, never a real identifiable person who has not agreed to it.
The catalog documents different limits for each editor, so match the model to the clip and the kind of change before you generate.
| Model | Best for | What you give it | Documented limits |
|---|---|---|---|
| prunaai/p-video-edit | Products, garments, props, backgrounds | Text prompt plus up to 4 reference images | Clips up to 15 seconds, draft mode for previews |
| runwayml/aleph-2 | A change that must hold across several shots | Text prompt plus up to 5 keyframe images | Clips from 2 to 30 seconds, under 16 MB |
| prunaai/p-video-replace | Putting a different person into the same motion | The video plus 1 to 3 photos of the person | 720p or 1080p output |
| wan-video/wan-2.2-animate-replace | Swapping a character, such as a mascot or illustrated figure | The video plus one character image | 720 or 480 processing |
For most object swaps the starting point is P Video Edit. It takes a plain instruction like change the sky to sunset or make the jacket red, accepts reference images when the new object has to match something exact, and has a draft mode that renders a fast lower quality preview first. Aleph 2 earns its place when the clip has cuts, because it applies one instruction across a multi-shot sequence and lets you pin keyframe images at the first frame, the last frame or any timestamp. Try the two on the same clip and keep whichever respects the shot better.
This walkthrough runs entirely in the toolkit and assumes you have a short clip and, ideally, a clear photo of the new object.
Each run costs credits and the price depends on the model and the settings, so the pricing page is the place to check numbers, and new accounts start with free credits to test a swap.
The prompt does two jobs: it says what changes and it says what must stay. Most failed swaps come from only doing the first. These habits are what the shipped examples in the catalog lean on, such as asking to change only the paint of an SUV and to keep the geometry, glass, reflections and camera movement unchanged.
Tip: if the swap drifts, shorten the prompt instead of lengthening it. A sentence that names the old object, the new object and one thing to keep usually steers better than a paragraph of adjectives, and the edit you can read in one breath is the one you can debug.
Seeds help once you have a result you like. On models that expose a seed, fix it and change only the wording, so a difference in the output comes from your edit and not from chance.
Replacement is good at a single object that stays mostly in view and does not touch much. It gets unreliable in a few specific places, and knowing them saves credits.
Reflections are the first. If the old object sits on glass, water or a polished table, its reflection is part of the picture, and the model may swap the object and leave a ghost of the old one in the mirror, or the other way round. Check every shiny surface frame by frame, and if the reflection matters to the shot, expect to fix it with a second pass or accept a different composition.
Hands are the second. An object being held, poured or turned is a moving contact between fingers and surface, and the swapped version can warp, slide or change grip between frames. Swaps work best on objects that sit on a surface or are worn, such as a bottle on a shelf or a jacket on a shoulder, and worst on a phone being typed on.
Brand logos are the third. Models redraw small printed text and marks, so a logo on a new package can come out with changed letters or a distorted shape. Use a reference photo where the logo is large and sharp, check the result at full size, and if the exact mark matters, apply it in your editor afterward instead of trusting the model. Fast motion, objects that leave the frame and come back, and a new object far larger or smaller than the old one cause drift too, and clips longer than the model limit simply cannot be edited in one pass.
Yes, for many shots. You upload the clip, describe what should change and optionally attach a reference photo, and a video editing model rewrites that object across the clip while the camera movement, the timing and the rest of the scene stay as filmed. It works best on one clearly visible object in a short clip. It is not a guarantee for every shot, so test a few seconds first and keep the original footage.
Start with P Video Edit. It takes a text instruction and up to four reference images, which is what a product swap needs, since the new package has to match a real photo. If the clip has several shots, try Aleph 2, which applies one instruction across cuts and accepts keyframe images. Running the same clip through both and keeping the better one costs a little more and is the fastest way to learn which fits your footage.
It depends on the model. The catalog lists clips up to 15 seconds for P Video Edit and between 2 and 30 seconds, under 16 MB, for Aleph 2. A longer video has to be split into shots, edited one by one and joined back in your editor. Cutting to the few seconds where the object appears is a good habit anyway, because shorter clips are cheaper to retry.
Yes, with the models built for it. P Video Replace puts a person from one to three reference photos into the motion of your clip, and Wan 2.2 Animate Replace swaps in a character from a single image. Use only your own likeness, a fictional or illustrated character, or someone who has clearly agreed. Never place a real identifiable person into a video without their permission.
Usually you can keep it. P Video Edit and P Video Replace have a save audio option that is on by default, and Wan 2.2 Animate Replace merges the source audio by default. The Aleph 2 listing does not say how it treats audio, so listen to the result and, if it matters, put the original track back in your editor. Check the first run before you rely on it.
Yes, this is one of the cleaner swaps, because a sky is a large area with little fine detail. Ask for the new sky by name, such as a warm sunset with soft clouds, and add that the subject and the lighting on it should stay unchanged. Expect the mood on the subject to lag behind the new sky unless you mention it, so say whether the light on faces and objects should follow.
You cannot fully guarantee it, because models redraw small marks. Use a reference image where the logo is large and sharp, keep the object facing the camera in the clip, and check the result at full size. If the exact logo matters for a launch or an ad, replace the object with AI and add the mark afterward in your editor, where the letters stay exactly as designed.
Replacing puts a new object where the old one was, so the prompt names both. Removing makes the object disappear and fills the space with background, which needs different models and prompts. If you only want something gone, read remove objects from videos. If the whole look should change, as in a painted or animated style, that is restyling and a separate edit.
Be careful. The picture must stay truthful about the item you sell, so do not swap in a product that looks better than the one a buyer will receive, and do not change a color or feature the listing promises. Each platform has its own rules on edited and AI media, so check its current requirements before you publish. For product footage, see AI product demo videos.
Generations are paid in credits, and the cost depends on the model, the resolution and the settings you pick, so any number written here would go stale. Video edits cost more than images, and a draft pass is a cheaper way to check a prompt. New accounts get free credits to start, and the pricing page is the only place worth trusting for current plans.
Got a clip with one thing in it that is wrong? Open the Picasso IA toolkit, upload the shot and run a short draft of the swap today.
Choose your language
