Recreating a great shot usually means starting from scratch: planning it, staging the performance, getting the camera moves right all over again. That eats time and money. A newer kind of AI video tool skips most of that. You take a clip that already works and rebuild it around a different character, product or setting.
Pollo Recast is one of these tools, made by Pollo AI. You hand it a reference video for the action and camera work, then give it your own images for whatever should take over the scene. Below: how it works, how to prep your materials, how to get more consistent results, and the rights and labeling rules you can’t ignore.
So what does it actually do?
Recast rebuilds a video around new visuals while keeping what made the original shot good. According to Pollo, it follows the reference clip’s action, camera movement, cuts and timing, then swaps in the character, product, prop or setting you supply as images.
The new subject can be pretty much anything: a different person, an AI character, a mascot, your dog, a product, or a whole different world. Say you’ve got a clip of a model showing off a product while the camera slowly pushes in. Recast lets you keep that presentation and that camera move, but with your character and your product in the frame.
Pollo is clear that this isn’t the same as its other tools. Its motion-transfer tool starts from a single image. Video-to-video restyles footage you already have. Face swap only touches the face. Recast rebuilds the entire subject through the whole performance instead of sticking a new face onto one patch of the frame.
The workflow, start to finish
The process is short. You upload a reference video, or grab one of Pollo’s templates. Then you add images for the character, product, prop or setting you want to bring in, write a quick direction, generate, and download. If you want something longer, you can stitch several results together in an AI video editor.
The reference handles the movement and rhythm, and your images decide what the new version stars. The clearer you are about what should stay and what should change, the better the tool handles the split between the two.
Pick your reference clip carefully
A good reference makes everything after it easier. Look for a short, focused clip with one clear subject doing one main thing: a walk, a product reveal, a gesture, a dance. It helps if the action has an obvious start, middle and end, and if the camera moves smoothly and predictably.
Steer clear of clips with lots of cuts, several things happening at once, heavy motion blur or busy crowds. The messier the reference, the harder it is for the tool to carry the shot over cleanly.
Choose based on the movement you actually need. A walking shot suits a fashion campaign. A slow reveal works for an e-commerce ad, and a gesture-heavy clip suits a virtual character.
And before you upload anything, ask yourself one thing: am I actually allowed to use this clip? We’ll come back to that.
Get your images right
Your source images shape the final video more than anything else.
For a character, use a picture that clearly shows what makes them who they are: face, hair, clothes, build, color palette. If you’re making a series, stick with the same character reference every time so they stay recognizable.
For a product, go for a clean shot with crisp edges and visible details, so the shape, colors, logo, and key design touches are easy to read. A plain background usually helps.
Bigger jobs can call for several references: a portrait for facial detail, a full-body shot for movement, a close-up of the product for branding, an outfit photo for styling, a scene image for the background. Whatever you use, don’t keep tweaking the design between generations. A character or product that shifts slightly every round makes consistent results a lot harder to get.
Keep your direction short and clear
You don’t need an essay. You need to say what changes and what stays. Something like this:
Replace the original performer with the uploaded virtual character. Keep the walking movement, camera tracking, and shot timing from the reference video. Use a clean studio background and maintain the character’s hairstyle, outfit, and facial features.
That one paragraph covers what to swap, what to keep, which details matter and what setting to use. Skip anything that doesn’t change the shot. Extra detail tends to blur the instruction rather than sharpen it.
Watch the whole thing, then fix one thing at a time
Play every result all the way through, because some problems only show up in motion. A character can look perfect in a still frame, then drift during a turn. A logo can start crisp and go wobbly in a close-up.
When something’s off, go after the most likely cause first. If the product keeps changing shape, use a better product image or simplify the action. If the character keeps slipping, try a clearer reference and cut back on unnecessary movement.
It’s tempting to rewrite the entire prompt after one bad result. Don’t. Change one big thing per round: the replacement image, how much motion there is, the camera move, the background, the outfit. That way you actually learn what’s working, and you don’t wreck the parts that were already fine. For complicated scenes, simplify the action before piling on detail. A clean, controlled shot beats a cluttered prompt full of competing instructions every time.
Where it earns its keep
The obvious win is getting more out of a shot that already works. E-commerce teams can reuse a strong product presentation for different products, packaging or seasonal campaigns without reshooting each one. Brands can drop the same virtual character or mascot into different actions, outfits and scenes, so a recurring series still feels like one series. Social creators can borrow a popular move or camera style for their own character. Regional marketing teams can spin up local versions of one idea, with different people, products or settings.
Every one of those comes down to reuse. The original shot’s structure becomes a template, and your creative effort goes where it actually counts.
Rights, consent and labeling: don’t skip this part
Tools that drop a new subject into someone’s existing performance use the same basic trick behind deepfakes. That doesn’t make them bad. It just means some rules apply.
Only use footage you have rights to. Recast keeps a clip’s movement, framing, camera work and pacing, so the reference really matters. Your own footage, licensed stock, or clips you’ve got permission for are safe bets. Building commercial work on someone else’s video without permission can land you in copyright trouble, even if every visible face and product has been swapped out.
Get consent from real people. Putting a real person’s likeness into a video, or recasting a real performer’s moves onto someone else, needs their permission. Using someone’s likeness without consent can cause legal headaches, especially in ads, and platforms take it seriously. YouTube, for one, lets people ask for AI content depicting them to be taken down.
Label anything that looks real. YouTube requires creators to disclose realistic altered or synthetic content, including content that uses a realistic person’s likeness. TikTok requires labels on AI content showing realistic scenes, and it bans AI deepfakes of young people and private individuals outright. In the EU, the AI Act’s transparency rules have applied since August 2, 2026: if you publish deepfake video, you have to clearly say it was artificially generated or manipulated. Obviously artistic or fictional work gets a lighter-touch disclosure, but not a free pass.
Recast is part of a bigger shift in AI avatars and digital identity. The better these tools get at mimicking real people, the more consent and transparency matter.
How it stacks up against similar tools
Recast isn’t the only option. Alibaba’s open-source Wan 2.2 Animate, released in September 2025, does both animation and character replacement, and you can run it yourself through workflows like ComfyUI. That’s handy if you want more control over the pipeline and don’t mind some technical setup. Kuaishou’s Kling has motion control that transfers movement from a reference video onto a character image, and Runway has its own character performance tools.
Recast’s pitch is convenience. It’s hosted and template-driven, it’s built around recasting a whole shot, and there’s nothing to install. Whichever tool catches your eye, test it on your own footage before building a workflow around it.
Quick tips
Start with a short reference clip and stick to one main action per shot. Use clear source images and keep them consistent. Keep prompts short and to the point, and change one major thing per revision. Generate a few versions before you pick a winner. And before anything goes public, make sure you have rights to the reference footage and consent from any real people in it.
Quick answers
Q. What is Pollo Recast?
It’s an AI video tool from Pollo AI that rebuilds an existing clip around your own character, product, prop, or setting, while keeping the original’s action, camera movement, and timing.
Q. How is it different from face swap?
Face swap only replaces a face. Recast rebuilds the whole subject through the entire performance, and it can swap products, props and settings too.
Q. Can I try it for free?
Pollo says new users can test it with free account credits. Paid plans give you more generations.
Q. Do I have to label videos made with Recast?
If the result looks realistic, often yes. YouTube and TikTok require labels on realistic AI content, and in the EU, deepfakes have to be disclosed under the AI Act.
Q. Can I use any video as a reference?
Stick to footage you own, have licensed, or have permission to use, especially for anything commercial.
Final Thoughts
Shot recasting with AI can save a lot of reshooting and turn one strong idea into a whole set of versions. With clean references, steady character designs, and proper consent and labeling, it’s a genuinely practical tool for creators and marketers.
Related: AI Video Generation’s Next Battleground Is Camera Motion
