How to Replace a Product in a Video with AI

To replace a product in a video with AI, you upload the existing clip, drop in a photo of the new product with a label describing what it replaces, and let an editing model repaint just that region frame by frame — everything else in the shot (the person, the hand movement, the camera path, the background) stays exactly as filmed. On VdoBloom this is the Object Swap preset inside Genjutsu, and it runs on ByteDance’s Seedance 2.5 video-editing mode, which keeps the output at the same length and aspect ratio as your source clip.

Disclosure: VdoBloom is our platform. Figures below are from our own generations on 20 September 2026.

Why this is different from generating a new ad from scratch

If you already have a video ad that works — good pacing, a hook that lands, a hand reaching for a product on a table — reshooting it every time you launch a new SKU, a seasonal colourway, or a regional variant is expensive. Object swap is built for that exact situation: you keep the footage and change only the thing you label. It is not text-to-video and it is not a full remix. It is closer to a very precise digital reshoot of one object in an existing take.

This matters because the alternative — asking a general video model to “change the drink can to a different brand” — tends to also redo the lighting, nudge the camera, or reanimate the hand holding it. Editing-mode models like Seedance 2.5 are trained to leave untouched pixels untouched, which is the whole point when you need brand consistency across ten product variants of the same 15-second spot.

What you need before you start

• A source clip: 2–30 seconds, 480p–720p, 24–60 fps, under 200 MB, MP4 or MOV.
• One reference photo per product you want to swap in — up to 10 per job — each with a short label saying what it replaces, e.g. “the drink can on the table” or “the football kit.”
• An optional note field for extra instruction (useful for tricky scenes — more on this below).
• A choice of output resolution: 480p, 720p, or 1080p.

Genjutsu lives at /dashboard/video-creation/genjutsu/. The Object Swap preset is one of the options inside it; the tool doesn’t touch anything you don’t label, which is also why labelling well is most of the work.

Step by step

1. Upload the source clip. VdoBloom checks it against the spec (length, resolution, fps, file size) and refuses out-of-spec clips before you’re charged anything.
2. Add your reference photo(s). One clean, well-lit product shot per target, ideally with a plain background and any label or logo facing the camera. This is the single biggest factor in swap quality.
3. Label each photo. Be specific about which object it replaces — “the can on the left,” not just “the drink.” The label is how the model tells your product apart from everything else in frame.
4. Use the note field for context if the scene is busy or the product is partially obscured — for example, noting that the product is only visible for the first half of the clip.
5. Pick your output resolution and generate. A job takes roughly 4–7 minutes.
6. Review, then re-run at full resolution if needed. We recommend drafting at 480p first (see below) and re-running at 720p or 1080p only once the swap itself looks right.

The model carries over the original object’s physical behaviour: if a hand lifts the old can in the source footage, the same hand lifts the new can in the same way, at the same time, in the same spot. You’re not re-animating the scene — you’re re-skinning one object inside it.

What it costs

Genjutsu Object Swap is priced per second of clip at the output resolution you choose, and the price is shown from your clip’s length before you upload anything, so there are no surprises after the fact. Failed jobs are refunded.

Clip length	Output resolution	Credits
4 seconds	480p	41
15 seconds	480p	212
15 seconds	720p	473
15 seconds	1080p	852
30 seconds	720p	912

Credits come from one-time packs starting at $2.49 that don’t expire, and there’s no subscription required to run this tool. New accounts get 10 free credits, which is enough to see the interface but not enough to complete even the cheapest job (41 credits) — plan to top up before your first real test.

Because pricing scales with output resolution rather than a flat per-job fee, the cheap move when you’re testing a new product photo or scene is to draft at 480p first. A 15-second 480p draft costs 212 credits versus 852 for the same clip at 1080p — a four-times difference for a check you might redo two or three times before you’re happy with the label wording and the reference photo.

What we actually tested

On 20 September 2026 we ran a swap on a clip of a footballer wearing a team kit, replacing the kit with a leather coat using a single product-style flat-lay reference photo. The face, the motion, and the rest of the scene were unchanged — the coat took on the kit’s fit and movement rather than looking pasted on.

We also ran a harder case: two targets in one scene, both pointed at the same coat photo. In that test, in a crowded wide shot, the model touched a nearby player who wasn’t one of our labelled targets. That’s a real limitation worth planning around, not a hypothetical edge case — it happened in our own generation, not in a support ticket from someone else.

How to get a clean swap

• One distinct photo per target. Don’t combine two products into a single reference image, and don’t reuse one photo for two targets.
• Specific labels. Name the object by its position or visible detail, not just its category.
• Use the note field to flag anything unusual — multiple people in frame, a product that’s partially hidden, quick cuts.
• Draft at 480p first. It’s a quarter of the cost of 1080p and enough to judge whether the swap itself worked before you pay for final resolution.
• Keep the product photo clean. Plain background, good lighting, label or logo facing the camera. This does more for output quality than any setting in the tool.
• Be cautious with crowded or multi-person shots until you’ve tested one similar to yours — that’s where our own two-target test ran into trouble.

When you don’t have footage to swap into

Object swap needs an existing clip. If you’re starting a new product line with no hero ad yet, or you want variants described from scratch rather than built on one take, VdoBloom’s Ad Variants tab generates video ads from a text brief instead of editing footage. The two tools solve different problems: reuse an ad you already have (Object Swap) versus generate new ad concepts (Ad Variants).

Where this fits for ecommerce

The common pattern is one hero ad reused across a product line: the same 15-second spot re-run for each SKU or colourway, the same seasonal ad re-skinned with holiday packaging, or a handful of A/B variants built off one base take to see which product presentation performs. Object swap is built for exactly that — keep the ad, change the product, without booking a reshoot for every variant.

Related reading

• AI Object Swap for Video
• Genjutsu AI Video
• How Genjutsu Works, Explained
• Ad Variants (text-to-ad alternative)

Bottom line

Object swap is a genuinely narrow tool and that narrowness is the feature: it changes what you label and leaves the rest of the take alone, at the same length and aspect ratio as your source. It works cleanly on single-product swaps with a good reference photo. It can misfire on crowded, multi-person, multi-target scenes — we saw that ourselves. Start with one target, a clean product photo, a specific label, and a 480p draft, and you’ll know within one 4–7 minute job whether it’s going to work for your ad.