Ecommerce video ads for your whole catalogue

You have more products than you can make videos for

Ask in any seller community and the same complaint comes back in slightly different words. One merchant with a large catalogue said: “Creating videos for every SKU isn't practical, especially if you have a large catalogue.” Another: “Since it takes a lot of time for the team to create high-quality videos for each SKU, we currently concentrate mostly on video content for our bestsellers.” A third named the problem exactly: “Creating videos for every SKU upfront usually turns into a production bottleneck.”

The constraint is not that sellers cannot make a video. It is that they cannot make enough of them.

So they triage. The same threads describe how: “We usually prioritise videos for products that already show demand rather than trying to cover the entire catalogue.” That is a sound decision. Demand is proven, so the return is easy to measure, and the effort goes where the evidence is.

But triage has a price.

So the real question is not which product deserves a video. It is how a validated angle reaches the rest of the catalogue without a new production for each item. This page answers that question. The answer is not “work faster”, because speed is a capacity argument and capacity is what you are short of. The answer is to change the unit of work. Stop counting one video per product, and start counting one format across the whole catalogue.

What a product demo video has to do

A product demo video runs in a social feed, organic or paid, alongside everything else in that feed. It has about three seconds to explain why anyone should stop. The whole tool is built for that slot, and the whole page follows from it.

Those three seconds are where most ecommerce video budgets get wasted. They are also where structure beats spend. A product demo video lives or dies on structure, not on production value.

The short-form product videos that perform are rarely the most expensive ones to shoot. They are the clearest ones, and the order is usually the same. A specific problem or result in the first two to three seconds. One or two seconds explaining why that happens. The product in use for the longest stretch. A single verifiable fact. A close that asks for one action and introduces nothing new.

That order leaves out the brand story, the mission and the montage. The structure does the selling and the product does the showing.

This is the part BrandU does not ask you to invent. The structure is read out of a product video that is already running, second by second, into a production sheet. Then it is rebuilt around your product. You do not have to know why the format works before you can use it. You only have to find one video where it demonstrably does.

One format, every product

A catalogue video is not a video of your catalogue. It is the master you generate the catalogue from, and it has two halves. The first is the structure: script order, how shots connect, where captions sit, pacing, aspect ratio. All of it comes from a reference video that already works. The second is your brand: name, visual language, tone of voice. Set it once and reuse it on every product you make.

The only thing that changes from one product to the next is the product itself: image, name, price, selling points. Everything upstream stays where it is. So the expensive part happens once. Deciding what the video is and how it is built is the expensive part. What repeats is the cheap part.

Before you can make the master, you have to decide what it is arguing. This is where most teams reach for a list of ideas, and it is the wrong tool. A seller who runs their own ads described a better method in one sentence: “I look for SKUs with proven demand and a large creative gap. Then I'd use reviews, support questions and comments to generate 3-5 assumptions rather than video styles. For example: objection, use case, demonstration.”

The distinction is the whole point. Assumptions, not styles. Changing a style usually means rebuilding the video. Changing an assumption usually means changing the opening segment only.

So the best source of product video ideas is not inspiration. It is your own returns policy, your support inbox and the two-star reviews.

Change one product, not the whole video

Most generation tools are built on one chain. A prompt goes into a model, and a finished video comes out. Every single edit goes back through the model, which is why changing one line of dialogue costs the same as making the video again. That is a structural property, not a pricing decision.

BrandU separates the two halves of that chain. A reference video is broken down second by second into a production sheet. Your brand and product information are folded into that sheet, and the model generates a reusable intermediate: script, components, visual style. The final video is rendered from that intermediate. The model has already done its work by then.

So there are three kinds of edit, with three different costs. Segment order, element styling and swapping in different assets re-render only. The copy in one segment re-runs that segment and its voiceover. Changing all of the copy is the only full-cost edit.

That has a consequence worth spelling out. The more you edit, the wider the gap gets. Edits are not free. It is that most of the edits an ecommerce team actually makes do not touch the model at all: a new product image, a different opening, a different order, a different aspect ratio. The fiftieth variant of a format is not priced like the first one, because it is not doing the work the first one did.

This is the structural reason a catalogue-wide rollout is affordable. Not a discount. Not a credit bundle. The work simply is not repeated. Because the work is not repeated, the cost does not scale with the catalogue.

Two more things follow from the same split. The video is generated as independent segments rather than one flat file, so changing one thing re-times the rest. Swap the hook, rewrite a line, change the language or replace a component, and the timeline and the related elements follow. You never adjust a timeline by hand. The attention this saves is attention you get to spend on what the ad argues, instead of on cutting and re-cutting assets.

Bringing your catalogue in

Your product library takes products in by link. Paste a product page URL and it becomes a reusable product entity: images, name, description, price, promotion. You can paste several at once, one per line or comma-separated, so a two-hundred-product catalogue does not have to be entered one at a time. Each link is validated individually, and if one fails, the error tells you which line it was.

Two things are worth being precise about. First, products and assets are different libraries. A product is the thing you are promoting, a reusable entity with its own images and pricing. Assets are files: images, voiceovers, b-roll. Keeping them apart matters because they are replaced for different reasons.

Second, importing a product is not the same as generating a video. The import puts the product in your library. The video is still built from the structure of a reference video, as described above. Import is how your catalogue gets in. It is not a shortcut past the structure.

Once a product is in the library, it becomes one of the axes you can vary. A concept can be rebuilt along five of them: the hook opening, the presenter or on-camera persona, the product SKU, the language, and the aspect ratio. Swapping the product is swapping one axis. The script structure, the pacing, the caption placement and the visual style stay where they are.

So the product library is not storage. It is the input to variation, and it is why a catalogue of two hundred is a different problem from two hundred videos.

Where these videos run

The output here is promoting video for social feeds: organic posts and paid placements. That is where ecommerce brands actually compete for attention, and the surfaces differ enough that one asset cannot serve them all unedited.

Two things in the requirements table are worth pulling out. TikTok Shop's creative guide puts the hook in the first 3 to 6 seconds, and its compliance rules bar slideshows and looping still images from shoppable video. That ban is not a problem here, because what gets generated is motion, not a carousel. Meta's own number is twenty, not three.

That asymmetry is the reason a single master matters. You are not choosing between twenty videos. You are choosing how fast a proven format can be rebuilt into the twenty the platform is asking for.

Start with the master at 9:16, because that is where most of this traffic lives. For TikTok, use it as it is. For Reels and Shorts, keep the ratio and replace the opening segment, because each feed rewards a different hook. For paid Meta placements, change the ratio to 1:1 or 4:3 and keep the rest. For Shopify and DTC pages, it is the same asset, embedded where shoppers browse.

This is where a Shopify video maker should earn its keep: one format rebuilt per placement, not one video per placement.

The aspect ratio is a field on the create screen with six options: 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, vertical by default. Changing it falls into the re-render-only category. The model does not run again.

One detail saves more time than it sounds like. The video-to-clone field accepts an ordinary web page URL. The placeholder says it plainly: “Paste a video link (a page URL is fine)”. You do not have to download the video first, and you do not need a direct .mp4 link. If a TikTok Shop video is working, you paste the page it is on.

Test angles, not styles

Ecommerce creative testing fails in one specific way, and it decides whether your testing compounds or resets every time. The mistake is testing styles when you meant to test assumptions.

Start with what the platform asks for, because the number is larger than most teams assume. Meta's guidance for Advantage+ shopping campaigns is to keep at least 20 diversified ads running, and to put 20-30% of budget into testing new ones. Twenty, not three.

The same guidance defines diversified: ads that are genuinely distinct in their visuals, messaging and formats. Not twenty versions of one ad with the colour changed. So the platform is asking for volume and for difference at the same time.

Those two goals pull against each other. Volume feeds the algorithm. Controlled variables feed your judgement. A test that only serves the first one teaches you nothing. The seller quoted earlier drew the line precisely: “I'd generate 3-5 assumptions rather than video styles. For example: objection, use case, demonstration.”

The last row of that comparison is the ecommerce advantage. A style is worth something once. An assumption is worth something across the whole catalogue, because the same objection shows up on many products. Test “will it fit” on one item, and you have a hypothesis you can spend on thirty others.

One rule keeps a test readable: do not change several things at once. If you change the opening, the persona, the offer and the landing page together and the result improves, you know that something worked and nothing about what. The five axes above are exactly the list of things to move one at a time: hook, persona, product, language, aspect ratio.

What stays, what changes

This is what cloning means here, component by component.

The left column answers what cloning copies: structure, not pixels. The right column answers what gets replaced: product, copy, voice, footage, brand. Nothing on the right is the reason the reference video worked. Everything on the left is.

What this looks like in practice

One proven format, every product on the catalogue.

By category

The same clone-and-swap loop fits most catalogue types. Only the example changes.

Fashion & jewelry

Show the piece on a model; clone a look-book format.

Pets & supplements

Before/after and testimonial formats, rebuilt with your product.

Food, hotel & travel

Turn a destination or dish video into your offer.

What BrandU does not do

A generation system can tell you what it made. It cannot tell you how the auction will receive it. So, plainly:

• It does not guarantee the video performs. Nothing does.
• It does not measure the result. Producing variants is not measuring them. The read still comes from your platform's own testing.
• It does not invent your angle. It rebuilds one that has already been tested, which is the whole reason it starts from a reference video instead of from a blank prompt.
• It does not fix a weak offer or a broken landing page. Volume gives the auction more to choose from. It does not choose for you.

How to run this

You do not have to start from a reference video. If you have one, that is the strongest path, because the structure is already proven. If you do not, the create screen also takes a one-sentence description. Two ways in, same editor afterwards.

• Find a product video that is already working: a TikTok Shop video in your category, a competitor ad you keep seeing, or your own best performer. Paste its page link. No download, no direct file URL needed.
• Bring your products in. Several links at once, into the product library.
• Generate up to 10 script variants in one pass. Each one is an angle hypothesis, not a style.
• Keep the one that runs, and treat it as your master.
• Change the product, the opening, the language or the aspect ratio. Only re-render.
• Roll the master out across the catalogue.