The advice is sound. The problem has always been cost. A half-day shoot with a videographer, a few edits and a round of revisions can easily run to more than a small online shop spends on marketing in a quarter.
AI video tools have changed the arithmetic. A short clip that shows a product in a styled setting can now be generated from a single product photograph in minutes. For a business with two hundred SKUs and a marketing team of one, that is a significant shift. It is also easy to get wrong, and the mistakes are the kind that cost customer trust rather than just money.
What the tools actually do well
The most useful capability for retailers is image-to-video: you supply an existing product photo, describe the scene and the motion you want, and the model produces a short clip that begins from your image. A candle on a table with the flame flickering and evening light moving across the room. A trainer rotating slowly against a clean background. A cup of coffee with steam rising in a café setting.
These are atmosphere shots, and that is where the technology shines. They give a product page or a social post movement and mood, without a studio, props or a photographer.
Newer models have also become better at longer clips and synchronised sound. Earlier tools produced a few seconds at most, which meant stitching fragments together for anything resembling an advert. Some current models can generate a continuous clip of around thirty seconds with audio, which is long enough for a complete social advert or a product page hero video.
Where small businesses get into trouble
Misrepresenting the product. This is the serious one. Generative models are good at plausible detail and bad at exact detail. They may subtly change a logo, alter stitching, add a feature that does not exist or change the colour under different lighting. If a customer buys based on a video and receives something that looks different, that is a return, a bad review and potentially a trading standards issue. Under UK consumer protection rules, marketing must not mislead about the main characteristics of a product, and “the AI did it” is not a defence.
The safe rule is to use generated video for setting and mood, and to keep the product itself as close to your real photography as possible. Many retailers use generated clips for social posts and adverts, while product pages continue to rely on genuine photography and, where possible, real footage of the item.
Generic results. Text-only prompts produce footage that looks like everyone else’s footage. Starting from your own product images and brand photography keeps the output recognisably yours.
Underestimating the time. The generation is fast; the selection is not. Expect to produce several versions for every clip you are happy with, and budget time for checking each one carefully for errors in the product.
A sensible workflow for a small team
Pick the right products. Start with items where atmosphere sells more than detail: home fragrance, food and drink, décor, gifts. Leave technical products, where buyers scrutinise specifications, until you are confident.
Use your best photography as the starting point. A clean, well-lit product image gives the model the most accurate reference.
Describe small, believable motion. Slow camera moves, gentle light changes and natural movement work far better than dramatic action.
Check every frame of the product. Logo, colour, shape, labels. Reject anything that misrepresents the item.
Label honestly where appropriate. Several platforms now expect AI-generated advertising content to be disclosed, and customers increasingly appreciate transparency.
What it costs
Most AI video services price by the second of output, often with higher rates for higher resolution or added audio. Because you will discard a good share of what you generate, the useful budgeting figure is the cost of all attempts, not the cost of one clip. Even so, most small retailers find that a month of experiments costs less than a single traditional shoot day.
Businesses that want to build video generation into their own systems, for example to create a standard clip automatically for each new product added to their shop, usually connect to the model through an API rather than using a consumer app. If that is on your roadmap, it is worth comparing per-second rates and clip lengths across models before committing; the published pricing for the ByteDance Seedance 2.5 API, which generates native 30-second clips with audio, gives a realistic benchmark for what longer-form product video costs at volume.
Measuring whether it works
Treat AI video like any other marketing test. Run it on a handful of products or campaigns, compare against your existing photography or adverts, and look at click-through, add-to-basket and return rates rather than views alone. A clip that drives more sales but also more returns has not done its job, and returns are often the first sign that a generated video has oversold the product.
The bottom line
AI video gives small retailers access to a kind of marketing that used to require a production budget. Used for mood, setting and social content, it can make a small brand look considerably bigger than it is. Used carelessly on the product itself, it risks the one thing a small business cannot afford to lose: customers believing that what they see is what they will get.
Start small, keep your real photography at the heart of your product pages, check every clip, and let the results rather than the novelty decide how far to go.
