Your Next Sale Needs an Image That Doesn’t Exist Yet.

Imagine a shopper opening your outerwear webshop on a rainy afternoon in Copenhagen. They see a yellow waterproof jacket styled for a wet city commute. That colour is in stock locally, and the matching trousers can arrive tomorrow.

Someone opens the same page in Madrid. They see a lightweight windbreaker in a dry evening setting, selected from the Spanish assortment.

Neither image existed before the visit. Each was created for that specific moment and may never be needed again.

This is a near-future ecommerce model we call Just-in-Time Imagery. A prepared creative blueprint meets live data to produce an image when there is a reason to show it.

Brief first. Image later.

Long before a shopper arrives, the brand prepares a detailed prompt containing its visual guidelines, approved product assets and conversion rules.

This blueprint defines lighting, styling, composition and product accuracy requirements. It establishes which products to prioritise, what to pair together and what must never appear.

Certain fields remain open. When someone lands on the page, live data completes them:

• Market: Local assortment, language and delivery options.
• Context: Weather, time of day and approximate location, where available and permitted.
• Stock: Available variants and products to exclude.
• Priorities: Products or colours that deserve greater visibility.

The completed prompt then guides production of the image. Much of this relevance comes from the immediate shopping context, without needing to identify the individual.

Show what’s available.

A traditional campaign image can keep promoting a sold-out colour for weeks. The shopper clicks, discovers the problem and has to start again.

With live stock data, the system can exclude sold-out variants before generating the image.

Excess local stock? Give that colour more visibility. Sudden cold snap? Show an available mid-layer alongside the jacket.

The image becomes an active part of merchandising, helping shoppers discover relevant products they can actually buy. If a shopper has already filtered by size, that detail can further refine which products appear.

The commercial effect still needs testing. Product clicks, add-to-basket rates, completed purchases and generation costs will determine whether the approach earns its place.

Protect the product.

A relevant setting means little if the jacket has the wrong zip.

Product colour, construction, materials and details must remain accurate. At Scenes, combining real product photography with AI-generated environments provides a useful foundation: protect verified product assets while adapting the surrounding scene.

Accuracy checks and clear brand rules remain essential. A change in weather should never make the webshop look like a different brand.

Don’t make shoppers wait.

An approved baseline image can load immediately while contextual generation happens in the background. The new image appears only once it is ready and checked. If generation fails or stock data is uncertain, the baseline stays in place.

Useful results should also be cached and reused. Thousands of shoppers may share the same market, weather and available assortment. There is little reason to generate the same visual repeatedly.

Created when needed. Reused while relevant.

Produce or stockpile?

Creating an image for a single visit can sound wasteful. Repeated generation consumes resources and needs a clear commercial purpose.

But an archive has a production cost too. Brands create images for possible seasons, markets and campaigns that may never materialise. When a specific need finally arises, the archived image may show the wrong product, colour or setting.

Just-in-Time Imagery is not automatically more efficient. The balance depends on generation costs, reuse and commercial results. A practical approach combines enduring campaign assets with contextual generation where it adds value.

It leaves ecommerce teams with a question worth asking:

Should you create images for demand that may never arrive, or generate them when there is something relevant to show?