Where AI works in product photography and where it doesn’t

October 2, 2026

A message arrives, and lately it arrives often: “we generated the whole catalogue with AI, could you check it before we upload?”

The files open. A bag with two seams the brand doesn’t manufacture. A trainer whose sole pattern changes between photo three and photo five. A beautiful bottle wearing the cap of a different model.

None of those images is banned for being generated. All of them sink the listing for a different reason, and it’s the reason almost nobody is looking at.

The question that matters

Before publishing an image, check one basic thing: does it show the product your customer will receive?

A generated background can change the setting while leaving the item intact. But if the stitching, sole, cap or finish changes, the image no longer represents what you sell.

The difference lies in what you generate and what you preserve from the real product.

Where AI works in product photography and where it doesn’t

Where AI genuinely works

There are four places where it earns its keep.

Extending scenes from a real product already photographed. A pair of shoes shot properly can live in twenty different environments without going back to the studio. Shape, colour and detail come from the original.

Social content between launches. Volume that doesn’t have to serve as proof of purchase.

Background and seasonal variations. The same still life in summer and at Christmas.

Concept tests before a campaign. Showing three directions without building three sets.

The quality of the result depends on the source material. If the original photo does not capture colour, shape or detail accurately, generation can repeat those errors or introduce new ones.

That is why our AI content projects begin with a real product shoot. We then check each generated image against the original and record where generation has been used.

In our AI content work the starting point is always a real shoot of the product. And we log which images involved generation, so you can make that call with actual information.

Where real photography is needed

The main product listing image.
The priority is to represent the item accurately. A real photograph provides a reliable starting point for showing its shape, proportions, colour and contents.

Materials and finishes that define the product.
Patent leather, polished metal and stones change with the light. When those reflections are part of what you sell, they should be photographed and preserved in the final image.

Details that decide the purchase.
Stitching, fabric texture, fastenings, engravings and finishes. These are the areas buyers zoom into to assess the product, and they require a real reference.

Products without a physical sample.
Generation can help visualise an idea before production. That image should be treated as a concept, rather than a verified photograph of the finished item.

Textures and swatches. A texture is how a substance behaves under one particular light, and that’s where generation improvises. It’s set out in full in the shot almost nobody gets right.

When AI works out more expensive than shooting

It happens more than people expect.

It costs more when the product line has heavy reference variation: sixty models that resemble each other mean correcting every image by hand, and the correction time overtakes shooting the collection in one go.

It costs more when the output has to be redone for accuracy. Two rounds of corrections across twenty generated images cost more than the shoot day you were avoiding.

And it costs more when the return arrives. That cost isn’t in anyone’s budget.

How to decide

Use	AI?	Why
Marketplace main image	Only if it's exact	The rule looks at the result, not the method
Environments for the same reference	Yes	Starts from real material
Detail and texture macro	Better not	It's exactly what the buyer zooms into
Seasonal backgrounds	Yes	Doesn't decide the purchase
Jewellery and watches	With care	Metal and stone get approximated: check against the original
Feed between launches	Yes	Volume, not proof of purchase
Product with no physical sample	No	With no real reference it resembles yours without being yours
Pre-campaign concept	Yes	Not published as product

What to ask before commissioning generation

What real material do the images start from?
Who checks accuracy of shape, colour and detail, and against what?
Is it logged which images involved generation?
What happens if a platform rejects an image?
Do the files work for the listing, or only for social?

If the source material does not let you verify colour, shape and detail, ask for a photograph that does before proceeding.

Before publishing

Check the platform’s current rules and any transparency requirements that may apply to generated content, especially when it includes AI-generated people.

Identifying synthetic content and representing the product accurately are separate issues: a label does not correct an inaccurate image. Both should be checked before publication.

First the product. Then the possibilities.

A good shoot establishes the foundation: shape, colour, materials and detail. From there, AI can expand the possibilities with new settings, seasonal variations and content between launches.

In our studio, we combine real photography and generation according to what each image needs. The aim is to create more useful content while preserving the product’s identity.