AI product images: when to use them and when not to
Generating a background is ordinary image editing at a different scale. Generating the product, or generating a claim about it, is something else entirely — and the line between them is where the legal risk sits.
Shops have retouched product photography for decades — removing a distraction, cleaning a background, adjusting light. Generation is a faster version of the same work until it crosses one line, and then it becomes a misleading commercial practice.
The line is not technical and it is not about how the image was made. It is about whether the picture makes a claim the product does not support.
Where it is straightforwardly fine
- Backgrounds and settings. Placing your real bottle on a real-looking marble counter that does not exist. The product is unchanged.
- Lighting and season. The same shot as a summer morning and a winter evening.
- Format variations. Extending a background so one photo works as a square, a story and a banner.
- Mood and concept imagery that is clearly not a product depiction — a texture, an abstract header, a seasonal illustration.
In all of these the shopper's understanding of what they will receive is unaffected, which is the test that matters.
Where it is not
- Generating the product itself. A shape, colour or finish that came from a model rather than from the item. This is a picture of something you do not sell.
- Changing the packaging or label. Including "cleaning up" a label into text that is subtly different from the real one.
- Implying a result. A skin texture after using a cream, a room after using a paint, a meal the ingredients do not produce.
- Inventing scale. A bottle that reads as 100 ml when it is 30.
- Fabricating people as customers. A generated person appearing to endorse or use the product, presented as real.
The first four are misleading about the product. The fifth is misleading about who recommends it, which EU rules treat at least as seriously.
What the rules actually require
The Unfair Commercial Practices Directive prohibits commercial practices that deceive or are likely to deceive the average consumer about the main characteristics of a product — which explicitly includes its nature, composition and the results to be expected from its use.
Two practical consequences:
- Intention is not the test. Whether the image is likely to deceive is, and "the model did it" is not a defence.
- Image edits are not exempt because everybody does them. An industry norm that misleads is still a misleading practice.
Separately, the AI Act introduces transparency obligations around certain generated content. The safe posture, and the one that ages well, is to be able to say what was generated and to never generate the product.
The four checks before publishing
Every generated image should be looked at by a person, in this order:
- Is the product identical to the real one? Colour, shape, label, finish. Compare side by side with the source photo, not from memory.
- Is any text real? Generated lettering is the most common and most visible failure. Add text afterwards, as text.
- Does the scale read correctly against anything else in frame?
- Does the image claim anything? A result, a context, an endorsement. If it does, can you support it?
A workflow that cannot show the original and the generation together will approve things a person would have rejected, which is an argument about tooling rather than about care.
A practice that holds up
- Product from photography, always. Generation places and lights it; it never creates it.
- Prices, sizes and claims from the shop's own data, added as text, never baked into a picture.
- Keep the source photo linked to the output so any image can be traced back to what it was made from.
- Nothing publishes without a person approving it. Not as a policy statement — as a step the system actually enforces.
Sources and further reading (4)
- EU — Directive 2005/29/EC on unfair commercial practices
- EU — Regulation 2024/1689 (AI Act)
- European Commission — Consumer rights and complaints
- Google — Product structured data (image guidance)
Checked on 23 September 2026. Provider prices, mailbox rules and legal guidance change — verify anything you plan to act on.
Side by side, before anything is published
Auralata generates settings around your own product photograph, shows the original and the result together, and publishes nothing until a person approves it — with prices and product details read from WooCommerce rather than drawn.