Do I need to label AI product photos?

The EU rule: Article 50 of the AI Act

Article 50(4) of the AI Act says that a business which uses AI to create or change a “deepfake” image must disclose that the image was artificially generated or manipulated. It has applied since 2 August 2026. The Digital Omnibus on AI, agreed this summer, gave AI tool makers extra time for some of their own duties. It did not delay this one.

A deepfake, in the Act’s definition (Article 3(60)), is AI-made or AI-changed content that looks like real people, objects, places or events and would falsely appear to someone to be authentic. “Objects” includes consumer goods, so product photos are covered.

Some details that catch stores out:

• You are responsible if your team made the image. A store whose staff or freelancers use Midjourney, ChatGPT, Firefly or Photoshop Generative Fill is the “deployer” and owes the label. If you commission an agency and have no say over whether it uses AI, the agency is the deployer (Guidelines, paragraph 14). Ask agencies to tell you anyway.
• Stores outside the EU can be covered. The Commission’s view is that posting images on the open internet, where EU shoppers can see them, is enough (Guidelines, paragraph 13). A UK or US store that ships to the EU should assume the rule applies.
• Older images are not caught. Images made before 2 August 2026 do not have to be labelled, though the Commission encourages it (Guidelines, paragraphs 153 and 154).
• Fines are capped by size. The maximum is EUR 15 million or 3% of worldwide turnover (Article 99(4)). For small and medium-sized businesses it is whichever of the two is lower (Article 99(6)).

What counts, using the Commission’s own examples

The European Commission published guidelines on Article 50 on 20 July 2026. They are not binding (only the EU Court of Justice can give a binding reading), but they are how regulators will read the rule, and they include examples from advertising and product photography.

The Guidelines say these are deepfakes:

• An AI image of a product that could mislead about how it looks, what it is like or what it does, for example making it look different from the real product, more appealing, or better quality than it is.
• A realistic AI-generated person, such as a model, a “customer” or an influencer. The Guidelines treat realistic AI avatars as persons (paragraph 113).

And these usually are not:

• A real product (their example is a car) shown against an AI-generated background, as long as the ad does not mislead about the product.
• AI colour correction, background extension, background replacement for clearly aesthetic purposes, arranging existing products, or re-scaling (paragraph 116).
• Obviously unrealistic images. Their example is AI-generated mice arguing about cheese in an advert for a cheese maker.

Two points from the Guidelines make the test stricter than it first looks. It is objective, so what you meant the image to do does not matter (paragraph 114). And it asks about the whole audience, including children, older people and people who are not confident online, not an imagined average shopper (paragraph 115).

Applied to common product images, our reading is:

There is a carve-out for “evidently artistic, creative, satirical or fictional” work, but it only changes how you disclose, not whether. The Guidelines say content that is plainly commercial is outside it (paragraphs 119 to 123), so it rarely helps with product images.

What the label has to look like

The Act says the disclosure must be “clear and distinguishable” at the first exposure at the latest (Article 50(5)). The Guidelines add:

• Shoppers must see it without doing anything. A label hidden behind a hover, a menu or the terms of use does not count (paragraph 142).
• Metadata on its own is not enough. You cannot rely on watermarks or Content Credentials in the file to tell shoppers (paragraph 117).
• First exposure applies to every person and every output (paragraph 143). In practice, treat each place the image appears as its own first exposure: product page, category grid, quick view and zoom.

The voluntary Code of Practice on Transparency of AI-Generated Content, published on 10 June 2026 and found adequate by the Commission, goes further on design. It recommends the letters “AI” as the main element of the label, placed where nothing else covers it, such as the top right corner of the image. An overlay that appears to sit on the image is acceptable. A second layer, opened on tap or hover, can say whether the image was generated or edited, and what changed. Labels should work with screen readers and keep high contrast.

The Commission also publishes EU icons for labelling AI content, free to use without attribution. Using them is optional, and it does not mean you have signed the Code.

The UK: no labelling law, but misleading images are banned

The UK has no equivalent of Article 50. The Advertising Standards Authority says plainly that there is no blanket legal requirement to disclose AI use in ads.

What does apply is the general rule against misleading consumers:

• The Digital Markets, Competition and Consumers Act 2024 (legislation.gov.uk) bans misleading actions (section 226) and misleading omissions (section 227). Since 6 April 2025 the CMA can fine directly, up to 10% of global turnover.
• The CAP Code covers marketing on your own website, product images included. You need to disclose AI use where leaving it out would mislead. CAP’s own example is an AI image showing cosmetic results the product cannot achieve: misleading even with an AI disclaimer. The ASA’s 2026 note, AI and deepfakes: four things advertisers need to know, repeats that the rules apply in full.

The government set up a taskforce on labelling AI-generated content in March 2026, with an interim report promised for autumn 2026. We have not seen it published yet. And if you sell into the EU, the EU rule above applies to you too.

A label does not fix a misleading image

In the EU, the Unfair Commercial Practices Directive (Articles 6 and 7) bans images that mislead shoppers about a product’s main characteristics, including how it looks. In the UK, the DMCC Act does the same. Both apply whether or not the image is labelled. The Commission’s Guidelines say the AI Act does not make misleading advertising acceptable just because it carries a label.

Other rules to check

• Google Shopping. Google’s image requirements ask for AI metadata on AI-generated product images and do not allow overlays or stamps on feed images. Put the label over the image on your site, never into the file you send to Google.
• Marketplaces. Amazon and Etsy have their own AI disclosure flags for sellers. Check their current seller policies.
• New York. Since 9 June 2026, ads that knowingly use AI-generated “synthetic performers” must say so (S8420-A).

A checklist for your store

1. List the product images made or changed with AI, including ones from agencies and freelancers.
2. For each, ask: was the whole picture made by AI? Was a real photo changed? Could it make the product look different from the real thing? Does it show a realistic person made by AI?
3. Fix any image that misrepresents the product. A label will not help.
4. Label the images that could pass as real. Decide whether to label other AI images too; some stores label all AI involvement for trust.
5. Put the label where shoppers see it without clicking, everywhere the image appears.
6. Keep a record of who decided, when, and why.

Sources

• Regulation (EU) 2024/1689, the AI Act, EUR-Lex. Articles 3(60), 50 and 99.
• Commission Guidelines on the transparency obligations under Article 50, 20 July 2026.
• Commission FAQ on Article 50.
• Code of Practice on Transparency of AI-Generated Content, 10 June 2026.
• EU icons for labelling AI-generated content.
• Unfair Commercial Practices Directive 2005/29/EC, EUR-Lex.
• Digital Markets, Competition and Consumers Act 2024, legislation.gov.uk.
• ASA: Disclosure of AI in advertising.
• ASA: AI and deepfakes, four things advertisers need to know.
• UK written statement HCWS1416, 18 March 2026.
• Google Merchant Center image requirements.
• New York Senate bill S8420.