Skip to content
disclosura

Law explained

AI-generated images on your website: what the law wants from you

You made product visuals with an image generator, or edited photos with generative fill. What now?

Layer 1 — machine-readable marking (Art. 50(2))

Your image tool’s provider must ensure outputs carry machine-readable marks (metadata/watermark) — a duty that applies since 2 August 2026, with a transitional period until 2 December 2026 for tools already on the market before that date (2026 AI Omnibus). That duty is not yours, unless you built the tool, or had it built, and run it under your own name (Commission guidelines, section 2.3). Good practice on your side: use tools that mark, and keep the marks when you upload (many CMS strip metadata — the Kit’s marking note gives a short test: upload one file, download it again, and check the mark is still there).

Layer 2 — visible label for a deep fake (Art. 50(4))

A deep fake is AI-generated or manipulated image, audio or video content that “resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful” (Art. 3(60)). The Commission’s guidelines (section 6.1.1) read “existing” broadly: it is enough that the person, object or place exists, can plausibly exist or could plausibly have existed. So a realistic invented person, product, shop or street can count too, if it would falsely appear authentic. Out of scope: content that is clearly unreal — their examples include a sphinx flying over the Eiffel Tower.

If your image is a deep fake, it must be visibly disclosed as AI-generated or AI-manipulated. This duty is yours, the deployer’s, and the metadata from layer 1 does not count for it (section 6.1.2). A fantasy scene that could not pass as real, or an obviously abstract graphic, is not a deep fake: no visible-label duty under Art. 50(4).

Need the label itself? Our free EU AI-label icon picker gives you the Commission’s official icon for this case plus a copy-paste snippet — and says plainly which duty that snippet does not cover.

Product images: where the guidelines draw the line

The guidelines (section 6.1.1) give both sides for products:

  • A deep fake: an AI-generated product image in an ad or on packaging that can mislead about the real product’s appearance, characteristics or use — for example, making it look better than it is in real life.
  • Not a deep fake: a real product shown against an AI-generated background, as long as the ad does not mislead about the product itself. Colour correction, background swaps for clearly aesthetic purposes, arranging existing products and re-scaling are likely to have only a minor effect.

A photorealistic image is more likely to be a deep fake, but photorealism alone does not decide it (section 6.1.1): what counts is whether it could mislead your audience. Consumer-protection law also applies on top: an image that implies “this is a real photo of the real product” when it isn’t is a problem older than the AI Act. Our stance in the Kit: when in doubt, a small “AI-generated image” caption costs you nothing and reads as confidence, not confession.

Practical labels that don’t ruin design

For a deep fake, put the label on the picture, where it is seen at once without a click: the Commission’s code of practice on AI-generated content (voluntary) gives the top right corner as its example. A caption under the picture is the form for evidently artistic, creative, satirical or fictional work, and a fair choice for AI images that are not deep fakes, where a label is good practice. Ready captions per media type and a copy-paste label snippet are in the Kit.

Which of these duties apply to you?

Free, 60 seconds, instant result with exact citations. No email required.

Run the free check

Sources are cited in place; official text: Regulation (EU) 2024/1689. This guide is a technical explanation, not legal advice.