Label the Generated Images in Your Marketing

A generated lifestyle shot shows a person holding your product. The lighting looks photographic, and the image goes into an ad with the same layout your team uses for real photos. A viewer could take the scene as something that happened in front of a camera.

For certain systems already on the market, the transitional deadline for the provider marking duty under Article 50(2) is December 2, 2026, according to FPF's timeline. That extension concerns machine-readable marking.

Decide how the scene will be disclosed

Could a viewer mistake your generated scene for an authentic photograph?

Under Article 50(4) of the AI Act, deployers must disclose image, audio or video content that is a deepfake: generated or manipulated content resembling real people, objects, places or events that would falsely appear authentic. A photoreal scene of a person using your real product could well fall under it. A company selling into the EU should check its own position with counsel.

Work that's evidently artistic, creative, satirical or fictional gets lighter treatment in how disclosure appears. The Article allows it in an appropriate manner that preserves the display or enjoyment of the work. So assess the campaign's creative context alongside the image and its intended placement.

The person approving publication has to know the asset was generated. If that information is lost when the file changes hands, an ad can reach publication with nobody having assessed its disclosure.

The label has to reach the viewer

The disclosure is due by the first interaction or exposure, under Article 50(5). For an ad, that means in the ad itself.

Check the placement. When the image moves into another layout, its disclosure has to remain visible there. A crop can change which part of the presentation the audience sees, so the final ad is part of the check.

Jones Day's account of the Commission's voluntary Code of Practice notes that labels can use the publicly available EU icon or equivalent labels. Choose a treatment that works with the design, then check the published presentation against what was approved.

Let the asset carry its history

Record that the lifestyle shot was generated and which tool produced it. The approved disclosure in the asset record tells the next person using the file how it was cleared for publication.

That internal record travels with the asset through the team. A colleague reusing the image can see how it was made before deciding how to present it in another campaign.

Before the December deadline for existing systems, ask the image-tool providers your team uses how they'll mark outputs and what your publishing process has to preserve.

Test what survives the export

The law requires a machine-readable, detectable mark. The voluntary Code of Practice calls for at least two layers where necessary, such as metadata plus a watermark, as Jones Day explains.

Visible disclosure under Article 50(4) has applied since August 2, 2026, in FPF's timeline. The December transition above applies to the marking duty for certain existing systems.

An image's route to publication can include resizing or compression. Test whether the marks survive those steps with an asset from your own process, using the export settings for the ad.

Compare the file the tool supplied with the file you published. Preserve the original output as a reference for that comparison. If the expected marking is missing, examine the export settings and the provider's marking method. Check the visible disclosure in the same published presentation.

If the publishing process makes another resized version after upload, that version goes through the check too. The file visitors receive may have passed through more steps than the export the designer approved.