Updated September 2026. Reading time: about 6 minutes. General information, not legal advice.
Nobody in fashion thinks of their campaign imagery as deepfakes. The word belongs to election interference and celebrity scandals, not to a lookbook. But deepfake is not an insult in European law. It is a technical category with a definition, and the definition was written wide enough that a large share of commercial AI fashion imagery falls inside it.
This matters because the category carries a duty: content classified as a deepfake must be visibly labelled by whoever publishes it. Since 2 August 2026, with no transition period. So the question "is this image a deepfake?" is not philosophical. It decides whether your published image is missing a legally required label right now.
The definition, taken apart
The law defines a deepfake as AI-generated or manipulated content that resembles existing persons, objects, places or events, and would falsely appear to a person to be authentic or truthful. Three elements do the work.
- Resemblance. The content looks like a person, object or scene from the real world. The Commission's final Guidelines read this broadly: a photorealistic person who does not exist is still covered, because the image resembles someone who plausibly could exist. The synthetic model defence, "she is not a real person," fails on exactly this point.
- Plausibility. Stylised, illustrated or obviously artificial content falls outside: a sketch, a cartoon, an avatar that no one would take for photography. The closer your output sits to photorealism, the more firmly it sits inside.
- Mistakable for authentic. Would a viewer take it for an authentic photograph? Commercial fashion imagery is built to be believed. That craft is precisely what completes the classification.
Why "it's creative work" does not save advertising
The AI Act softens the labelling duty for evidently artistic, creative, satirical or fictional work. Fashion imagery is creative work, so many teams assumed the exception covers them. The final Guidelines closed that door for commercial content: where content mixes information and creativity, the informative character prevails. An advertisement informs a buyer about a product, whatever its artistic merit, and every advertising example in the Guidelines carries a label. Editorial and artistic projects can still argue the exception. A product page cannot.
The five image types
Applied to the imagery fashion actually produces, the classification sorts into five recurring cases.
| Type of image | Deepfake under Article 50(4)? | What the publisher must do |
|---|---|---|
| Product on an AI-generated background | No | Nothing extra. Machine-readable marks in the file still apply. |
| Synthetic photorealistic model wearing a garment | Probably yes | Add a visible label (EU icon or equivalent). |
| Virtual try-on applied to a photo of a real model | Yes | Add a visible label. |
| Photography with retouching only | No | Nothing extra. |
| A render that makes the product look better than reality | Yes, explicitly | Label it, and treat it as a consumer-protection question too. |
Two boundaries in this table deserve a closer look, because they are where teams get surprised in both directions.
The relieving surprise: an AI-generated background behind a real product photo is generally not a deepfake. The product is real, the person is real or absent, and the scene-setting does not turn the image into a false depiction of a person or event. It still needs machine-readable marks inside the file, but no visible label. The full operation-by-operation breakdown is the subject of the next piece in this series.
The unwelcome surprise: the flattering render. The Guidelines' own examples treat a render that makes the product look better than the reality a buyer will receive as a deepfake requiring a label. And the same image walks straight into consumer-protection law on misleading advertising, which existed long before the AI Act. Render fidelity, how far you let generation drift from the physical garment, has quietly become a legal setting, not a taste setting.
What the label has to be
A label that satisfies the duty is clear, distinguishable and visible at the first interaction: someone scrolling past should not have to expand a caption or visit a details page to learn the image is AI-generated. The European Commission publishes official icons for this purpose, free to use, in three versions, with no attribution required, designed to remain legible against any background. A brand can use its own equivalent marking, but the official icons are the option no authority will argue with. We will publish a practical placement guide, channel by channel, in October.
The three pushbacks, answered
- "Our model does not exist, so no one is depicted." That is the argument the Guidelines pre-empted: resemblance to a plausible person is enough. Nonexistence is not a defence; photorealism is the trigger.
- "Everyone knows ads are constructed." Context helps but does not classify. The test is whether the image itself would be taken for authentic photography, and commercial imagery is engineered to be. The advertising examples in the Guidelines are labelled for exactly this reason.
- "Our platform already marks the images." Those marks satisfy the platform's duty, not yours. The Commission has stated explicitly that a publisher cannot rely on the provider's machine-readable marking to meet the visible-label obligation.
What to do with your catalogue
- 1. Sort your published AI imagery into the five types. Most catalogues concentrate in two or three of them, so this is an afternoon, not a project.
- 2. For everything in the yes rows: add the visible label to your publishing workflow now. The duty has applied since August; every day is exposure.
- 3. Write the render-fidelity rule down. Decide how far generation may improve on reality, and make it a documented production setting. It protects you under two bodies of law at once.
The complete legal framework, with dates, duties and the FAQ, is in our Article 50 guide.
