Does the EU AI Act apply to a US or Canadian brand?

It can. The guidelines say a deployer outside the EU is covered when it directs AI content at people in the EU, including by posting deepfakes on the open internet. Content that reaches the EU only through channels you did not foresee or control is not your duty.

Does a platform’s AI label count as mine?

It can, inside that platform. The guidelines say deployers may rely on labeling tools that very large platforms provide, if they give a clear and visible disclosure. You stay responsible, so check that the label shows in every placement where the image appears. Each platform’s tools are covered in the ad labeling guide.

Do I need to label images made before August 2, 2026?

No. The guidelines say deepfakes generated before that date do not need a label after the fact, though labeling them is encouraged where it takes little effort. Anything generated on or after August 2, 2026 needs a label if it meets the four tests.

Who enforces this, and how large are the fines?

National market surveillance authorities, on their own initiative or after a complaint. Fines for breaking Article 50 reach €15 million or 3% of worldwide annual turnover, whichever is higher. For small and medium firms the lower of the two applies.

What does the EU AI Act require for AI marketing images?

Checked September 24, 20268 min read

Short answer

Since August 2, 2026, Article 50 of the EU AI Act requires any business that uses AI to make a realistic image of people, products, places or events that would pass as real to label it visibly as AI-generated. The Commission’s July 2026 guidelines count an AI product image that makes the product look different or better than it is, but usually not a real product photo on an AI background. Fines reach €15 million or 3% of worldwide turnover.

What applies now, and what was delayed

Article 50 of the AI Act (Regulation (EU) 2024/1689) sets transparency duties for AI-generated content, and it has applied since August 2, 2026. The Digital Omnibus on AI, adopted as Regulation (EU) 2026/1744 and in force since July 27, 2026, pushed back the rules for high-risk AI systems. It did not change the duty to label deepfakes.

Item	Date	What it means for images
Article 50 applies	August 2, 2026	Deepfakes shown to people in the EU must carry a visible AI label.
Grace period for machine-readable marks	Until December 2, 2026	AI tools already on the market before August 2, 2026 have until then to embed hidden marks in their outputs. It does not delay your label.
Commission guidelines on Article 50	Approved July 20, 2026	The Commission’s reading of who must label what, with advertising examples.
Code of Practice on Transparency of AI-generated Content	Final version June 10, 2026	A voluntary way to comply. The Commission and the AI Board confirmed it is adequate, and about 190 organizations had signed by the end of July 2026.
Fines	From August 2, 2026	Up to €15 million or 3% of worldwide annual turnover, whichever is higher. For small and medium firms, whichever is lower.

Status on September 24, 2026.

Providers mark files, deployers label images

The Act splits the job. A provider is the company that makes an AI system and offers it under its own name, such as an image model maker. Providers must embed a machine-readable mark in every output (Article 50(2)). A deployer is anyone who uses an AI system in their work, and deployers must visibly disclose deepfakes (Article 50(4)). A brand or agency making ad images with someone else’s tool is a deployer.

The guidelines answer two questions brand teams ask. The agency is usually the deployer: a company that only commissions an ad, without deciding whether or how the agency uses AI, is not one. A brand that decides the agency will use AI and controls how can be a deployer too. Being outside the EU does not help either. A company in a non-EU country is covered when it directs content at people in the EU, which the guidelines say includes posting deepfakes on the open internet.

The provider’s hidden marks do not count as your label. The guidelines say deployers cannot rely on them, because people cannot see them without tools.

When is a marketing image a deepfake?

The Act defines a deepfake as AI-generated or manipulated image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful (Article 3(60)). The guidelines turn this into four tests, and an image must meet all four.

Resemblance. The image closely resembles its subject. It does not need to be identical.
Existing. The subject exists or could plausibly exist. Realistic people, products and places count, including a realistic AI-generated person. A dragon or a person flying unaided does not.
Subject. People, objects (the guidelines name consumer goods), places, animals, and events such as someone using a product.
False appearance. A viewer could take it as authentic or truthful. Photorealism makes this likelier without settling it; context and audience decide, and intent to deceive is not required.

Image	Deepfake?
An AI image of a product in an ad or on packaging that shows it "not identical to the real product, more appealing or with improved quality than in real life"	Yes
A real product, such as a car, shown against an AI-generated background, if the ad does not mislead about the product	No
AI color correction, extending or replacing a background for looks, arranging existing products, or resizing, in product ads or packaging	Likely no, because the effect on authenticity is minor
An AI-generated celebrity influencer in an advertising video	Yes
A realistic synthetic influencer testing a sponsored real product	Yes, and not a creative work
A teleshopping-style video of AI people using a product to persuade viewers to buy it	Yes, and not a creative work
Mice arguing about cheese in a cheese maker’s ad	No, because it is not realistic

Examples from the Commission’s guidelines of July 20, 2026, in plain words.

Two points follow for product imagery; they are our reading of the guidelines. A photoreal AI model wearing your product counts as a realistic person, so an EU lookbook or ad built on AI models is likely to need a label. And the more an AI render improves on the real product, the closer it moves to the first row. See whether AI models need model releases.

The creative-work exception rarely fits an ad

Article 50(4) softens the duty for deepfakes that are part of an evidently artistic, creative, satirical or fictional work: the label can go where it does not spoil the work, such as in credits or beside the image. The guidelines read "evidently" strictly. Content that is purely commercial and recognizable as such falls outside it, and ads qualify only in specific situations, judged case by case. Even then a label is still required; it can sit in the credits or next to the image.

How to label: the Code of Practice

The Code of Practice gives the working details. It is voluntary, but the guidelines say signatories get a predictable route, while others should expect to explain their own measures, for example with a gap analysis against the Code.

Use an icon or label whose main element is the capital letters "AI". The EU publishes free icons in black, white and two semi-transparent versions. The icons are optional; the label is not.
You may add a second layer saying whether the image was "generated" or "modified", and what was changed.
Place it where nothing covers it, for example the top right corner of the image, and embed it in the image itself where you can.
Show it at first exposure. A label hidden in terms of use or a menu does not count as clear.
For video, show it at the start and, where possible, at intervals and after breaks.
Content generated before August 2, 2026 does not need a label after the fact, and you are not expected to change packaging already printed.

What a brand advertising in the EU should do now

List every AI-made or AI-edited asset running in the EU, with the tool and model behind it.
Decide who is the deployer for each one, your team or your agency, and write it into the agency contract along with who applies the label. See telling clients you used AI.
Run each image through the four tests, and record the answer and the reason.
Label every image that meets all four with an "AI" icon or label inside the image, visible at first exposure, in every placement.
Keep the provider’s hidden marks intact. The Code asks providers to forbid deployers, in their terms, from removing them on purpose. See Content Credentials.
Check the image against EU consumer law as well. A label does not make a misleading ad lawful, and the Unfair Commercial Practices Directive judges an ad by its overall presentation.
If a real person appears, get their consent. The label does not replace personality rights or data protection law; see AI images of real people.

Decision record for one EU image

Asset and placement:
Made or edited with (tool, model):
Deployer (our company or the agency, by name):
1. Resembles a real or plausible person, product, place or event? (yes/no, why)
2. Would a viewer take it as authentic? (yes/no, why)
3. Does it change how the product looks, works, or what it includes? (yes/no)
4. Evidently artistic or fictional? (yes/no, why)
Deepfake: yes/no
Label used, and where it sits:
Checked by, and date:

Questions people also ask

Does the EU AI Act apply to a US or Canadian brand?: It can. The guidelines say a deployer outside the EU is covered when it directs AI content at people in the EU, including by posting deepfakes on the open internet. Content that reaches the EU only through channels you did not foresee or control is not your duty.
Does a platform’s AI label count as mine?: It can, inside that platform. The guidelines say deployers may rely on labeling tools that very large platforms provide, if they give a clear and visible disclosure. You stay responsible, so check that the label shows in every placement where the image appears. Each platform’s tools are covered in the ad labeling guide.
Do I need to label images made before August 2, 2026?: No. The guidelines say deepfakes generated before that date do not need a label after the fact, though labeling them is encouraged where it takes little effort. Anything generated on or after August 2, 2026 needs a label if it meets the four tests.
Who enforces this, and how large are the fines?: National market surveillance authorities, on their own initiative or after a complaint. Fines for breaking Article 50 reach €15 million or 3% of worldwide annual turnover, whichever is higher. For small and medium firms the lower of the two applies.

Where Overs fits

Overs does not publish to ad accounts or stores; you export the files, so the label decision sits with you or your agency when an image goes out. Each export records the prompt, model and reference pictures behind every photo, which helps you fill in a decision record like the one above.

Free for 40 photos a month. The AI that makes the photos is billed separately, on your own key, with no markup from Overs.

Start free How it works

Sources

Checked on September 24, 2026. Prices, specs and rules change; follow the links for the current versions.