Labelling AI images: what the EU AI Act asks of brands since August 2026
When an AI image needs a label, who has to add it and how to do it properly.
By The Vela Team
Short answer: since 2 August 2026, brands must label AI images when they are deepfakes: realistic images that resemble existing people, products, places or events and could falsely appear real. For product imagery it comes down to one question: does the image mislead about what your product looks like, what it does or how it is used? A real product in an AI-generated setting generally does not need a label. A product that looks nicer, bigger or better than it really is does. So do realistic AI-generated people promoting your product.
This article is a practical summary, not legal advice. If you are unsure about a specific campaign, check it with a lawyer.
What does Article 50 of the AI Act say?
Article 50 of the EU AI Act sets out the transparency obligations for generative AI. For brands producing images and video, two paragraphs matter (full text of Article 50):
- Paragraph 2 (providers): providers of AI systems that generate images, video, audio or text must ensure the output is marked in a machine-readable format as artificially generated. That is the invisible layer: watermarks and metadata. This duty sits with the makers of the AI tools.
- Paragraph 4 (deployers): anyone using an AI system to create image, audio or video content that constitutes a deepfake must visibly disclose that the content has been artificially generated or manipulated. That is the duty that can fall on you as a brand.
These obligations have applied since 2 August 2026. The European Commission stresses that the accompanying code of practice is voluntary, but the transparency requirements of Article 50 are legal obligations (European Commission, Code of Practice on Transparency of AI-generated Content).
What is a deepfake under the AI Act?
Under the AI Act, a deepfake 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.
The word suggests fake videos of politicians, but the definition is broader. A product is an "object" too. That is why the guidelines the European Commission published on 20 July 2026 (guidelines on the transparency obligations under Article 50, C(2026) 5054) explicitly address product images in advertising and packaging.
Do AI product images need a label?
The Commission's guidelines give concrete examples. Summarised for product content:
| Situation | Deepfake? | Label needed? |
|---|---|---|
| Real product against an AI-generated background or setting, without misleading about the product | No | No |
| Colour correction, extending or replacing backgrounds for aesthetic purposes, arranging products, re-scaling | Probably not (minor impact) | Usually not |
| AI image showing the product as different, more appealing or of better quality than in reality | Yes | Yes |
| Realistic AI person or synthetic influencer testing or promoting your product | Yes | Yes |
| Clearly fantastical scene nobody takes for real (e.g. talking animals in an ad) | No | No |
Source: European Commission guidelines on Article 50 of the AI Act, paragraph 116 and the examples that follow.
The principle is simple: the setting can be generated, the product has to be accurate. As soon as an image makes a customer believe something untrue about your product, you have a labelling obligation, and frankly a problem with misleading advertising and returns as well.
Watch out for AI people and UGC-style videos
UGC-style videos, where a realistic person unboxes or uses your product, are popular in ads. If that person is AI-generated and looks real, the guidelines treat it as a deepfake. The Commission literally gives the example of a realistic synthetic influencer testing a sponsored product. Such a video needs a visible label, ideally right at the start.
Who has to label: the brand or the agency?
The obligation sits with the deployer: whoever uses the AI system under their own authority. The guidelines draw a useful distinction. A company that merely commissions an advertising agency, without deciding whether and how that agency uses AI, is not a deployer. If you create images yourself in an AI studio, you are.
In practice, brands and studios often work on the same images together. So agree in advance:
- who assesses whether an image is a deepfake;
- who adds the label;
- what the label looks like on each channel.
How do you label AI images correctly?
The AI Act does not prescribe wording, but it does set two conditions: the disclosure must be clear and distinguishable, and it must appear at the latest at the moment of first exposure. The guidelines give the start of a video and the moment someone encounters a deepfake while scrolling social media as examples.
In practice, this works well:
- Image in a feed or ad: a short, legible label in the image itself ("Image created with AI") or in the first line of the copy.
- Video: the label in the first few seconds, not tucked away in the end credits.
- Product page: a note next to the image, not hidden in the terms and conditions.
- EU icons: the European Commission provides free icons for labelling AI content. They are optional and do not in themselves prove compliance, but they make a label recognisable (European Commission, EU icons for labelling AI-generated content).
- Platform labels: also use the AI labels offered by Meta, TikTok and others, but do not rely on them to meet your obligation.
Note: metadata alone is not enough for a deepfake. Machine-readable marking is the tool provider's duty; your duty under paragraph 4 is a visible disclosure for the person seeing the image.
What is the risk of not labelling?
Under Article 99 of the AI Act, breaches of the Article 50 obligations can lead to administrative fines of up to 15 million euros or 3% of worldwide annual turnover, whichever is higher (Article 99, AI Act). Those are maximums; how enforcement plays out in practice is still taking shape. Rules on unfair and misleading commercial practices continue to apply as well.
For most brands, though, the biggest risk is not the fine but trust. A customer who receives a product that looks different from the image does not come back.
Checklist: is your AI image ready to publish?
- Does the image show the real product, with the right colour, shape, proportions and label?
- Does the setting avoid promising any function or use the product does not have?
- Is there a realistic AI person in the image? Then label it.
- Is the label visible at first exposure?
- Is it agreed who assesses and labels: brand or studio?
- Do you keep the source photo of the real product so you can show the image is accurate?
At Vela, every image starts from your own product photos, and your labels and colours are learnt from your own files. The product stays accurate, which is exactly the line the guidelines draw. More on how we protect brand consistency in Brand consistency at generative scale.
AI content that stays true to your product
Want AI images that show your product accurately, in every format and in line with EU regulation? Send us one product line and within a week we will show you what Vela makes of it.
Frequently asked questions
Do I need to label every AI-generated image?
+
No. Under Article 50(4) of the AI Act, a brand only has to visibly disclose deepfakes: realistic AI-generated images, video or audio that resemble existing people, objects, places or events and could falsely appear real. A clearly imaginative or illustrated image does not need a label.
Is an AI product photo a deepfake?
+
It depends on whether the image misleads about the product. According to the European Commission's guidelines, an AI image that shows the product as different, more appealing or of better quality than in reality is a deepfake. A real product against an AI-generated background is not, as long as it does not mislead about the product's appearance, characteristics or use.
Since when does the labelling obligation apply?
+
The transparency obligations in Article 50 of the AI Act have applied since 2 August 2026. The European Commission states that they are legal obligations, while the accompanying code of practice is voluntary.
Who is responsible for the label: the brand or the agency?
+
The deployer, meaning whoever uses the AI system under their own authority. According to the Commission's guidelines, a company that merely commissions an advertising agency, without deciding whether and how the agency uses AI, is not a deployer. So agree on labelling in your contract.
Where and when must the label be visible?
+
At the latest at the moment of first exposure, in a clear and distinguishable way. For a video that means at the start; for an image in a social feed, the moment someone comes across it while scrolling.
What is the risk of not labelling a deepfake?
+
Under Article 99 of the AI Act, breaches of the Article 50 obligations can lead to administrative fines of up to 15 million euros or 3% of worldwide annual turnover, whichever is higher. Rules on misleading advertising also continue to apply.
Ready to ship content that catches the wind?
Book a live intro and see the studio in action.
