How to label AI-generated images for EU audiences: the complete guide

Learn how to label AI-generated images for EU compliance: visible labels, machine-readable marking, and the official EU icons. Step-by-step for creators and…

· 4 min read

How to label AI-generated images for EU audiences

If you publish AI-generated images in the EU, you must label them with a visible notice and a machine-readable mark. This is required by the EU AI Act’s Article 50(2), which applies from 2 August 2026 to providers of generative AI systems. The label must be clear, distinguishable and accessible at first interaction. For images, this means a human-readable badge or text plus an embedded, tamper-evident marker such as C2PA or an invisible watermark. The EU provides a set of voluntary icons you can use for the visible part.

What counts as AI-generated and who must label it

The duty to mark synthetic images applies to providers of generative AI systems that place the output on the EU market. If you are a studio, agency or creator using such a system and you publish the image, you are acting as a deployer and must ensure the marking is present. The marking must identify the content as artificially generated or manipulated. There is a narrow exemption for business-to-business or industrial contexts, but most public-facing images must be marked.

Visible labels: text, badges and the EU icons

The visible label should be placed on or next to the image so it is immediately apparent. Use plain language such as ‘AI-generated image’ or ‘Image created with AI’. The EU has published three icons you can use: one for fully AI-generated content, one for partially AI-modified content and one for general AI involvement. These icons are voluntary but align with the legal requirement for a clear, distinguishable notice. You can download them from the Commission’s digital strategy pages.

Machine-readable marking: C2PA, watermarks and metadata

Article 50(2) requires that synthetic images are marked in a machine-readable format so they can be detected as artificially generated or manipulated. The marking must be effective, interoperable, robust and reliable as far as technically feasible. C2PA (Content Credentials) is an open standard that binds provenance and edit history to the file via a cryptographically signed manifest. Invisible watermarks can be embedded in the image data itself so the mark survives cropping or compression. Combining a C2PA manifest with an invisible watermark gives the strongest protection against stripping. Providers and deployers can rely on the Code of Practice on Transparency of AI-Generated Content, which recommends a multi-layered approach and digitally signed, time-stamped metadata.

Step-by-step: how to apply both layers

Start by generating or editing the image with a tool that supports C2PA. When you export, enable the Content Credentials option so the manifest is embedded. Next, add an invisible watermark using a tool that meets the robustness and interoperability criteria. Then, overlay or place next to the image a visible label: either a text notice or one of the EU icons. If you use a third-party certification service, ensure it issues a tamper-evident provenance certificate and embeds both a C2PA manifest and an invisible watermark. For platforms that strip metadata, the invisible watermark will still be detectable. Always check that the marking persists after upload and sharing.

When you do not need to label AI images

Content generated and published before 2 August 2026 does not need to be labelled retroactively. Artistic, creative, satirical or fictional works only need disclosure in a way that does not hamper the display or enjoyment of the work. The business-to-business or industrial exemption applies only to the marking duty and only when the output is used in those contexts. If in doubt, mark it: the safe approach is to apply both visible and machine-readable layers.

Platforms and practical tips

Major platforms have their own tools for labelling AI content, but these do not replace the legal duty. Meta, for example, requires labels on AI-generated images and has detection systems for unlabeled content. To stay compliant, add your own visible label and machine-readable mark before upload. Use a consistent workflow: generate with C2PA support, add an invisible watermark, apply a visible badge, then upload. Keep a record of the provenance certificate for each file. If you use a validator, test that the marking is still present after platform processing.

Checklist for EU-compliant image labelling

Confirm the image is AI-generated or manipulated. Add a visible label: text notice or EU icon, placed on or next to the image. Embed a machine-readable mark: C2PA manifest plus invisible watermark. Verify the marking persists after compression, cropping and platform upload. Use the Code of Practice as a reference for technical standards. Document the provenance certificate for each file. If the image is artistic or fictional, ensure the disclosure does not hamper display or enjoyment. For pre-2 August 2026 content, no retroactive labelling is required.

Frequently asked questions

Do I have to use the EU icons for AI-generated images?

No. The EU icons are voluntary, but they are designed to meet the Article 50 requirement for a clear, distinguishable label. You can use your own text or badge as long as it is immediately apparent and accessible.

What happens if I do not label AI-generated images?

Breaching Article 50 obligations can lead to fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. Lower ceilings apply to SMEs.

Can I rely on platform labels alone?

No. Platforms may have their own labelling, but the legal duty is on the provider or deployer to ensure the marking is present. You must add your own visible label and machine-readable mark before upload.

Is C2PA enough for machine-readable marking?

C2PA is a strong foundation, but its metadata can be stripped. Combining it with an invisible watermark ensures the mark survives platform processing and user actions like screenshots.

Are there any exemptions for creative works?

Yes. Artistic, creative, satirical or fictional works only need disclosure in a way that does not hamper the display or enjoyment of the work. However, most public-facing images should still be marked.

Put provenance on your next file

L'AItelier signs images and video with C2PA, adds an invisible watermark, and issues a certificate anyone can check. Your file is deleted as soon as the certificate is issued.

General information about the law as it stood when this was published, not legal advice for your situation.