AI labelling and the EU AI Act: what creators should know

The EU AI Act (Regulation (EU) 2024/1689) entered into force on 1 August 2024 and applies in stages. Its transparency rules for AI-generated content are in Article 50, which applies from 2 August 2026. EU institutions have discussed adjusting some AI Act deadlines, so check the current official text before you rely on a date.

Two different duties

Article 50 splits the work between the companies that make AI tools and the people who publish AI content.

• Providers (the companies offering the AI system) must make sure generated images, video, audio and text are marked in a machine-readable way, so they can be detected as artificially generated or manipulated (Article 50(2)).
• Deployers (people and organisations who use an AI system in a professional context) must disclose deepfakes: realistic AI images, audio or video that look like real people, objects, places or events and could falsely appear to be authentic (Article 50(4)).

Purely personal, non-professional use falls outside the definition of a deployer. If you publish as a creator, business or brand, assume the disclosure duty applies to you.

What counts as a deepfake?

The Act defines a deepfake as 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. A photorealistic street scene that never happened can qualify, not only fake videos of famous people.

• Probably a deepfake: a realistic photo of a real politician at an event they never attended; a realistic "news" video of a flood that didn't happen.
• Probably not: a clearly stylised anime portrait, a claymation scene, an obviously impossible fantasy image.
• In between: a realistic AI product photo or influencer post. Disclosing it is the safe choice.

The artistic exception

When the content is part of an evidently artistic, creative, satirical or fictional work, the duty is lighter: you still disclose that AI-generated or manipulated content is present, but in a way that doesn't spoil the display or enjoyment of the work. A line in the caption or credits is a common way to do this.

AI-written text

Text generated or manipulated by AI and published to inform the public on matters of public interest must also be disclosed, unless a person has reviewed or edited it and a person or organisation holds editorial responsibility for it.

How results from this studio are marked

Images made in the studio get machine-readable metadata following the IPTC standard (Digital Source Type "trained algorithmic media") plus the name of the tool. The pixels are not changed. Some model makers add their own marks, such as invisible watermarks or content credentials (C2PA). The details are on our AI disclosure page.

Metadata can get lost when a file is edited, screenshotted or uploaded to a platform that strips it. That is why the visible disclosure by the person publishing matters.

Platform labels

Big platforms have their own rules, and they apply worldwide, not just in the EU. Instagram and Facebook show an AI label, TikTok asks creators to label realistic AI content, and YouTube asks creators to disclose realistic altered or synthetic content when uploading. Rules change often, so check each platform's current help pages.

A practical routine

1. Keep the original files, so the metadata stays intact where possible.
2. Turn on the platform's AI label whenever the content is realistic.
3. Add a short line in the caption, such as "Made with AI" or "AI-generated image".
4. For a virtual influencer, state clearly in the bio that the character is AI.
5. Never use AI to make real people appear to say or do things they didn't, unless they agreed and it is clearly disclosed.
6. Don't remove or hide AI marks to make content look real. It is also against our acceptable use policy.

What if you don't?

National authorities enforce the AI Act. For breaches of the transparency obligations the Act allows fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher, with lower caps for small businesses. Beyond fines, undisclosed deepfakes can break other laws, such as rules on defamation, consumer protection and personality rights.