MenuClose
  1. 01Does your persona look like a real photo? Could a viewer think it is a photograph of a person?

    Clearly stylised or physically impossible images are treated differently. The rule starts when an image looks like a real photo.

  2. 02Do you offer the content to people in the EU?

    What counts is where your audience is. Where you live does not matter.

  3. 03Do you resize, compress or re-encode your images before you post them?

    Most image optimisers remove embedded metadata by default. That includes the provenance data that shows where an image came from.

  4. 04Is the AI label switched on for every platform account your persona uses?

    Instagram, Fanvue and others each have their own label. Switching on one does not switch on the others.

  5. 05Was any part of your persona trained on images of a real person, or based on them?

    This includes a look "inspired by" someone. The key question is whether you have consent you can prove.

  6. 06Does a viewer see the AI label the first time they see your content, and not only in a bio or a footer?

    Art. 50(5) sets the timing. The label has to be "clear and distinguishable", at the latest the first time someone sees the content.

What applies to you

0 of 6 answered

Answer the questions above. The rules that apply to you will show up here.

    The full reasoning is in the EU compliance guide, with the article numbers and the date each rule started. See what changed and when on the changelog.

    Every result this check can give

    The questions above decide which of these apply to you. You can read all of them here without answering anything.

    AppliesYou have to label your images under Art. 50
    Photorealistic AI images of people count as deepfakes under Art. 3(60), as the Commission guidelines read it. That is true even when no real person is shown and you do not intend to deceive anyone. The rule has applied since 2 August 2026.
    AppliesThe rule applies to you
    The rule applies when you offer content to people in the EU. Where you are based does not matter. A US-based operation with EU subscribers is covered.
    Often missedYour AI label probably comes too late
    Art. 50(5) says the label has to be "clear and distinguishable" at the latest the first time someone sees the content. A line in a bio that a viewer never opens does not meet that. A visible mark on the image or video does.
    Often missedCheck whether your workflow removes provenance markers
    AI generators embed machine-readable provenance data in each file. The AI Act does not ban you from removing it. But platforms read it to label your posts automatically, and your generator's terms may forbid removing it. Most image optimisers delete metadata by default. Open one exported file in a manifest tool to check. It takes five minutes and almost nobody does it.
    Often missedPlatform labels are separate from the law, and from each other
    As of 31 August 2026, Instagram shows unlabelled AI persona accounts to fewer people. Following the AI Act does not switch on a platform label. Switching on the label on one platform does not switch it on anywhere else.
    StopA real person in the training data is a much bigger risk
    If the look comes from a real person and you have no documented consent, labelling is no longer the main issue. You are now dealing with consent and personality rights. Several jurisdictions now have specific offences for intimate deepfakes made without consent. Get legal advice at this point. A guide is not enough.
    AlwaysThe rule that always applies
    Nothing that looks underage, in any context. In Germany, § 184b StGB covers sexual depictions of children, including realistic computer-generated ones. What is judged is the impression the image gives. A prompt, a stated age or a disclaimer changes nothing. This applies whatever you answered above.

    How to use the self-check

    1. Answer each question for one persona. If you run two, check each one on its own.
    2. Pick Not sure when you do not know. The check treats it like the risky answer, so you see the rule and can find out.
    3. Read the result from the top. Rules marked Stop come before anything else.
    4. Open the guide linked under the result for the article numbers and the dates.
    5. Use Copy link to this result to keep your answers. The link opens the check with the same answers ticked.

    A worked example

    A typical setup: a photorealistic persona with fans in the EU. You compress your images before posting. You switched the AI label on for Instagram but not on Fanvue. The persona is not based on a real person. Your only AI label is a line in your bio.

    01 Does your persona look like a real photo?
    Yes
    02 Do you offer the content to people in the EU?
    Yes
    03 Do you resize, compress or re-encode your images before you post them?
    Yes
    04 Is the AI label switched on for every platform account your persona uses?
    No
    05 Was any part of your persona trained on images of a real person, or based on them?
    No
    06 Does a viewer see the AI label the first time they see your content, and not only in a bio or a footer?
    No

    The check returns 6 results for this setup:

    • Applies: You have to label your images under Art. 50
    • Applies: The rule applies to you
    • Often missed: Your AI label probably comes too late
    • Often missed: Check whether your workflow removes provenance markers
    • Often missed: Platform labels are separate from the law, and from each other
    • Always: The rule that always applies

    The rule about a real person does not show, because the answer was No. The first thing to fix is the label in the bio. Put a visible mark on the image or video itself.

    Open this example in the self-check

    What the self-check is based on

    Article 50(4) of the AI Act. If you publish a deepfake, you have to say it is AI. A photorealistic AI image of a person counts, even an invented person:

    Art. 50(4): "Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated." Art. 50(5): "The information referred to in paragraphs 1 to 4 shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure."

    EU AI Act Art. 50(4) and 50(5) (opens in a new tab)checked

    The start date. The rule has applied since 2 August 2026, as the Commission's page on its Article 50 guidelines states:

    "Article 50 of the AI Act applies from 2 August 2026."

    European Commission guidelines on Article 50 transparency obligations (opens in a new tab)checked

    An invented person still counts. The guidelines read the rule broadly, as the EU compliance guide explains with its sources:

    The Commission's guidelines of 20 July 2026 read the definition broadly: the labelling duty applies without any intent to deceive, and it applies even where no real person is depicted.

    European Commission guidelines on Article 50 transparency obligations (opens in a new tab)checked

    Not sure counts as the risky answer. This is our rule, not the law's. A checklist that leaves out a rule you needed is worse than one that shows a rule you can rule out.

    Platform labels are a separate question. The check asks about them because people mix the two up. Each platform's own rule is in the AI label matrix.

    Who checked this, and when

    Gus, who runs AI OFM City, checked this page. The wording of Article 50 and the Commission's start date were read again against the sources below, and the worked example was run through the tool. Last checked: .

    If a figure changes, the page changes with it and the date above moves. How we check our sources.

    Questions people ask

    Does the EU AI Act apply to me if I live outside the EU?

    It can. What counts is whether you offer the content to people in the EU. Where you live does not matter. A US-based account with EU subscribers is covered.

    Since when do I have to label AI images in the EU?

    Since 2 August 2026. That is the date Article 50 of the AI Act started to apply. It covers photorealistic AI images and videos of people, including invented people.

    Is an AI label in my bio enough?

    Usually not. Article 50(5) says the label has to be clear and distinguishable at the latest the first time someone sees the content. Many viewers see a post without ever opening your bio. A visible mark on the image or video reaches them.

    Does a platform AI label count as the EU label?

    No. A platform label follows that platform’s own rules, and the EU rule is separate. You need both, and a label on one platform does not switch on the label anywhere else.

    No. It is a checklist an operator wrote from the published rules. It shows which questions to take to someone qualified. It cannot tell you that you follow the law.

    Post your result in #question-thread

    5,857+ members who run AI influencers

    EU AI Act self-check: answer the six questions above and your result shows here.

    Ask what to change first. Paste the link too, so people see the same result you do. The Discord is free, and you can read everything right away.

    Join the Discord