Deepfake labelling sounds like a rule for bad actors, but it applies to anyone who deploys an AI system to produce image, audio, or video content that resembles real people, places, or events. Marketing teams generating realistic product scenes, agencies producing AI voiceovers, and publishers using synthetic imagery all fall inside it.
What counts as a deepfake here
In the Act, a deepfake is AI generated or manipulated image, audio, or video content that resembles real persons, objects, places, entities, or events and would falsely appear to a person to be authentic or truthful. The test is resemblance to reality, not malice. Synthetic content that is clearly artistic or obviously stylised sits differently, but realistic synthetic media of real looking scenes is squarely in scope.
What the label has to do
The deployer must disclose that the content has been artificially generated or manipulated. The disclosure has to be clear and distinguishable, given at the latest at the first exposure. For creative or fictional work there is a lighter touch: the disclosure should not spoil the experience, but it still has to be there.
Deepfakes vs the general synthetic marking rule
Do not confuse two obligations. Article 50(2) asks providers to mark generative output in a machine readable way (think embedded metadata). Article 50(4) asks deployers to disclose deepfakes to the viewer in a way they can see. One is under the hood, the other is on the surface. If you publish realistic AI media, you care about the visible disclosure.
Making it routine
The reliable pattern is a reusable label you apply wherever synthetic media appears, in the viewer's language, at first exposure. Handling it per asset by hand is where labels get forgotten.
Common questions
Does an obviously cartoon style AI image need a deepfake label?
The rule targets content that could pass as authentic. Clearly stylised or artistic work is treated more lightly, but realistic synthetic media of real looking people or events is in scope.
Who applies the deepfake label, the tool or me?
The deployer, meaning you, when you publish the content. The generating tool may add machine readable marks, but the visible disclosure to viewers is your duty.
Label synthetic media consistently
The paid plans include reusable labels for AI generated content and deepfakes, localized and applied at first view.
See pricing