EU AI Act: Labelling AI images and videos

From 2 August 2026, key transparency duties under the European AI Act apply. A common question follows: must every image or video involving AI carry a visible label?

The short answer is no. Whether disclosure is required depends on who is acting, how substantially the content was changed and whether the result could falsely appear authentic.

This article provides practical orientation for businesses based on the legislation and current European Commission guidance. It is not legal advice.

When do the rules apply?

The Article 50 transparency duties generally apply from 2 August 2026, as confirmed by the Commission's questions and answers on the transparency code.

The Code of Practice published in June 2026 is voluntary. Article 50 itself is legally binding. In July 2026, the Commission and AI Board assessed the code as an adequate EU-wide instrument for demonstrating compliance, although signing it is not conclusive proof of compliance.

Two duties that are often confused

The official EU AI Act text distinguishes between providers of AI systems and organisations or people deploying those systems professionally.

1. Machine-readable marking by providers

Providers of systems generating synthetic text, images, audio or video must generally ensure that outputs are marked in a machine-readable format and can be detected as artificially generated or manipulated.

This provider duty is not the same as adding a visible badge to a social media post.

The Act also limits the duty where an AI system only assists with standard editing or does not substantially alter the input data or its meaning.

2. Visible disclosure by professional deployers

A professional deployer using AI to generate or manipulate image, audio or video must disclose its artificial origin when the result constitutes a deep fake.

The Act defines a deep fake as AI-generated or manipulated image, audio or video content resembling existing persons, objects, places, entities or events and falsely appearing authentic or truthful.

The disclosure must be clear and distinguishable no later than the first exposure.

When should images or videos be labelled?

For marketing teams, the effect of the result matters more than the name of the software.

Labelling will regularly be required

A fully AI-generated photorealistic kitchen is presented as a completed customer project.
A real photograph of an empty room is realistically furnished with AI and could be understood as the completed room.
A real person is replaced or convincingly reproduced by an AI avatar.
A person's voice, facial expressions or statements are artificially generated or substantially altered.
A real event, location or product is manipulated in a way that can create a false impression of reality.

The Commission's page on EU icons for AI-generated content specifically uses an authentic photo of an empty apartment furnished with AI as an example of partially AI-modified content.

Not automatically subject to visible labelling

Correcting exposure, white balance or contrast.
Cropping, resizing or adapting an image to a social format.
Removing noise, minor technical defects or insignificant distractions.
Creating subtitles or technically compressing a video.
Using AI assistance without substantially changing the depicted facts or meaning.

These actions may qualify as standard editing. Whether a specific edit crosses the threshold of substantial alteration still requires an individual assessment. Platform rules, advertising law, copyright and personality rights may impose additional requirements.

Not every AI-generated item is a deep fake

The Commission expressly explains that not all AI-generated or manipulated content requires visible labelling. For image, audio and video disclosure by deployers, the deep-fake threshold is central.

Where content forms part of an evidently artistic, creative, satirical or fictional work, disclosure may be designed so that it does not hamper the display or enjoyment of the work.

In practice, a clearly recognisable illustration differs from a photorealistic image claiming to document a customer project that never existed.

What should a label look like?

The AI Act requires clear and distinguishable information no later than first exposure. The voluntary Code of Practice and accompanying EU icons provide practical direction:

Show the disclosure at first exposure.
Do not allow overlays to cover it.
Where possible, preserve it when the content is shared or downloaded.
Use plain, understandable wording.
Make the information accessible.

Suitable wording can include:

Created with AI
Fully AI-generated
Partially modified with AI
AI-generated voice

The EU icons are free and optional. Using an icon alone does not establish legal compliance. The Commission notes that understanding improves when the icon is accompanied by a short text label.

Businesses should not wait until upload to decide whether labelling is needed. The decision belongs in the production workflow:

Document the source: customer material, camera capture or generative model?
Assess the alteration: technical enhancement or a newly created reality?
Check deceptive potential: could an average viewer mistake it for a real person, place or event?
Set the disclosure: apply visible wording, metadata and platform declarations consistently where required.
Keep evidence: retain source material, editing steps, approval and the final version.

Our approach at Virello

For Virello as an AI marketing agent, transparency is part of the content workflow rather than an afterthought. Routine optimisation such as exposure correction, cropping or careful cleanup is separated from substantive alterations. For photorealistic generation or material manipulation, artificial origin must remain traceable and, where the legal threshold is met, visibly disclosed.

The guiding principle is simple: AI may improve quality, but it should not quietly create a false reality.

Official sources

Last updated: 19 July 2026. This article provides general information and does not constitute legal advice.

EU AI Act: When must AI images and videos be labelled?