WITHOUT.

Do I need to label AI images?

1. Could the image be mistaken for a real photograph?
2. Only technical editing (crop, noise, exposure)?
3. Context?
Answer all three questions to see the self-assessment.

Whether you should visibly label an AI image mainly depends on whether it could look like a real photograph, whether it was only technically edited and the context in which it appears.

  • Purely personal and non-professional, only minimally technically edited or clearly artificial: metadata could suffice as technical documentation.
  • Clearly artistic, satirical or fictional: a perceivable disclosure in the immediate context could be suitable.
  • Other deceptively realistic publication: an additional visible disclosure could be appropriate.

When is an image a deepfake?

The EU AI Act defines a deepfake as AI-generated or manipulated image, audio or video content that appreciably resembles existing or plausible persons, objects, places, entities or events and would falsely appear authentic or truthful. Not every AI image meets those elements. Context and the foreseeable audience may matter; the first question is only a conservative, simplified proxy.

What counts as purely technical editing?

This means minor, non-substantive steps such as cropping, small colour or exposure corrections, sharpening, compression, noise removal, format conversion or dust and red-eye cleanup. Adding or removing people or objects, replacing a face, major relighting or colour changes and composites can go further. If in doubt, “No” is the more conservative answer.

What applies to advertising and artistic works?

Advertising or information does not automatically trigger a visible disclosure merely because AI was used. A deceptively realistic image may still fall within the deepfake rules, and a label does not cure misleading advertising. Article 50 provides a less intrusive disclosure regime for evidently artistic, satirical, creative or fictional works, not a blanket exemption. The disclosure should remain perceivable without unnecessarily hampering enjoyment of the work.

Frequently asked questions

Do I need to label AI images?

Not every AI image would automatically need a visible label. Relevant factors could include whether it looks authentic, whether you use it professionally and whether only minor technical edits were made. This path provides only a self-assessment calculated from your answers.

What is the EU AI Act image labelling requirement?

Providers of generative AI systems have machine-readable marking and detection duties, while professional deployers may have to disclose deepfakes clearly and perceivably. Those are different roles and duties. Metadata added by a user can document processing but does not by itself replace a perceivable deepfake disclosure.

What are the EU AI Act disclosure rules for images?

Article 50 applies from 2 August 2026 and covers specific transparency duties rather than every use of AI. Deepfake disclosure should be clear, distinguishable, accessible and perceivable by the first exposure. Content created before that date generally does not have to be labelled retroactively under the Commission Guidelines.

Do I need to label AI-generated content on Instagram?

Article 50 is not limited to a particular platform. Whether an Instagram post is purely personal or professional, and whether its image could be a deepfake, depends on the particular use and audience. Current platform rules should be checked separately.

Tools

Open the AI labelling tool to carry over your answers, mark an image locally and save the self-assessment record as a PDF. View EXIF can inspect existing metadata locally.