Labelling AI images in e-commerce: what applies in 2026

Who has to label what?

What the AI Act requires

Provider or deployer?

According to the European Commission's guidelines of 20 July 2026, a provider offers an AI image tool in the EU under its own name. A deployer uses such a tool professionally and under its own authority; employees and contracted freelancers do not count separately. A company that merely commissions an agency, without deciding on its use of AI, is not a deployer. Most shops are therefore deployers.

Machine-readable marking (Article 50(2))

Providers must ensure that their system's outputs are marked in a machine-readable format and detectable as artificially generated, and as far as technically feasible effective, interoperable, robust and reliable. The guidelines (point 73) list watermarks, metadata, cryptographic provenance methods, logging and fingerprints, alone or combined.

Exempt are standard editing and changes that do not substantially alter the input data or its meaning. Examples from the guidelines (point 92): minor cropping, colour and brightness correction, sharpening, deleting a background visible in the original. Marking is required when objects or persons are inserted, replaced or removed in a way that changes the message, and for composites. The safe choice is to keep every newly generated image element marked, including an AI background.

Visible disclosure of deep fakes (Article 50(4))

Under Article 3(60) AI Act, a deep fake is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to be authentic. The guidelines (points 113 to 116) count consumer goods as objects and give retail examples:

• Deep fake: an AI product image in advertising or on packaging in which the product looks different, more appealing or of higher quality than in reality.
• Deep fake without the artistic exception: a realistic synthetic influencer testing a real product (examples after point 124).
• Not a deep fake: a real product in front of an AI background, as long as the ad does not mislead about appearance, characteristics and use.

Photorealistic AI models wearing your product will usually meet the definition. The notice must be perceivable by people without technical tools; the provider's metadata expressly does not suffice (point 117). It must be clear and distinguishable and appear at the latest at first exposure (Article 50(5)).

Deadlines after the Digital Omnibus

Under Article 113 AI Act, Article 50 has applied since 2 August 2026. Regulation (EU) 2026/1744 (the Digital Omnibus on AI), in force since 27 July 2026, changes only one thing: the new Article 111(4) gives providers whose systems were placed on the market before 2 August 2026 until 2 December 2026 for the marking under paragraph 2. Deep fake disclosure has not been postponed. Content generated before 2 August 2026 does not need retroactive labelling (guidelines, point 154). Fines under Article 99(4)(g): up to EUR 15 million or 3 percent of worldwide annual turnover.

Code of Practice and guidelines

The Commission published the final Code of Practice on marking and labelling AI-generated content on 10 June 2026. It is voluntary, has one section each for providers and deployers, includes an optional EU icon and has been confirmed by the Commission and the AI Board as an adequate tool to demonstrate compliance. The guidelines C(2026) 5054 are not binding.

The technology: IPTC and C2PA

• IPTC Digital Source Type: a metadata field with fixed values. trainedAlgorithmicMedia means “created with an AI model”, compositeWithTrainedAlgorithmicMedia stands for a photo augmented with generative AI (such as inpainting).
• C2PA (Content Credentials): an open standard backed by Adobe, Amazon, Google, Meta and Microsoft, among others. A digitally signed manifest records origin and edits and makes later changes evident.
• Weak spot: metadata is lost on re-encoding, in screenshots and in some shop systems. C2PA therefore also provides for invisible watermarks or fingerprints.

What Google, Amazon, Zalando and Meta require

Two official Meta sources contradict each other: the February 2025 announcement plans detection of images from third-party tools, while the current help page lists such ads as a case without a label. If your image is a deep fake, label it yourself.

Misleading claims: the image must show the real product

Regardless of the AI Act, a practice is misleading if it deceives about the main characteristics of the goods, such as execution, composition or specification (Article 6(1)(b) Directive 2005/29/EC; in Germany Section 5(1) and (2) no. 1 UWG). Both sets of rules apply side by side: even a correctly marked AI image can mislead. An image that makes the goods look better than they are has no place in the listing.

Checklist for AI product images

How the ecomDNAI image modules work

The ecomDNAI Image Generator always starts from your uploaded product photo: the product is placed into the scene you describe, not reinvented. Every generated image carries a machine-readable AI label and is marked as AI-generated in the asset library. Results of the Image Modifier are labelled as AI-generated there as well. Whether an image is a deep fake and shows the product accurately is for you to check before publishing.

Note: this article reflects the situation as of 24 October 2026 and is not legal advice.