The EU AI Act and Your Product Content:What Actually Changes

Straight answers on what the EU's new AI transparency rules mean for AI-generated product images and text, and what, if anything, you need to do about it.

Since August 2, 2026, the EU AI Act's transparency rules (Article 50) require certain AI-generated content to be disclosed. If you're using Dyver to enrich your catalog, here's exactly where that applies, and where it doesn't.

The two things the law actually asks for

There are two separate obligations, and they land on different people:

The platform that builds the AI (that's us)

has to make AI-generated output detectable: invisible, machine-readable marking baked into the content itself.

The business that publishes the content (that's you)

only has to add a visible "this is AI-generated" label in one specific case: realistic images, video, or audio that could pass as a genuine, unaltered photo or recording of something real.

That second point is the one worth understanding, because it's narrower than most people assume.

Product images

Know the difference between "edited" and "deepfake"

The law's disclosure trigger is a legal concept called a deepfake: content realistic enough that someone could mistake it for an authentic, unedited photo of a real person, place, or thing.

Low risk, no label needed:

Removing or replacing a background
Cropping, color correction, minor retouching
Removing a small defect or blemish for cleanup purposes
Clearly stylized or illustrative graphics that nobody would mistake for a real photo

Higher risk, label your listing:

Fully AI-generated "lifestyle" shots that place your product in a photorealistic scene designed to look like a genuine photograph
Removing or altering a real, identifiable object, person, or element in a way that changes what the image appears to show

The rule of thumb: if a shopper could reasonably mistake it for an unedited photo, add a simple visible note, something as light as "AI-generated image" near the product photo. If it's obviously a rendered or stylized graphic, you're already in the clear.

Product text

This one's easy

Titles, descriptions, bullet specs, category copy: none of this needs a label. The law's text-disclosure rule only covers content published to inform the public on matters of public interest (politics, public health, safety, and similar topics). Ordinary commercial product content sits outside that entirely.

On top of that, if a person reviews AI-generated text before it goes live, which most enrichment workflows already do, that content is exempt from the labeling duty even in edge cases.

Bottom line: no changes needed to how you publish product copy.

What this means for Dyver

We're building compliance into the platform rather than leaving it to guesswork:

Machine-readable marking embedded in AI-generated images and text at the point of creation, so the technical obligation is handled automatically.
Clear guidance inside your workflow flagging which generated images fall into deepfake territory, so you know exactly when a visible label is worth adding.
Optional label tooling to make adding that disclosure to a listing a one-click step, not a manual process.

What this means for you

If you're only using Dyver for text enrichment, deduplication, categorization, or attribute mapping, no action needed. This content isn't in scope.
If you're using AI-generated or AI-edited product imagery, do a quick pass on realistic, scene-generated photos and add a simple visible disclosure where it applies. Straightforward background removal and minor cleanup don't require this.
If you're unsure about a specific image or workflow, reach out to your Dyver contact and we'll help you make the call.

This page is a plain-language summary, not legal advice. The EU AI Act's guidance and Code of Practice are still being finalized, and requirements may be refined. For a compliance determination specific to your business, we recommend speaking with legal counsel familiar with EU AI regulation.