On 2 August 2026, the transparency obligations in Article 50 of the EU AI Act took effect. Among other things, they require that AI-generated images, video and audio be marked as synthetic where they are made available to people in the EU, with penalties reaching EUR 15 million or 3% of worldwide annual turnover.
If you generate marketing visuals with AI and sell into the EU, this applies to you and it applies now. If your images are produced another way, it may not - and the distinction is worth understanding properly rather than guessing at.
This is an explainer, not legal advice. We are not lawyers, the rules are new, and how they apply to any particular business depends on specifics. Treat this as a map of the questions to ask, and get proper advice on your own situation.
What actually changed
Article 50 of the EU AI Act sets transparency obligations for certain AI systems. The parts that matter to marketers:
Synthetic content must be identifiable. Where an AI system generates image, audio or video content, that output needs to be marked as artificially generated or manipulated - in a machine-readable form, and visibly disclosed to people encountering it.
Deepfakes require disclosure. Content depicting real people or events that has been generated or manipulated needs to be disclosed as such.
The obligations began on 2 August 2026. The Act came into force earlier, with provisions phasing in over time; these transparency rules are among those now live.
Enforcement carries real weight. Penalties for non-compliance with transparency obligations run to EUR 15 million or 3% of global annual turnover, whichever is higher.
The European Commission has published guidelines and a code of practice covering marking and labeling, including a proposed taxonomy that distinguishes fully AI-generated content from AI-assisted content. That distinction is where most marketing work will actually land, and it is the least settled part.
Generated versus rendered: a distinction worth understanding
Not everything produced by software is AI-generated, and the difference is not a technicality.
AI-generated images are created by a generative model from a prompt or an input. The model produces pixels that did not previously exist, based on patterns learned from training data. Two runs of the same prompt give different results. This is what the transparency rules are aimed at, because the output is synthetic - it depicts something that was never photographed or drawn.
Rendered images are produced by compositing assets you supply according to a layout you defined. A template engine placing your logo, your background, your chosen font and a name from your database into positions you specified is doing the same kind of work as a word processor laying out a page. Nothing is invented. The output is deterministic - the same inputs give the same image every time.
On the face of it, the second category is not synthetic media in the sense the transparency rules address. No model generated the imagery; software arranged materials you already had.
This is the reasonable reading, and it is why template-based personalization sits in a different position from AI image generation under these rules. It is not a loophole - it is a genuine difference in what the software does.
Where it gets genuinely uncertain
Be careful, because most real marketing images are not purely one thing or the other.
AI-generated elements inside a rendered image. If you generated a background with Midjourney and then used it as a layer in a template, the finished image contains synthetic content. Whether that makes the composite subject to labeling, and whether it counts as "AI-assisted" rather than "AI-generated" under the Commission's taxonomy, is exactly the kind of question that needs an actual legal answer rather than a blog post's opinion.
AI-written copy rendered into an image. Text generated by a language model and then placed into a graphic. Different obligations attach to AI-generated text, and the interaction with image rules is not obvious.
AI-enhanced photography. Background removal, upscaling, generative fill. There is a spectrum here from routine editing to substantial synthesis, and the line is not crisply drawn.
Stock imagery of unknown provenance. Increasingly, stock libraries contain AI-generated material. If you licensed an image without knowing how it was made, you may not know your own position.
That last one is worth acting on regardless of your view of the rest: it is worth knowing where your visual assets came from.
Practical questions worth asking
Not a compliance checklist - a starting point for a conversation with someone qualified.
Do we sell into or market to the EU? The obligations attach to content made available to people in the EU, so this is not only a question for EU-based companies.
Which of our images involve a generative model at any stage? Ads, social graphics, product imagery, blog headers, email visuals. Most teams have not audited this and will be surprised.
Do we know the provenance of our stock imagery? Including anything licensed in the last two years.
Who signs off on visual assets? If nobody owns this question, it will not get answered.
What does our tooling actually do? Worth asking vendors directly whether their output involves generative models. "AI-powered" appears in a great deal of marketing copy without meaning that images are model-generated - and in some cases it does mean exactly that.
Where OKZest sits
Since the question will come up: OKZest is a template rendering engine. You design a layout, define which parts are dynamic, and each image is produced by placing your assets and your data into that layout. No generative model produces the imagery.
That is a narrow claim and we will not stretch it. If you place an AI-generated background into an OKZest template, the finished image contains AI-generated content and your position depends on that, not on us. And nothing about a rendering engine addresses AI-generated copy, AI-edited photography, or any other part of your stack.
What we can say plainly is that the personalization itself - your design, rendered with a recipient's name in it - does not involve generating synthetic imagery.
Frequently asked questions
Do I have to label AI-generated marketing images in the EU? The EU AI Act's Article 50 transparency obligations, in effect since 2 August 2026, require AI-generated image, audio and video content to be marked as artificially generated and disclosed to people encountering it. How that applies to your specific content is a question for legal advice.
When did this take effect? 2 August 2026. The AI Act entered into force earlier with obligations phasing in; these transparency rules are among those now applicable.
What are the penalties? Up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher, for non-compliance with transparency obligations.
Does it apply if my company is outside the EU? The obligations attach to content made available to people in the EU rather than only to EU-based companies. If you market or sell into the EU, it is a question worth asking.
Are template-generated personalized images AI-generated? A template engine composites assets you supply into a layout you defined - it does not generate imagery with a model. On the face of it that is a different thing from synthetic media. But if any element inside the image was AI-generated, that element is what matters.
What if I use an AI-generated background inside a template? Then the finished image contains AI-generated content, and how the rules treat that composite - including whether it falls under "AI-assisted" in the Commission's taxonomy - is genuinely unsettled and worth proper advice.
What about AI-enhanced photos - background removal or generative fill? There is a spectrum from routine editing to substantial synthesis and the line is not crisply drawn. Generative fill that invents new content sits closer to the synthetic end than a background removal does.
What should I do first? Audit where generative AI appears anywhere in your visual production, including stock imagery whose provenance you may not know, and give the question an owner. Then take advice on what your specific situation requires.
Know what your images are made of
The practical takeaway is not about any particular tool. It is that a lot of marketing teams cannot currently answer the question "which of our images were generated by AI?", and that has become a question with a deadline attached.
Auditing it is a morning's work and worth doing before someone asks.
OKZest renders images from templates you design, using assets and data you supply. If you want personalized visuals without introducing generative imagery into your pipeline, that is what it does.
Create a free account, no credit card required.
This article is general information about a new regulation, not legal advice. Take advice on your own circumstances.
Sources
- Guidelines on transparency obligations for providers and deployers of certain AI systems - European Commission
- Commission publishes second draft code of practice on marking and labelling AI-generated content - European Commission
- The EU AI Act's Transparency Rules: A Practical Guide to Article 50
- European Commission adopts final Guidelines on AI Act Article 50 transparency obligations - Bird & Bird
- The AI Act's Transparency Obligations: Rules, Scope and Timeline - Stibbe
Related reading: AI image generation vs template-based generation, personalized images explained.