Does the EU AI Act Apply to My Jewellery Brand? Yes, Here is What Changed.
On 2 August 2026, the part of the EU AI Act that most jewellery brands can actually feel came into force. From that date, the European Commission gained the power to request documentation, run evaluations, restrict models, and issue fines of up to 3% of global annual turnover or €15 million against the largest AI providers. The obligations themselves are not new — they applied from August 2025 — but the enforcement machinery only switched on now.
For a small jewellery brand, the fines aimed at AI providers are not your problem. The transparency rules aimed at deployers are. If you use AI to make images, video, or copy that reaches EU customers, you are a deployer. This article explains what changed, why it matters for jewellery marketing specifically, and what to do before it becomes a dispute.
What the EU AI Act Actually Says About AI Content
The EU AI Act is the first broad law regulating artificial intelligence. It sorts AI systems into risk tiers and phases in obligations between 2024 and 2027. Most of it targets high-risk uses like medical devices and hiring tools. One article targets everyone who publishes AI content: Article 50.
Article 50 sets transparency obligations. In plain terms, it says people have a right to know when they are looking at AI-generated or AI-manipulated content. These obligations apply as of August 2026. Two parts matter for jewellery brands:
AI-generated output must be markable as AI-generated. Providers of generative AI systems have to build machine-readable marking into their outputs. That is mostly the AI tool's job, not yours.
Deployers who create deepfakes must disclose it. A "deepfake" here has a specific legal meaning. Under Article 3(60), it is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places or events and would falsely appear authentic.
That second point is where jewellery brands need to pay attention.
Why This Lands on Jewellery Brands Specifically
Jewellery is one of the most AI-photographed product categories on the market. The reason is simple economics. A model shoot with a real person, a real studio, and real diamonds is expensive and slow. AI product images and AI "models" are cheap and fast. So a lot of independent jewellery brands now run campaigns where the ring is real but the hand, the face, and the setting are synthetic.
Here is the concrete problem. If your AI-generated image shows a synthetic person wearing your jewellery, and that image is built to look like a real photograph of a real model, you are close to the deepfake definition — content that resembles an existing person and would appear authentic. The disclosure duty is triggered. Deployers must disclose deepfake content to the viewer, clearly and distinguishably, at the latest on first exposure, using visible or audible labels a normal person can perceive without special tools.
A few observable facts sharpen the picture:
The rules apply to providers and deployers outside the EU where the output is used in the EU. A UK or US jewellery brand selling to EU customers is in scope.
A UK marketing function generating synthetic content for EU campaigns engages Article 50(2), and where existing people are depicted, the deepfake labelling duty in Article 50(4). This is a direct mapping, not a theoretical one.
The final Code of Practice on marking and labelling of AI-generated content was published on 10 June 2026, and assessed as adequate by the Commission and the AI Board on 8 and 9 July 2026. Following it is the recognised route to showing compliance.
There is relief on timing for the provider-side marking duty. The AI Omnibus package gives providers of generative AI systems already on the market until 2 December 2026 to meet the Article 50(2) marking requirement. That buys the tool-makers time. It does not remove the deployer disclosure duty that sits with the brand.
The Carve-Outs That Actually Help You
The law is not asking you to stamp "MADE BY AI" across a fantasy render of a dragon wearing a necklace. Two carve-outs matter for creative work:
Clearly fantastical content is out of scope. The draft guidelines confirm that obviously impossible content — humans flying unaided, for example — falls outside the deepfake definition. A surreal, plainly artificial campaign image is treated differently from a photo-real fake.
Artistic and creative work gets a lighter duty. Where AI content forms part of an evidently artistic, creative, satirical or fictional work, the obligation is limited to disclosing the AI content in a way that does not spoil the enjoyment of the work. You are not forced to plaster a warning over the middle of the frame.
The trap sits in the middle: photo-realistic AI images built to pass as genuine photography of a real person. That is exactly the aesthetic much jewellery marketing chases, which is why the category is exposed.
What Jewellery Brands Can Do About It — 5 Practical Routes
Below are five ways to handle AI content under the new rules. They range from full disclosure to removing the risk entirely. Most brands will mix them.
1. Label AI-Generated Imagery Clearly
Add a visible, plain-language label to synthetic images and video that could be read as real photographs of real people. The disclosure has to be perceivable without special tools and present at first exposure. A caption, an on-image tag, or a consistent icon in your product galleries all work. Keep it consistent across your site and social channels so it reads as a policy, not an accident.
To make this easier, the Commission has published a free set of EU icons for labelling AI-generated content. They are part of the official Code of Practice, come in four colour variations (black, white, and both at 50% transparency), and can be downloaded and used without attribution. User-testing found the icons performed better across every measure when paired with a short text label such as "modified", so use the icon and a word, not the icon alone. There are three to choose from:
Icon | When to use it | Jewellery example |
Basic AI icon | AI was involved in creating the image, video or text | A campaign image where AI touched part of the shot, tagged "created with AI" beside the basic icon |
Fully AI-Generated | The whole image, video or text is AI-made, with no human-created elements beyond prompting | A model shot where the person, hand and setting are entirely synthetic |
Partially AI-Modified | Real, human-made content was partly altered with AI | A genuine photograph of a real hand with a ring swapped or a background furnished by AI |
The display rules are practical. The icon should be clearly perceivable at first exposure, placed where nothing overlays it, embedded in the content, and still visible when the image is reshared or downloaded. For accessibility, add alt text or an ARIA label stating the content is AI-generated, and keep any wording in plain language. One caveat worth noting: using the icons is optional and does not by itself prove compliance — the Article 50 duty is what binds you, and you remain responsible for meeting it.
Operational cost: low. A downloaded icon set, a template, and a workflow rule. The pain is remembering to apply it on every asset, every time.
2. Keep a Content Inventory and Editorial Sign-Off
The guidance repeatedly points to documented workflows. AI-assisted text escapes labelling only where it has undergone human review or editorial control and a named person holds editorial responsibility. Practically, that means a simple register: which assets used AI, who reviewed them, who signed off. This is the record a market surveillance authority would ask to see.
Operational cost: low to medium. A spreadsheet and a habit. The value is that it turns "we think we're fine" into evidence.
3. Use the "AI-Assisted" Route Instead of "Fully AI-Generated"
Classification carries a hidden commercial consequence. If you label a work as fully AI-generated, third parties may read that as limited human authorship, which under European copyright concepts can affect whether copyright protection exists at all — potentially leaving the content free for competitors to reuse. Genuine human creative direction, documented, supports an "AI-assisted" classification and protects the asset.
Operational cost: medium. Requires real human involvement in the creative, not a label swap. The upside is you keep ownership of what you make.
4. Favour Fantastical or Evidently Creative Concepts Where It Fits
For campaign and editorial work — as opposed to catalogue product shots — leaning into surreal, stylised, plainly artificial imagery moves you toward the lighter-touch carve-out. This is a creative-direction decision, not a legal loophole, and it only fits certain brand positions.
Operational cost: low, but limited use. It suits mood and campaign work, not the honest packshot a customer needs to judge a clasp or a stone.
5. Work With a Specialist Who Handles Both Real and AI Photography — Chocianaite
For the images where a customer is deciding whether to spend money — the product shot, the model shot showing scale and drape on a real hand — real photography sidesteps the deepfake disclosure question entirely. There is nothing to label because nothing is synthetic. It also protects the thing weak AI images quietly erode: buyer trust in what the piece actually looks like.
This is the territory Chocianaite works in. As a creative house working exclusively with jewellery brands, the studio produces real product, model, and campaign photography built for conversion — and also produces AI jewellery photography for brands that want it, handled transparently and labelled in line with the rules above.
The point is matching the method to the job: real photography for the images that carry the purchase decision, where accurate colour, true light on metal, and real scale on a hand can't be faked; AI imagery where it genuinely helps and the disclosure is managed properly.
Working with a specialist means both are done correctly, so the compliance question is handled rather than discovered later. For brands selling into the EU, that removes a category of risk from the most commercially important images they publish. You can see the approach and discuss a project on the Chocianaite contact page.
Operational cost: varies by method. Real photography costs more per asset but carries no disclosure question; AI costs less but must be labelled. A specialist helps you spend each where it pays off. Contact us for a quote; start with a FREE introduction call.
Quick Comparison: Your Options at a Glance
Route | Disclosure needed? | Cost to run | Best used for |
Label AI imagery | Yes, visible label | Low | Synthetic content that reads as real |
Inventory + sign-off | Supports compliance | Low–medium | Every AI-touched asset |
"AI-assisted" classification | Yes, but lighter | Medium | Protecting copyright on AI work |
Fantastical concepts | Light-touch only | Low, limited fit | Campaign and mood imagery |
Real photography (Chocianaite) | None — nothing synthetic | Higher per asset | Product and model shots that drive the sale |
Why Act Before This Becomes a Dispute
The enforcement timeline is no longer abstract. GPAI obligations have been live since August 2025, prohibited practices since February 2025, and from 2 August 2026 the Commission's supervision and enforcement powers apply, with high-risk system obligations arriving on the same date. National market surveillance authorities are operational. The AI Act and GDPR apply concurrently, so any AI content that also processes personal data sits under two regimes at once.
For a jewellery brand, the realistic risk is not a headline fine on day one — those powers point at the AI giants. The realistic risk is a complaint, a customer challenge, or a marketplace requirement that asks you to prove your AI imagery is disclosed and documented. Brands that inventory their content and label honestly now will answer that in an afternoon. Brands that don't will scramble. The cheapest moment to build the habit is before anyone asks.
The EU AI Act's transparency rules are now enforceable, and they reach any brand whose AI content is seen by EU customers, wherever the brand sits. Photo-realistic AI imagery of synthetic people is the exposed case. Label it, document it, and reserve real photography for the images where authenticity is the whole point. Handled properly, this is a workflow change, not a crisis — and it happens to reward the brands whose product images were honest to begin with.
FAQ
Does the EU AI Act apply to my brand if I'm based in the UK or US? Yes, where your AI-generated output is used in the EU. The rules reach providers and deployers outside the EU when EU customers see the content.
What counts as a "deepfake" under these rules? AI-generated or manipulated image, audio or video that resembles a real person, place or event and would falsely appear authentic. A photo-realistic AI image of a synthetic model can fall inside this definition.
Do I have to label every AI image I use? Not every one. Clearly fantastical content is out of scope, and evidently artistic work gets a lighter duty. Photo-realistic content built to look like a genuine photograph of a real person is the case that triggers disclosure.
When do the rules start? Article 50 transparency obligations apply from August 2026. Providers of generative AI systems already on the market have until 2 December 2026 to meet the technical marking duty, but the deployer disclosure duty applies to brands now.
Do I have to use the official EU icons? No. The EU icons are free and optional, and pairing them with a short text label works best. They help you label consistently, but using them does not prove compliance on its own — the Article 50 disclosure duty is what you have to meet.
What's the safest option for my main product images? Real photography. If nothing in the image is synthetic, there is no AI disclosure question, and the customer sees the piece as it actually is.
This article is general information, not legal advice. For your specific obligations, consult a qualified adviser.
Comments