Does CGI need an AI label under the EU AI Act?

In most cases, no. As of October 2026, our reading of the EU AI Act and the European Commission’s guidelines is that a product image rendered from a 3D model built by artists is not AI-generated content, so it does not need an AI label. A label can become necessary when a generative AI tool creates or substantially changes realistic parts of the image, and above all when it creates or changes the product itself. This article explains where that line sits. It is not legal advice; for a specific campaign, check with your own counsel.

What does the AI Act require since August 2026?

Two transparency duties in Article 50 matter for brand imagery, and they apply to different people.

Article 50 has applied since 2 August 2026, the general application date set in Article 113 of the regulation. The relevant parts are:

Rule	Who it applies to	What it requires	What it means for a CGI buyer
Article 50(2)	Providers of generative AI systems (the companies that build the tools)	Mark AI output in a machine-readable way so it can be detected	Mostly the tool maker’s job, not yours or your studio’s
Article 50(4)	Deployers who use an AI system to generate or manipulate a deep fake	Disclose that the content is artificially generated or manipulated	The rule that can require a visible label on a campaign image or film

The key term is “deep fake”. Article 3(60) defines it as AI-generated or manipulated image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. Note the word “objects”: a product is an object, so this is not only about fake celebrities.

Is a rendered CGI image “AI-generated”?

No, not by itself. The Commission’s guidelines on Article 50 explicitly name “a rendered frame” as an example of content that falls outside the marking rule, because it comes from simple data processing rather than AI generation.

That matches how CGI is made. Artists model the bottle, set up materials and lights, and a render engine calculates how light behaves in that scene. Nothing is invented by a model trained on other people’s images. The guidelines also say that intermediate 3D artifacts, such as a mesh, do not need to be marked.

Studios that build the product from its real files can usually show element by element what is and is not AI. Not Important Studio, for example, has written that it uses AI mainly for research and processing, not to create the work itself.

What about AI tools inside a CGI pipeline?

Most AI tools in a normal CGI workflow fall under the exceptions for standard editing, but generative tools that add or change visible content need a closer look.

Article 50(2) does not apply where an AI system performs an assistive function for standard editing or does not substantially alter the input. The guidelines give examples, and several map directly onto CGI work.

Likely outside the rules, according to the guidelines’ examples:

• AI noise reduction, which many render engines use to clean up a render
• Rescaling, format conversion and compression
• Pixel filling to adapt an image to another aspect ratio
• Minor color correction, sharpening and removing dust spots

Worth a closer look:

• Generated backgrounds, props or sets. The guidelines treat a real product shown against an AI background as not a deep fake, as long as the ad does not mislead about the product.
• Generated or AI-altered people, hands, faces or skin.
• Any AI step that creates or changes the product, its liquid, its texture or what it appears to do.

The guidelines also say that outputs used only inside closed production workflows for film, animation, games or advertising do not need marking until they become a final output. In plain terms: what counts is what ends up in front of the audience.

When does a beauty image become a deep fake?

When AI has generated or changed something realistic in a way that could mislead viewers about what is real, and in advertising that most often means the product.

The guidelines break the definition into four tests: the content resembles something, that something exists or could plausibly exist, it is a person, object, place, entity or event, and it would falsely appear authentic or truthful. Clearly unrealistic content, such as a dragon or a flying sphinx, does not count. Photorealism makes a deep fake more likely but does not decide it alone.

For beauty brands, the clearest example in the guidelines is this: an AI-generated image of a product in an advertisement or on packaging that could mislead about the product’s real appearance, characteristics or use, for example by making it look better than it is in real life. That is listed as a deep fake. A real product placed in an AI-generated setting is listed as not a deep fake, provided the ad does not mislead about the product.

So the risk is highest when AI touches the parts of the image that carry the claim: the texture of a cream, the effect of a serum on skin, the color of a lipstick. Those are also the details that advertising regulators already watch. The UK’s ASA, for example, warns against production techniques that exaggerate what a cosmetic can do, whatever the tool.

Does the creative-work exception cover advertising?

Rarely, and it would not remove the label anyway. Article 50(4) only softens the duty for evidently artistic, creative, satirical or fictional work: you still disclose, but in a way that does not spoil the work.

The guidelines interpret “evidently” strictly. Content that is purely informative or commercial and recognizable as such is excluded. Advertising might qualify in specific situations but not in others. The examples that do not qualify include a teleshopping-style video with synthetic people demonstrating a product, and a synthetic influencer testing a real product. If your campaign uses AI to show a product working, do not plan around this exception.

Who is responsible, the brand or the studio?

The deployer is whoever decides to use the AI system and how. According to the guidelines, a brand that simply commissions an agency, without deciding whether and how the agency uses AI, is not the deployer. The agency or studio is.

That changes if the brand directs the AI use, for example by asking for an AI-generated model or an AI-extended set. The guidelines also say that employees and freelancers working under a company’s control are not separate deployers; the company is. Because the answer depends on who decides, it is worth writing down in the contract who chooses AI tools and who handles labeling.

What should you ask your studio now?

A short written answer per deliverable covers most of the risk:

1. Which AI tools were used on this image or film, and for what?
2. Did any tool generate new visible content, or substantially change existing content?
3. Was the product itself, its liquid, texture or effect, generated or changed by AI?
4. Were any people, faces, hands or skin generated or altered by AI?
5. Did any tool embed provenance metadata, and has it been kept intact in delivery?
6. If a label is needed, what wording and placement do we use?

The Commission published a voluntary Code of Practice on marking and labeling on 10 June 2026, and offers optional EU icons. Both help if you end up needing a label. If you are still choosing between generated and built imagery, our guide on CGI or photography for a beauty campaign and our brief checklist cover how to set the project up.

What is still uncertain?

Several things, and they are worth tracking:

• Guideline status. The Commission approved the content of its Article 50 guidelines on 20 July 2026 and said formal adoption would follow once all language versions are ready. The guidelines are non-binding and call themselves a first interpretation that will be reviewed. Only the Court of Justice of the EU can give the final, authoritative interpretation.
• The Digital Omnibus. Regulation (EU) 2026/1744 amended the AI Act in 2026. According to the Commission’s guidelines, it gives providers of generative AI systems already on the market before 2 August 2026 until 2 December 2026 to meet the machine-readable marking rule. As far as we could find, it did not change the deep fake disclosure duty for deployers.
• Enforcement. National market surveillance authorities enforce Article 50, and how strictly they treat advertising is not yet visible.

For a specific campaign, especially one where AI touches the product or people, ask your legal team before launch.