AI Model Disclosure Rules: What Sellers Must Label, and Where

Do you have to say a model is AI-generated?

In some places yes; elsewhere a label is the safe choice. New York requires it in ads, Amazon asks for a metadata tag, and the EU now has labelling rules for content that could pass as real. Everywhere, an ad must not mislead buyers about the product.

Our main guide explains what AI fashion models are; our brand cases show how labels have landed with the public.

Read next: AI fashion models explained · Brands using AI models: dated cases

New York: the synthetic performer law

New York requires anyone who makes an ad for a commercial purpose to conspicuously disclose a synthetic performer when they know one is used. The bill, S8420-A, was signed on 11 December 2025 and took effect 180 days later, on 9 June 2026. It sits in the General Business Law as section 396-b.

A synthetic performer is a digitally created asset, made with generative AI or other software, that looks like a human performer but is not recognizable as any real person. Most AI fashion model tools make exactly that. Ads for films, TV shows and games are exempt when the use fits the work, as are audio-only ads and AI used only to translate. The law sets no wording, but the disclosure must be conspicuous, so small print in a corner is a risk.

California and Pennsylvania are next

California and Pennsylvania have passed similar rules. California's Business and Professions Code section 17610, signed on 16 September 2026, applies from 1 January 2027 to ads that prominently include a synthetic performer without a clear and conspicuous disclosure. Prominent means, for example, in the foreground showing the product. Penalties reach $2,500 per violation.

Pennsylvania's Artificial Intelligence Deceptive Advertising Act, signed on 22 October 2026, applies from 21 December 2026. It covers ads where AI-generated or heavily AI-edited content creates a false impression of the product, unless clearly and conspicuously disclosed. Fines reach $3,000 per violation.

Copies of real models need written consent

If an AI image copies a real model, New York's Fashion Workers Act requires their clear written consent first. Since 19 June 2025, that consent must cover the scope, purpose, rate of pay and length of use of the digital replica, and a new campaign needs new consent. A power of attorney can no longer cover digital replicas.

Our tools make generated people, not copies of real ones. Even so, a generated face can resemble someone, so look at every face before you publish and make a new one if it looks like a real person.

Amazon: tag AI-generated people in the file

Amazon asks sellers to tag photoreal AI-generated people in listing images, videos and A+ content, in all its stores, citing recent laws that require disclosure in ads. The tag goes inside the file, not in the listing text. Use an IPTC-compatible metadata editor to add the keyword contains-synthetic-performer to the dc:subject (XMP) field. Amazon says it will add an indicator to the listing where it applies.

No tag is needed for images of only real people, even if AI edited them, or with no photoreal people. Some editors strip metadata on export, so add the tag last, just before upload.

Read next: Amazon main image checker · Marketplace policy explainer

Etsy and Google Merchant Center

Etsy now requires at least one unedited photo of the actual item against a real background in every listing. Its 22 October 2026 update, as reported by Digital Commerce 360, allows AI-assisted editing in other photos if they show the item accurately, and requires sellers to disclose items created with AI. The reporting does not mention AI models, so treat an on-model AI image as an extra photo, never the only one.

Google Merchant Center asks that AI-generated product images keep their IPTC DigitalSourceType metadata set to TrainedAlgorithmicMedia, so do not strip it when you resize or compress.

European Union: AI Act Article 50

From 2 August 2026, the EU AI Act requires people who use AI to make deep fakes to disclose that the content is AI-generated or manipulated. A deep fake here means content that looks like real people, objects, places or events and could falsely seem authentic. Makers of AI tools must also mark output in a machine-readable way; tools already on the market have until 2 December 2026 for that. Fines can reach 15 million euros or 3% of worldwide turnover.

How Article 50 applies to an ordinary product photo with a generated model is still debated. If you sell into the EU, a visible label is the cautious choice.

United States: FTC truth-in-advertising basics

Under FTC rules an ad must be truthful and not deceptive, and any disclosure must be clear and conspicuous. An ad is deceptive if it is likely to mislead a reasonable buyer about something that matters to their decision. An AI model showing a different color, pattern or fit from the real garment can do that, label or not.

The FTC's fake reviews rule, announced in August 2024, bans testimonials from people who do not exist, including AI-generated ones. Never present an AI model as a real customer.

How to label AI models: plain examples

A good label is short, sits where the buyer looks, and says what is AI and what is real. None of these is official wording.

• Listing image caption: "Model is AI-generated. The garment is real; check the size chart for fit."
• Ad overlay, on the image in readable text: "AI-generated model".
• Social post, first line of the caption: "This model was created with AI. The jacket is our real product."
• File metadata: contains-synthetic-performer for Amazon, and TrainedAlgorithmicMedia kept for Google.
• Avoid: tiny grey text in a corner, a label buried in hashtags, or a fake name and review for the model.

Read next: Clothes on a model without hiring one · Best AI fashion model generators · Compliance claims checker