Do You Have to Disclose AI Fashion Images? The 2026 Rules for Brands

For a couple of years, using AI-generated models and imagery in fashion advertising sat in a grey area. You could run an AI campaign, put an AI "model" on a product page, and nobody was legally required to say so. That grey area closed in 2026. Two major disclosure laws took effect within weeks of each other, and both reach fashion brands directly.

Here's the plain-English version of what changed, who it applies to, and what you actually have to do. One note before the details: this is a practical summary, not legal advice. The specifics of your situation deserve a real conversation with a lawyer, and the rules are new enough that guidance is still settling.

What changed in 2026

Three pieces of law matter for fashion brands using AI imagery.

New York's Synthetic Performer Disclosure Law (effective June 9, 2026). If an advertisement uses a "synthetic performer" — an AI-generated figure standing in for a human performer — the ad has to carry a clear and conspicuous disclosure inside the ad itself. It applies to any company whose ads reach New York consumers, wherever that company is based, so a brand in Los Angeles or London running ads that a New Yorker sees is in scope. Penalties run $1,000 for a first violation and $5,000 for each one after. There are carve-outs, including audio-only ads and promotional material for films and video games, but on-model fashion advertising is squarely the kind of thing it covers.

The EU AI Act transparency rules (effective August 2, 2026). AI-generated or manipulated images, audio, and video have to be labeled clearly and at first exposure, not buried in metadata or a terms page. The reach here is important: it applies even when there's no intent to deceive, and content that resembles a real person needs labeling regardless of whether an actual individual is depicted. If you sell into the EU, this touches your AI imagery.

New York's Fashion Workers Act (model consent for digital replicas). This one is about the person, not the label. Models have to give consent for digital replicas of themselves. It grew out of a model suing a retailer over AI images generated from a likeness after the contract had expired. If your AI imagery is built on a real model's face or body, consent is now a documented requirement, not a handshake.

Who this actually affects

Short answer: most brands, sooner than they think. The New York and EU rules are written around reach, not headquarters. If your ads or product pages are seen by consumers in New York or the EU, the location of your studio doesn't get you out of scope. For an online fashion brand, that's effectively everyone.

The trigger is also broader than "obvious deepfake." The EU rule applies without any intent to deceive, and the New York rule is about using a synthetic performer at all, not about whether the image is misleading. A tasteful, clearly-commercial AI model in a clean product shot still counts.

What you actually have to do

The common thread across the laws is the same instinct: say so, clearly, where people will see it.

Disclosure has to be clear and conspicuous, and under the EU rule, at first exposure. That rules out the tricks brands reach for first — a line in the footer, a note in the image metadata, a mention buried in terms and conditions. The label belongs where the viewer meets the image, not somewhere they'd have to go looking.

If your imagery uses a real person's likeness, get and keep consent for the digital replica, and mind the term of that consent. The Fashion Workers Act case turned on an expired contract, so "we had permission once" is not the same as "we have permission now."

None of this bans AI imagery. It requires you to be transparent about it and clean about whose likeness you used. Those are very doable. They just have to be done on purpose.

The part that quietly favors doing it right

Here's where it gets interesting for brands that care about the craft. These rules are easiest to comply with when your AI imagery is built the careful way, and hardest when it's the cheap, scraped-together way.

Consent is a non-issue if your talent is original. When the model is a custom, fully synthetic figure created for your brand rather than a real person's scanned likeness, there's no expired contract to worry about and no digital-replica consent to chase. The whole category of Fashion Workers Act risk simply doesn't apply. That's a structural advantage of using original, brand-owned talent over face-swapping a real model.

Disclosure done well can actually build trust rather than cost it. The brands that will struggle are the ones hoping nobody notices the AI. The ones that will be fine are the ones whose AI work is good enough that a label doesn't embarrass them. If your AI images are accurate, honest about fit, and free of the tells that make people distrust AI, "AI-assisted imagery" reads as modern, not as a warning. If your images are the confidently-wrong kind, a mandatory label just points a spotlight at the problem.

That's the real filter these laws create. They punish the sloppy and the deceptive, and they barely inconvenience the transparent and the accurate.

A short checklist for right now

• Assume New York and EU rules apply to you if your ads or product pages reach those consumers, which for most online brands means yes.
• Add a clear, visible disclosure where AI-generated imagery appears — not in metadata, not in the footer, not in terms.
• If you use a real person's likeness, confirm you have current, in-term consent for the digital replica. Check the dates.
• Favor original, brand-owned AI talent over real-likeness face-swaps, which sidesteps the consent problem entirely.
• Get the imagery to a standard where a disclosure label doesn't hurt you — accurate construction, honest fit, none of the obvious AI errors.
• Talk to a lawyer about your specific setup. This summary is a starting point, not a ruling.

Where I land on it

I think this is good news for the studios and brands doing AI imagery properly, and a reckoning for the ones cutting corners. Disclosure isn't the threat. The threat was always shipping images that don't hold up, and now the law makes those images wear a label. The work that survives this is the work that was accurate and honest to begin with. That has been the whole point here from the start.

If you want AI imagery that's built to clear that bar — original talent, accurate construction, nothing that a disclosure label would expose — that's what we do.