Short answer. Two regimes now govern AI imagery in fashion advertising, and both are live. EU AI Act Article 50 applies from 2 August 2026: providers must embed machine-readable markings and offer a detection mechanism, deployers must disclose deepfake content, and penalties reach €15 million or 3% of worldwide annual turnover. New York's Synthetic Performer Disclosure Law has applied since 9 June 2026: any ad featuring a synthetic performer needs a clear and conspicuous in-piece disclosure, with penalties of $1,000 for a first violation and $5,000 thereafter. A third law — New York's Fashion Workers Act, in force since 19 June 2025 — governs consent for models' digital replicas. Most brands have a plan for none of the three.
The three rules at a glance
| EU AI Act, Art. 50 | NY Synthetic Performer Disclosure Law | NY Fashion Workers Act | |
|---|---|---|---|
| In force | 2 August 2026 | 9 June 2026 | 19 June 2025 |
| Governs | Marking and disclosure of synthetic content | In-ad disclosure of synthetic performers | Consent for models' digital replicas |
| Who it binds | Providers and deployers | The advertiser producing the ad | Model management companies and clients |
| Trigger | Synthetic content; deepfakes | Actual knowledge a synthetic performer is used | Use of a real model's digital replica |
| Penalty | Up to €15m or 3% of worldwide turnover | $1,000 first, $5,000 subsequent | Civil penalties; agency registration duties |
| Reach | Extraterritorial — outputs used in the EU | Any ad reaching New York consumers | New York |
EU AI Act Article 50: what actually applies to you
The European Commission adopted implementing guidelines on 20 July 2026. Article 50 splits obligations between two roles, and getting your role right is the first task.
Providers develop and place an AI system on the market. If you use a third-party generation platform, that platform is the provider — not you.
Deployers use an AI system under their own authority. A fashion brand generating campaign imagery is a deployer. So, note carefully, is your agency or production partner if they operate the tool under their own authority. Establish in writing who is the deployer for each workflow, especially where agencies and contractors are involved.
Article 50 covers four scenarios. Two are relevant to imagery:
- Systems generating or manipulating synthetic image, audio, video or text. Providers must embed machine-readable markings and provide a detection mechanism, subject to limited exceptions such as standard editing and non-substantial alterations.
- Deepfakes. Deployers must disclose that the content was artificially generated or manipulated — unless it has undergone substantive human editorial review with a person assuming editorial responsibility.
Key dates. Obligations apply immediately from 2 August 2026 to all in-scope systems, regardless of when they were placed on the market. Content generated and published before that date does not need retroactive labelling. A limited transitional period runs to 2 December 2026, applying only to the marking-and-detection obligation for generative AI systems already on the market.
How disclosure must be delivered. Per reporting on the EU rules, disclosure must reach the viewer clearly and at first exposure. It cannot be buried in terms and conditions, and it cannot be left to machine-readable metadata alone. The EU provisions apply even without intent to deceive — including, on this reading, content that looks like a real person even where no real individual is depicted and no deception was intended.
The Code of Practice. The AI Office has published a voluntary Code of Practice on Transparency of AI-generated Content, including a set of icons for labelling. Signatories get a degree of presumption of conformity and a more favourable enforcement posture; non-signatories face closer scrutiny and must demonstrate compliance by equivalently adequate alternative means. Ask your generation vendor whether they have signed. It is a one-line procurement question with real consequences.
New York: the disclosure law and the consent law
Synthetic Performer Disclosure Law. Signed by Governor Hochul in December 2025, billed as first-in-the-nation, effective 9 June 2026. It requires a clear and conspicuous disclosure within the piece in any advertisement featuring a synthetic performer.
Specifics worth knowing:
- It targets the advertiser that produces or creates the ad, not the platform.
- It applies where the advertiser has actual knowledge that a synthetic performer is used.
- Penalties: $1,000 for a first violation, $5,000 for each subsequent one.
- It reaches any company whose ads reach New York consumers, regardless of where the advertiser is based.
- Exemptions: audio-only ads; promotional material for expressive works such as films and video games; AI used solely for language translation.
- Publishers that merely disseminate a non-compliant ad are shielded.
The "actual knowledge" standard is worth pausing on. It places the burden on the advertiser to know what is in its own creative — which in practice means knowing what your agency and your production partners did. Not knowing is not a strategy, and it will not survive contact with a discovery process.
Fashion Workers Act. In force since 19 June 2025. Models must give separate, explicit written consent for use of a digital replica, specifying scope, purpose, rate of pay and duration. Consent for one campaign does not carry to another; new use requires new approval. Powers of attorney can no longer cover digital replicas, and pre-existing ones covering them were invalidated. Agencies faced a Department of Labor registration deadline of 19 June 2026, with civil penalties for non-registration.
A New York model sued Rainbow Shops this spring over AI images generated from an expired contract — the practical illustration of why replica lifecycle management is a compliance function, not an archiving preference.
What fashion brands should label, in practice
Work through it by asset category:
| Asset | Real person depicted? | Label needed? |
|---|---|---|
| Ghost mannequin, flat lay | No | Generally not a synthetic performer issue; provider marking still applies |
| AI colourway variant of a real garment | No | Low exposure; standard-editing exceptions may be relevant |
| On-model image, fully synthetic figure | No individual | Yes — NY in-ad disclosure; EU deployer disclosure where applicable |
| On-model image, licensed digital twin | Yes | Yes — plus separate written consent under the Fashion Workers Act |
| Campaign image with AI-generated background only | No | Depends on materiality; standard-editing exceptions may apply |
| AI-generated video with a human figure | Either | Yes — highest exposure category |
Two cautions. First, the "standard editing / non-substantial alteration" exception is real but narrow, and its edges are exactly where enforcement will land — do not build a programme on the assumption that your use falls inside it. Second, this table is an operating starting point, not legal advice; anything running at scale in the EU or New York warrants counsel.
The compliance work, in order
- Map roles. For every imagery workflow, establish in writing whether you or your vendor or your agency is the deployer. Contracts should say so explicitly.
- Inventory assets. Identify what in your existing library is synthetic. Content published before 2 August 2026 does not need retroactive labelling — but you still need to know which side of the line each asset falls on.
- Assess vendors. Does the provider embed machine-readable marking and offer detection? Have they signed the Code of Practice? Note the 2 December 2026 deadline for tools already on the market.
- Build the label into the asset pipeline. Disclosure has to be in the piece, visible at first exposure. Retrofitting labels across thousands of assets is the expensive path.
- Fix consent records. Separate written consent per digital replica covering scope, purpose, rate of pay and duration, with tracked expiry and an active retirement process.
- Brief upstream. The most effective compliance move is changing the brief. Caimera reports clients switching AI use toward flat lay and ghost shots without models, or moving it into design teams, precisely because of the disclosure line.
The commercial argument for disclosing more than required
The compliance floor is not the optimum. Caimera's 2026 survey of 502 US consumers found 75% believe AI imagery should be disclosed, and 79% would trust the brand that labels it when two brands both use AI. And 85% could not reliably tell AI images from real ones — meaning the risk was never being noticed, it was being found out.
The counterweight: a December 2025 Klaviyo/Datalily survey of 8,000 consumers across eight markets found that noticing AI in brand marketing makes people four times more likely to trust the brand less (31%) than more (7%). YouGov puts 55% of consumers as uncomfortable with AI-generated brand marketing on social.
Read together, these say something specific. Consumers penalise visible AI and penalise concealed AI more. The stable position is to use AI where it is uncontroversial, label it, and keep human craft on the imagery that carries brand meaning — which, conveniently, is also the allocation that maximises the savings and minimises the exposure.
FAQ
Do fashion brands have to disclose AI-generated images?
Yes, in two live regimes. Ads featuring synthetic performers reaching New York consumers require clear and conspicuous in-piece disclosure. In the EU, deployers must disclose deepfake content under AI Act Article 50, applicable from 2 August 2026.
When did the EU AI Act transparency rules take effect?
2 August 2026, with implementing guidelines adopted 20 July 2026. A limited transitional period to 2 December 2026 applies only to marking and detection for generative AI systems already on the market.
Do we have to relabel our existing image library?
No. Content generated and published before 2 August 2026 does not need retroactive labelling. You do still need to know which assets are synthetic going forward.
What are the penalties?
Up to €15 million or 3% of worldwide annual turnover under the EU AI Act, whichever is higher. Under New York's Synthetic Performer Disclosure Law, $1,000 for a first violation and $5,000 for each subsequent one.
Does the EU law apply to a US brand?
It can. The AI Act applies to providers, deployers, importers and distributors that place AI on the EU market or whose AI outputs are used within the EU.
Is a metadata tag enough?
No. Disclosure must reach the viewer clearly at first exposure; it cannot be buried in terms and conditions or left to machine-readable metadata alone. Machine-readable marking is a separate, additional provider obligation.
What if our agency generated the images?
Establish in writing who is the deployer. New York's law targets the advertiser producing the ad and applies where the advertiser has actual knowledge a synthetic performer was used — so knowing what your agency did is part of your obligation, not an excuse from it.
Do we need consent to use a model's digital replica?
In New York, yes — separate explicit written consent specifying scope, purpose, rate of pay and duration, per campaign. Expired contracts do not carry forward.
Keep reading
- AI Fashion Photography: The 2026 Guide
- What Is an AI Fashion Model? A Definition
- How AI Fashion Photography Actually Works
- What AI Fashion Photography Actually Costs
Sources
- Cooley LLP, EU AI Act: Transparency Obligations Take Effect 2 August 2026, 3 August 2026 — the four Article 50 scenarios, provider/deployer split, key dates, transitional relief, Code of Practice, penalties, extraterritorial scope.
- European Commission, Guidelines on transparency obligations for providers and deployers of certain AI systems — Article 50 applies from 2 August 2026; enforcement by national market surveillance authorities, the AI Office and the EDPS.
- European Commission, Code of Practice on Transparency of AI-generated Content.
- FashionUnited, How new AI disclosure laws are reshaping fashion advertising, 29 July 2026 — New York Synthetic Performer Disclosure Law provisions, penalties and exemptions; EU delivery requirements; Caimera survey; Rainbow Shops litigation.
- New York State Department of Labor, Fashion Workers Act FAQs — digital replica consent, power-of-attorney restrictions, agency registration.
- EMARKETER, Visible AI in marketing is four times more likely to cost brands trust than build it, 10 April 2026 — Klaviyo/Datalily survey, December 2025; YouGov discomfort figure.
This is an operating summary, not legal advice. Fashion N.U.T. is not a law firm; programmes running at scale in the EU or New York should be reviewed by counsel. Have a correction? Email desk@fashionnut.co.
