By Stuart Kerr, Technology Correspondent, LiveAIWire

A TikTok video calling out a Guess advert in Vogue’s August 2025 issue was viewed more than 2.7 million times before most readers realised what they were looking at: two models, Vivienne and Anastasia, who do not exist. Both were generated by the AI marketing studio Seraphinne Vallora for Guess, with a disclosure printed in type small enough that most readers scrolled past it entirely. Calls for a boycott followed within days.

Guess is not an outlier, and the studio behind the ad did use a real person as a reference, photographing a human model in the actual clothes before generating the synthetic versions. It is the most visible example of a shift that has been building quietly across the industry for two years, and it has turned AI in fashion modelling from a niche experiment into a live fight over who gets paid, who gets credited, and whether the person in a photograph was ever in the room, or exists at all.

How AI in Fashion Modelling Actually Works

H&M took a different approach in July 2025, building what it calls digital twins of 30 real, named models, generated from thousands of reference photographs and used to shoot campaign imagery set in cities the models never visited for that shoot. Unlike Guess, H&M’s models retain ownership of their digital likeness and are paid each time it is used, under terms the company says mirror a standard modelling contract. One model involved, Mathilda Gvarliani, described her digital twin simply: “She’s like me, without the jet lag.”

The pattern goes back further than either headline. Levi’s announced a partnership with the Amsterdam startup Lalaland.ai in March 2023 to generate AI models in a wider range of body types and skin tones than its existing photoshoots reflected. Within a week, after being accused of using AI to simulate diversity rather than pay a more diverse cast of real models, Levi’s issued a clarification stating the AI models would supplement rather than replace human photoshoots, a walkback that has become a template other brands now study before launching their own AI campaigns.

Mango ran what is widely considered the first fully AI-generated campaign from a major retailer, for its 2024 Sunset Dream teen collection, with no photoshoot at all behind the finished images. By 2026, brands including Zara, Burberry, Adidas and Patagonia were using AI-generated or AI-assisted model imagery in at least some part of their marketing, most often for the high-volume e-commerce product shots that never appear in a print campaign but make up the bulk of what a shopper actually sees on a retailer’s website.

The AI studio behind these campaigns is also consolidating. Lalaland.ai, the Amsterdam firm at the centre of the Levi’s controversy, had by 2025 built a client roster that also included Tommy Hilfiger, Calvin Klein and Patagonia before being acquired outright by the 3D fashion software company Browzwear in July 2025. An acquisition rather than a shutdown is usually a sign that a category has stopped being a novelty and started being infrastructure other companies want to own.

What This Means for You

If you shop online, there is now a reasonable chance that the person modelling the item in your basket is not a real person, and current disclosure rules do not reliably tell you which is which. Guess’s small-print label technically complied with existing advertising standards while functionally failing most readers. That gap is closing on paper at least: from 2 August 2026, the EU AI Act’s Article 50 requires a visible, not merely technical, disclosure label on AI-generated images used commercially, with fines that can reach 15 million euros or 3 percent of global turnover for non-compliance.

Until enforcement catches up with intent, the only defence a shopper has is to assume any campaign image could be synthetic and to rely on customer photos and reviews for an honest sense of fit and drape. For anyone whose income depends on being photographed for a living, the practical question is starker: which parts of this work are being replaced, and which are being preserved with consent and payment. H&M’s model is the version the industry’s advocates want to become standard. Guess’s is the version they are trying to prevent.

The Diversity Paradox at the Centre of This Fight

Every brand that has introduced AI models has framed the technology as a tool for inclusion, generating body types, skin tones and ages that a single photoshoot could never economically cast in full. Critics call this “artificial diversity,” the substitution of a synthetic image of a marginalised body for the paid, credited work of an actual person from that background. The debate is not new. Shudu, created in 2017 and often described as the world’s first digital supermodel, booked campaigns with brands including BMW and Louis Vuitton while drawing sustained criticism for being a Black woman created by a white founder, Cameron-James Wilson, years before H&M or Guess made the same trade-off at industrial scale.

There is a second, quieter version of this problem. AI image generators are trained on decades of existing fashion photography, an archive that already skews toward narrow beauty standards. A model built from that archive does not correct its biases by default. It reproduces them fluently, at scale, unless a company deliberately intervenes to broaden what the system generates, which is a design choice, not an automatic feature of the technology.

The pattern across Levi’s in 2023, H&M in 2025 and Guess later that same year is strikingly consistent: a brand announces an AI model initiative framed around representation or efficiency, a portion of its audience identifies the underlying substitution of real paid work for synthetic imagery, and the brand issues a clarification within days. That three-times-repeated cycle suggests the industry has not yet internalised the lesson, only learned to apologise for it faster each time.

Who Loses When the Model Is Synthetic

The clearest financial threat is not to the small number of models who walk in Paris or Milan. It is to the much larger group who earn their living from e-commerce catalogue work, the repetitive, unglamorous shoots that pay a working model’s rent between bigger jobs and that also happen to be the easiest category for an AI system to replicate convincingly. NBC News reporting on the Levi’s backlash found that critics worried professional models could be pushed out of jobs entirely as AI-generated images became harder for ordinary consumers to distinguish from real photography.

The wider crew a physical shoot requires, photographers, stylists, hairdressers and make-up artists, are not party to any digital twin licensing agreement and see none of the payment structure that H&M built for its models. Estimates of the size of the AI-in-fashion market vary widely between research firms, but Precedence Research put it at 2.23 billion dollars in 2024, projected to grow past 60 billion dollars by 2034. Whatever the precise figure, every credible estimate points the same direction, and that direction runs directly through the entry-level modelling work that has traditionally been how new, often diverse, talent built a career in the first place.

The Law Is Already Catching Up, at Least on Consent

Fashion models have more legal protection against unauthorised digital replicas than most other creative professions, at least in two states. New York’s Fashion Workers Act, in force since 19 June 2025, requires brands and agencies to obtain separate written consent before creating or using a model’s digital replica, spelling out the scope, duration and rate of pay for that specific use, with civil penalties of up to 5,000 dollars per violation. California’s AB 2602, effective from 1 January 2025, imposes a similar affirmative-consent requirement for digital replicas of a person’s voice or likeness. Neither law existed when Levi’s launched its Lalaland partnership in 2023.

The gap those laws are closing is exactly the one Guess’s small-print disclosure exploited: a contract that permits “routine photographic edits” does not, under New York’s new statute, automatically permit generating an entirely new AI image of a model. At least one model has already sued a retailer on precisely that distinction, arguing that a standard model release never covered the creation of an AI clone. How that case and others like it resolve will determine whether H&M’s consent-based approach becomes the industry norm or remains the exception brands point to when defending less careful competitors.

Is This Actually the End of Runways?

Not literally, and not yet. Physical runway shows remain the industry’s most valuable marketing spectacle precisely because they are live, and AI has not replaced that theatre in any commercially meaningful way. What is changing is the pipeline that feeds runway careers: the catalogue and e-commerce bookings that let a new model build a portfolio, earn a reputation with casting directors, and eventually get considered for a show. If that entry-level tier of paid work keeps shrinking, the runway does not disappear, but who gets to walk it, and how they got there, changes considerably.

Agencies themselves are starting to hedge both directions at once, signing new faces while also experimenting with licensing existing models’ likenesses for digital use, effectively becoming brokers for a hybrid workforce of real people and their AI-extended selves. That dual approach may turn out to be the actual long-term shape of the industry, not a full replacement of models by AI, and not a return to an all-human pipeline either, but a negotiated middle where a smaller number of established faces license their likeness widely while a shrinking number of newcomers fight over the paid bookings that remain.

Fashion has already run this experiment once, in the debate over Sora 2 and Hollywood likeness rights, where actor Bryan Cranston’s push for consent controls became a template other creative professions are now copying. H&M’s consent-and-payment model for digital twins looks like fashion’s version of that same fight, arriving early rather than after the damage is done. Guess’s small-print approach looks like the version everyone else is now under regulatory pressure to avoid repeating.

For related coverage of how generative tools are reshaping the wider industry, see LiveAIWire’s analysis of AI in fashion design and supply chains, our reporting on how AI tools trained on historical data reproduce bias, and our look at why independent and gig-style creative workers are simultaneously the most exposed and the most adaptable to this kind of disruption.

None of this resolves in a single season. Brands will keep testing digital twins because the cost savings are real and immediate, models and their advocates will keep pushing for consent and payment terms that match the value being extracted from their likeness, and regulators in New York, California and Brussels will keep tightening the rules a step behind whatever the next AI fashion campaign turns out to be. The runway itself will likely survive all of it largely unchanged. The much larger, much less photographed career that used to lead there is the part actually being rewritten.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.