AI video in fashion ads: big savings, bigger backlash, new labelling rules

In December 2025 Valentino posted an AI-generated video on Instagram for its Valentino Garavani DeVain handbag: models emerging from a gold version of the bag, the logo morphing into arms and swirling bodies. The film was labelled as AI-made and framed as a "digital creative project" with digital artists. It still drew hundreds of critical comments, with users calling the work cheap, lazy and disturbing. "Disappointing from a couture fashion house," one wrote, the BBC reported.

That reaction captures the central tension of generative video in fashion marketing. The savings are real where AI produces volume: Zalando says AI cut the production time of campaign imagery from six to eight weeks to three or four days and reduced costs by 90%. But in the hero campaign, the piece of marketing that is supposed to prove a luxury brand's craft, AI reads to many consumers as a cost cut. As Getty Images' Rebecca Swift put it to Glossy, "Consumers predominantly view AI-created works as less valuable than human-made images."

The stakes changed on 2 August 2026, when the transparency obligations in Article 50 of the EU AI Act became enforceable. Brands that publish AI-generated or manipulated video depicting realistic people now face labelling duties, while unions and new state laws are putting a price on the digital replicas of models and actors. The question for fashion executives is no longer whether AI video can be made cheaply, but where it is worth the reputational and legal cost.

Which fashion brands have used AI video in campaigns?

The list is longer than the controversies suggest. Glossy counted a series of luxury experiments in 2025: Jil Sander used AI-generated visuals in a conceptual campaign in February, MCM Worldwide used AI-generated surreal environments to extend its spring campaign, and Burberry built AI-enhanced movement from imagery of a 1980s campaign while keeping its core product imagery as conventional photography. Peek & Cloppenburg has tested AI imagery on its sites, often without announcing it.

Technology companies are using fashion to showcase their video models. For Google, the agency R/GA made From the Mountains to the City, a Moncler film that premiered at Google Marketing Live and was built from the outputs of Google's Veo 2 video model, orchestrated through a Gemini-powered tool the agency called ShotFlow. R/GA says the film took four weeks to make. In May 2026 Prada unveiled Satellites II, a six-minute AI-assisted short by game designer Hideo Kojima and film-maker Nicolas Winding Refn for the 14th edition of Prada Mode in New York. Online critics quickly pointed out that Kojima's own games warn about the misuse of AI.

Why did Valentino's AI video provoke such a backlash?

The criticism was less about the technology than about what it signalled. "The main issue is not the technology itself - it is the perception of what the technology replaces," Anne-Liese Prem, head of cultural insights at the agency Loop, told the BBC. "When AI enters the visual identity of a brand, people worry that the brand is choosing efficiency over artistry." Swift added that consumers "hold brands to a higher standard, especially expensive brands", and that "even full transparency about AI use wasn't enough to win them over".

Valentino was not the first case. In August 2025 a Guess advertisement in the August issue of American Vogue featured two AI-generated models, Vivienne and Anastasia, created by the London agency Seraphinne Vallora. Critics on TikTok accused both Guess and Vogue of promoting unrealistic beauty standards. "This doesn't make me want to buy anything," one user said. The agency's co-founder Andreea Petrescu defended the work as "meant to just supplement and to add, you know, a new avenue of marketing".

Seraphinne Vallora's other co-founder, Valentina Gonzalez, insisted that "the heart of fashion is photography. We will never challenge that." The psychologist Rachel Hawkins offered the counterpoint that worries brand teams: with synthetic models, "we start to basically idealize these images that don't actually exist". For a sector already criticised over body image, AI-generated people add a second layer of scrutiny on top of the question of who lost the job.

Gucci found in February 2026 that labelling does not neutralise the issue. Two days before Demna's first Milan show for the house, it posted four AI-generated images on X, each captioned "Created with AI". Critics argued that a brand whose handbags cost from USD 850 to USD 10,000 has little credibility cutting costs on imagery, and Gucci did not respond to requests for comment, DesignRush reported. The common thread is that luxury pricing is built on a story of human craft, and generative imagery in the most visible slot of the marketing calendar appears to contradict it.

How much money does AI video actually save?

The strongest evidence comes from volume production, not hero campaigns. In May 2025 Zalando told Reuters that about 70% of its editorial campaign images in the fourth quarter of 2024 were AI-generated, that production time had fallen from six to eight weeks to three or four days, and that costs were down 90%. "I see it rather that creative minds have now, instead of two hands, six hands," said Matthias Haase, Zalando's vice president of content solutions.

These are company figures and refer mainly to still imagery rather than film. Independent producers give a more cautious picture. Jill Asemota, founder of the Berlin studio Parallel Pictures, told Glossy that AI e-commerce imagery can reduce production costs by up to 70% compared with studio shoots, according to the studio's internal data, while campaign-related AI work saves closer to 50%. "For campaigns, it's almost always an add-on," she said: clients typically ask for 10 or 20 additional assets matching a real shoot. Video adds a technical limit. "When models start moving, the product can change or warp," Asemota said.

That last point matters more for fashion than for most categories. A handbag whose stitching shifts between frames, or a dress whose drape is invented by a model, is not just an aesthetic flaw: it risks showing customers a product that does not exist. The savings case for generative video is therefore strongest for mood, backgrounds and extensions of existing shoots, and weakest where the product must be shown exactly.

Asemota also described a less rational driver of adoption. "There's a certain FOMO," she told Glossy, adding that "AI helps with that scale, but it doesn't replace the main campaign." The Moncler film shows what the technology can do when a platform owner funds the experiment: a four-week production, a purpose-built orchestration tool and a launch on Google's own stage. Few fashion brands will have that support, and none of the parties has published what the film cost.

Do consumers accept AI-generated fashion ads?

Marketers think they do; consumers say they are less sure. A study by IAB and Sonata Insights, published in January 2026, surveyed more than 500 Gen Z and Millennial consumers and 100 advertising executives.

While 82% of the executives believed young consumers felt very or somewhat positive about AI-generated ads, only 45% of the consumers said so, and the gap had widened by five points in two years. 39% of Gen Z respondents reported negative feelings about AI. Disclosure helps rather than hurts on this measure: 73% said clear disclosure would increase or not affect their likelihood to buy. "There is a growing gap between advertisers' use of AI and consumer skepticism," said Jack Koch, IAB's senior vice president of research and insights.

What does the EU AI Act require from 2 August 2026?

Article 50 of the AI Act, enforceable since 2 August 2026, requires deployers of AI systems that generate or manipulate realistic images, audio or video, known as deepfakes, to disclose that the content is artificial. According to the Commission's final guidelines on transparency obligations, published on 20 July 2026 and summarised by the law firm Faegre Drinker, deployers must apply a clear visual label at the point of first exposure, and the Commission's standardised icon satisfies that duty. Advertising may be treated as creative content with a lighter disclosure regime only in narrow circumstances, and most advertising examples do not qualify.

Two details matter for fashion. Content whose audience does not expect it to be authentic falls outside the deepfake definition, which may cover overtly surreal work. But digital replicas of real persons still require disclosure, which captures any campaign that uses an AI twin of a named model. Under the adopted AI Omnibus, systems already on the market before 2 August 2026 have until 2 December 2026 to implement the machine-readable marking requirement in Article 50(2), and content generated before 2 August does not need retroactive marking. Fines for non-compliance can reach EUR 15 million or 3% of worldwide annual turnover.

What are models' unions and agencies demanding?

Consent and payment, in writing. When H&M said in spring 2025 that it would create 30 digital twins of its models, it stressed that the models would own the rights to their twins and be paid each time they were used. That did not end the debate. Paul W. Fleming, general secretary of the UK performers' union Equity, warned that the "race to 'innovate' in the area of artificial intelligence must also not be a race to the bottom to increase profits". Sara Ziff, founder of the Model Alliance, voiced "serious concerns about the use of digital replicas without meaningful protections in place".

In New York those protections now have legal force. The Fashion Workers Act, in effect since 19 June 2025, requires model management companies and clients to obtain clear, separate written consent before creating or using a model's digital replica, setting out its scope, purpose, rate of pay and duration. The labour commissioner can fine a model management company up to USD 3,000 for a first violation and USD 5,000 for subsequent ones, and models can claim damages. Model management companies had to register with the state by 19 June 2026, which brings the agencies that broker replica deals under direct supervision.

Actors' unions have set a price. SAG-AFTRA's 2025 commercials contracts, approved by its national board in April 2025, require consent before a digital replica is created and before it is used, and make generating a performance with a digital replica trigger a 1.5 times session fee plus holding and use fees. The union says the deal includes provisions to "remove economic advantages to the use of synthetic performers". The contracts also bar the use of performers' data to train AI without consent and are valued at USD 218.4 million in new earnings and benefit contributions over three years. The union's chief negotiator, Duncan Crabtree-Ireland, said the agreement "puts rules and a price tag on digital replica use".

What are the risks and limits of AI video for fashion brands?

The most cited savings are company claims about mostly still imagery, and nobody has published audited figures on the full cost of generative video campaigns, including prompting time, retouching, legal review and the reshoots needed when products warp. Brands should treat vendor case studies with the same scepticism they would apply to any marketing claim.

The reputational risk is asymmetric. A successful AI campaign rarely makes headlines for being AI; a failed one becomes the story, as Valentino, Guess and Gucci found. Labelling, now a legal duty in the EU, is necessary but not sufficient: Swift's research suggests that even full transparency does not win sceptical audiences over. There is a legal risk beyond the AI Act too. If generative video alters a product's appearance, the ad may misrepresent what the customer will receive.

The counter-argument is that the backlash is concentrated in visible, high-price moments and among engaged online audiences, while volume content quietly shifts to AI with little resistance. Zalando's 70% share of AI-generated editorial images shows how far that shift can go, and the IAB data show disclosure does not hurt purchase intent for most young consumers. The likely outcome is not a retreat from AI video but a split: human-made hero work, AI-assisted everything else.

What should fashion brands do now?

1. Map every AI video use across campaigns, social and e-commerce, and record which tools, which people and which products appear in each piece.
2. Apply the EU label consistently from first exposure for any realistic synthetic person or scene, using the Commission's icon where it fits, rather than case by case.
3. Secure written replica consent that specifies scope, purpose, pay and duration for every model or actor, in line with the New York Fashion Workers Act and union agreements.
4. Keep hero campaigns human-led where the brand sells craft, and use generative video for extensions, backgrounds and additional assets.
5. Check product fidelity frame by frame so that colour, cut, logo and hardware match the real item before any AI video is published.
6. Prepare a response before launch that explains why AI was used and who made the work, because silence, as Gucci found, becomes the story.