AI-designed fashion is on sale. Whether it sells, almost nobody will say

In November 2023 the online retailer Revolve put clothes designed with Midjourney and Stable Diffusion on sale, developed and manufactured with its own team in Los Angeles. The three winning labels of the first AI Fashion Week each launched with 10 to 12 pieces, priced from $228 for a one-shouldered top to $1,598 for a pleated taffeta dress, according to CNN. Three years later, the question any merchandiser would ask first still has no public answer: did they sell?

That silence is the pattern, not the exception. This publication reviewed the documented cases of collections designed with generative AI and sold commercially, from Revolve and the Spanish brand Desigual to the French menswear chain Jules and the Nigerian label I.N Official. In none of the cases reviewed did the brand or retailer publish sell-through, revenue or margin for the AI-assisted products, and none disclosed what share of the design work the software actually did. The only sales figure found, from a small Korean brand, is unaudited and undated.

The answer to the headline is therefore precise but unflattering. AI-assisted collections are being sold, mostly as limited capsules and marketing experiments. Whether they sell better or worse than human-designed lines is not known publicly, and the legal position on who owns the designs is now settled in one important respect: in the United States, only the human part can be protected.

Which brands have actually sold collections designed with generative AI?

The list is shorter than the volume of press coverage suggests. The best-documented examples have one thing in common: the AI generated images and ideas, and people turned them into garments.

Revolve and Maison Meta. AI Fashion Week took place in New York in April 2023, organised by the creative agency Maison Meta. More than 130 designers showed collections of 15 to 30 virtual looks made with AI software and tools such as Photoshop, and the event's website counted more than 350 submissions at the time, according to Fashion Dive. Revolve co-founder Michael Mente sat on the jury, and the retailer helped build an incubator for the winners. The top three were Paatiff, founded by a Portuguese designer with a background in architecture; Molnm, also Portuguese; and Opé, whose designer had previously worked at Betsey Johnson. Each worked with Revolve's team in Los Angeles on product development and material sourcing, and Revolve manufactured the garments itself. 'Technology is a big competitive edge for Revolve,' Mente said in a press release. CNN reported no sales figures, and none appear in the other sources reviewed.

Desigual. In July 2023 the Barcelona brand launched what it called its first on-demand collection, with garments and accessories devised by its creative team 'with artificial intelligence support', according to FashionUnited. Customers ordered through the website, production took an estimated 90 days, sales were limited to a set number of customers and the capsule was available only in Spain, Germany, France and Italy. Founder Thomas Meyer said: 'We are very excited to test these designs made by our team and supported by artificial intelligence.' Desigual framed the benefit as lower stock rather than higher sales, and published no results.

Jules and imki. The French menswear chain Jules sold an 11-piece capsule of T-shirts, polo shirts, jogging suits and teddy jackets from 17 September to 15 October 2024, in stores and on its website, with prices starting at EUR 19.99, according to The Interline. The Paris start-up imki supplied 'product' AI models trained on curated datasets reflecting the brand's codes, and supported the design team. This is one of the few cases of AI-assisted design at mass-market prices in physical stores. Again, no sales results were reported.

How much of the design work did AI actually do?

Nobody quantifies it, and the descriptions consistently put people at the decisive points. At Revolve, the winners generated images with Midjourney and Stable Diffusion, sometimes edited them in Photoshop, and then developed products and materials with the retailer's team. At Collina Strada, which showed its Spring 2024 collection at New York Fashion Week in September 2023, the label fed its entire archive into an AI system, refined the looks it produced, fed them back over several weeks so the system learned the designer's preferences, and then had human designers tame the wilder outputs before the studio made the garments, according to The National.

The Nigerian label I.N Official, whose collection developed with Meta AI was shown at Africa Fashion World London on 9 August 2025, used the tool from colour and fabric selection through to storytelling, according to Meta. Prompts produced images, such as a silhouette blending Agbada and Hakama styles, that served as inspiration and mood boards, and Meta AI suggested accessories. Meta's announcement does not say whether the collection was sold.

The more ambitious projects are internal tools. Norma Kamali is building an AI model of her own design style as a succession tool, so that her team can keep designing in her spirit after she steps back, according to The Business of Fashion. The system, installed by Maison Meta on a computer in her office rather than in the cloud, is built on Stable Diffusion XL and trained on her archive, with about 10,000 images for swimwear alone, the first category. MIT News, reporting in April 2025, described a closed-loop tool trained only on her work. Her AI-generated designs appear on her website alongside her traditional collections; neither source confirms that any have been produced or sold.

The Business of Fashion has also reported that Collina Strada and the Danish label Heliot Emil use image-generating AI trained on their past collections to push their creative boundaries. In both cases the brand's own history is the raw material, and the designers decide what survives. That is a long way from the popular image of a machine designing a collection on its own, and it explains why brands find it hard to put a number on AI's share: the software contributes options, not decisions.

In China the same idea is being pursued at scale. In July 2026 Gao Dekang, chairman and chief executive of the down-jacket maker Bosideng, said the company had built a vertical AI model, the 'Bosideng AI Aesthetic Brain', covering everything from design conception to fitting and delivery, according to China Daily. He gave no figures on how many products it had designed or how they sold.

Do AI-designed collections sell?

There is one published claim, and it should be read carefully. Apparel Resources reported that the Korean officewear brand Dali Hotel, bought by the generative AI company Designovel, achieved KRW 100 million in weekly sales, with AI-generated trend analysis and design suggestions guiding product creation and AI used to set prices. The article carries no date and cites no original source, and its claim of 'no marketing expenditure' in the headline sits awkwardly with the statement in the text that Designovel spends 2.5% of monthly sales on marketing. Designovel's chief executive Ki-Young Shin was quoted with a notably cautious line: 'We need to wait a little longer until generative AI model can make money.'

Beyond that, the evidence is circumstantial. The capsules were small: 10 to 12 styles per label at Revolve, 11 pieces at Jules, and a made-to-order online offer in four countries at Desigual. Several were explicitly framed as tests. Desigual described its capsule as a way to cut stock; Jules ran its capsule for four weeks. A brand with a clearly successful product line typically has reasons to say so; the absence of any disclosure across three years of launches is itself informative, although it does not prove failure.

How widely are fashion companies using gen AI in design?

Less widely than the attention suggests, at least on the last published industry-wide measure. In the BoF and McKinsey State of Fashion 2024 survey, published at the end of 2023, 73% of fashion executives said gen AI would be an important priority for their businesses in 2024, but only 28% said they had tried it in creative processes for design and product development, according to Just Style. McKinsey estimated that up to a quarter of the value gen AI could create in fashion would come from design and product development use cases, and earlier put the total potential addition to the sector's bottom line at $150 billion to $275 billion, as reported by Fashion Dive.

Even then, the tools mentioned in the coverage were mostly about speed of ideation rather than finished products. The Business of Fashion reported that the start-up Cala helps designers produce more than 100 sketches in a day and that the AiDA system reportedly produces a dozen templates in 10 seconds. The bottleneck in fashion is rarely the number of sketches. It is choosing which ones to produce, in what quantity, at what price, and that remains a merchandising decision.

Who owns a design created with generative AI?

In the United States, the answer has become clearer in the past two years. On 29 January 2025 the US Copyright Office published the second part of its report on AI, concluding that outputs are protected only where a human author has determined sufficient expressive elements. Protection can cover a human-authored work that is perceptible in the output and 'creative arrangements or modifications of the output', 'but not the mere provision of prompts'. Using AI as an assistant does not prevent protection; the Office saw no case for new legal protection for AI outputs.

The courts have agreed. On 18 March 2025 the US Court of Appeals for the D.C. Circuit held in Thaler v. Perlmutter that the Copyright Act requires a human author, and on 2 March 2026 the Supreme Court declined to hear an appeal, according to the law firm Baker Donelson. The court did not decide how much human involvement is enough, because the claimant had disclaimed any. That question is being litigated in Allen v. Perlmutter in Colorado, which concerns an image made with Midjourney through many iterative prompts.

For fashion, this has two consequences. A design produced by typing prompts and selecting the best result may be impossible to protect as a creative work, which matters for anyone hoping to enforce rights against copyists. And the human contributions, selection, modification and arrangement, need to be documented if a brand wants to rely on them. The Copyright Office's January 2025 report left the separate question of training models on copyrighted works for a later part; the closed models trained only on a brand's own archive, as at Norma Kamali and Collina Strada, are one way to reduce that risk.

Baker Donelson's advice to businesses is practical: document human contributions, disclose AI involvement when registering works with the Copyright Office, and consider alternatives such as trade secrets and contracts. For a fashion house, contracts may matter most. Agencies, start-ups and freelance designers that generate AI imagery for a brand should assign whatever rights exist in their contributions, and the brand should know which elements of a final design came from a human hand.

What are the risks and limits of AI-designed collections?

• No proof of commercial advantage. Without published sales data, there is no public evidence that AI-assisted designs outperform, or underperform, human ones.
• Weak IP protection. Prompt-only designs may not be protectable in the United States, which leaves them open to copying.
• Training data exposure. Models trained on other people's work raise questions the Copyright Office has deferred; archive-only training reduces but does not remove the risk.
• Brand dilution. Kamali's own caveat applies: 'When a designer designs a collection, it has to have a soul.' Output trained on a brand's past can also lock it into its past.
• Reporting gap. Vague claims about AI's role make it impossible for buyers, investors or customers to judge what they are getting.

There is a counter-argument. Several of these projects were never meant to prove that AI sells better. Desigual tested made-to-order production, Revolve used AI Fashion Week to find new designers, and Kamali is planning succession. Judged on those aims, the experiments may have done their job. Nima Abbasi, a partner at Maison Meta, summed up the lesson of AI Fashion Week to CNN: 'You need vision, you need creativity and then you need to learn the tools.'

What should fashion companies do now?

1. Set a clear commercial test Launch AI-assisted styles alongside comparable human-designed ones and measure sell-through, full-price share and returns over the same period.
2. Document the human contribution Record who selected, modified and arranged each design, so that the work can be defended under copyright law.
3. Train on what you own Prefer models trained on your own archive and licensed material, and keep a record of the training data.
4. Keep merchandising decisions human Use AI to widen the range of ideas, but leave choices on quantities, prices and assortment to buyers and merchandisers.
5. Be specific in public claims Say what the AI did, for example mood boards, prints or silhouettes, rather than calling a collection 'AI-designed'.
6. Review supplier and partner terms Check who owns AI outputs created by agencies, start-ups and freelance designers working for you.