Fashion
Last reviewed
May 13, 2026
Sources
40 citations
Review status
Source-backed
Revision
v2 ยท 4,275 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 13, 2026
Sources
40 citations
Review status
Source-backed
Revision
v2 ยท 4,275 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: Fashion ChatGPT Plugins
AI in fashion refers to the use of artificial intelligence, machine learning, and computer vision across the apparel, footwear, and accessories industry. Applications span personal styling, recommendation systems, demand and trend forecasting, generative design, AI-generated models for marketing imagery, virtual try-on, supply chain optimisation, counterfeit authentication, and sizing prediction. The industry was an early adopter of recommendation algorithms through companies like Stitch Fix, founded in 2011, and has more recently become a focal point for controversies involving copyright, labour displacement, and algorithmic copying.
Fashion combines large volumes of visual data, short product life cycles, and chronic forecasting problems, all of which make the industry well suited to AI. Retailers process millions of product images and customer photos a year; designers iterate on silhouettes, colours, and prints; brands try to predict demand for items that may sell out in days or sit in warehouses for seasons. Algorithms can sort through these volumes faster than humans, but they also introduce new problems: replication of copyrighted work, displacement of models and designers, and amplification of fast fashion's environmental footprint.
The field roughly breaks into seven application areas:
| Area | What AI does | Representative companies |
|---|---|---|
| Recommendation and styling | Picks items for individual shoppers based on preferences and history | Stitch Fix, Thread, Lyst |
| Trend forecasting | Predicts which colours, silhouettes, and prints will sell | Heuritech, WGSN, Edited, Stylumia |
| Generative design | Produces new garment designs from text or image prompts | CALA, Maison Meta, Resleeve |
| AI-generated models | Creates synthetic humans for catalogue imagery | Lalaland.ai, The Diigitals, Aww Inc. |
| Virtual try-on | Shows clothing on a shopper's body or a selected model | Zeekit, Google, Snap |
| Supply chain | Forecasts demand and routes inventory | Blue Yonder, o9 Solutions, SAS |
| Authentication and sizing | Verifies luxury goods and predicts fit | Entrupy, True Fit, 3DLOOK |
Fashion's early adoption of recommendation algorithms predates the current generative AI wave by a decade. Stitch Fix launched in 2011 with a hybrid model that combined human stylists and statistical recommendation, then expanded its data science team aggressively after hiring Eric Colson, former Netflix vice president of data science, in 2012. By the time the company went public on Nasdaq in November 2017 at a valuation around $1.6 billion, it had more than 80 data scientists reporting directly to the CEO and an in-house publication called "Algorithms Tour" that explained its modelling stack.
In 2016, IBM's research division collaborated with the New York fashion house Marchesa to design a "cognitive dress" for the Costume Institute Gala (Met Gala). The dress, themed "Manus x Machina," used IBM Watson to associate Twitter sentiment about the gala with five emotions chosen by Marchesa, then drove LED lights embedded in the fabric to shift colour in response to the social media feed. Karolina Kurkova wore the gown on the red carpet, and the dress later entered the permanent collection of the Henry Ford Museum.
Virtual influencers and digital models emerged on Instagram in the same period. Cameron-James Wilson, a British photographer experimenting with 3D modelling after hand-painting Barbie dolls, created Shudu Gram, often described as "the world's first digital supermodel," and launched her Instagram account on 21 April 2017. Wilson founded The Diigitals, an all-digital modelling agency, in 2018. The Japanese virtual human Imma, with her signature pink bob, debuted in 2018 under Tokyo studio Aww Inc. and went on to appear in campaigns for Porsche Japan, Magnum, Hugo Boss, Tommy Jeans, and IKEA.
Generative tools arrived in earnest after the release of DALL-E 2 and Stable Diffusion in 2022. On 20 October 2022, the fashion supply chain platform CALA announced one of the first enterprise integrations of the DALL-E API, letting designers generate apparel, accessory, and footwear concepts from text prompts and reference images. A wave of brand experiments followed in 2023, including the Levi's and Lalaland.ai partnership that became the year's most discussed AI fashion controversy.
Stitch Fix remains the standard reference for algorithmic personal styling. Customers complete a style profile, and the company's models combine collaborative filtering, content-based features (size, fabric, season, brand affinities), and merchandiser inputs to assemble a box of five items that a human stylist then reviews and ships. Returns and customer feedback feed back into the system. The company has published unusually detailed public explanations of its stack, including the "Algorithms Tour" microsite that walks through styling, demand, inventory, and warehouse routing models.
Thread, a London-based menswear platform acquired by Marks and Spencer in 2021, runs a similar hybrid. Its in-house engine, called Thimble, searches across roughly 500 partner brands to surface items that fit a customer's stated style, fit, and budget, then human stylists curate the final outfit. Co-founder Kieran O'Neill has framed the algorithm as a way to scale a stylist's expertise rather than replace it.
Lyst, the fashion search engine founded in 2010, indexes over 17,000 brands and uses machine learning to rank and personalise results. Its annual Lyst Index report ranks the hottest brands and products based on aggregated search and engagement data. The company has filed patents around visual search and outfit completion. Newer entrants include Glance AI, an AI shopping app launched by India-based Glance that generates personalised lookbooks and renders the user's likeness wearing recommended items.
Chatbot stylists also fall into this category. Sephora's Sephora Virtual Artist, built on technology from Toronto-based ModiFace (acquired by L'Oreal in 2018), launched in 2016 and added AI-powered colour matching in 2017. The app uses facial geometry detection to apply digital lipstick, eyeshadow, and other products through the phone camera. Sephora reports that customers who used the tool became significantly more likely to complete a purchase, and that returns on cosmetics fell after rollout. H&M and Walmart have both fielded AI chat assistants on their apps, and Walmart launched a generative shopping assistant called Sparky in 2025.
Forecasting which colours, prints, and silhouettes will be in demand has historically been the work of trend agencies like WGSN and its competitor Promostyl, which sent analysts to clubs, galleries, and street markets to write seasonal trend books. AI has not replaced these analysts but it has changed the inputs: a forecaster can now feed millions of social media images, search queries, and runway photos into a model and quantify a trend's trajectory.
| Platform | Headquarters | Founded | Approach | Notable clients |
|---|---|---|---|---|
| Heuritech | Paris | 2013 | Image recognition on roughly 3 million social posts daily, identifies more than 2,000 fashion attributes | Louis Vuitton, Dior, Adidas, Moncler, Paco Rabanne, Kontoor Brands |
| WGSN | London | 1998 | Trend reports and a data platform combining qualitative and quantitative inputs; serves 6,500+ brands | Mass and luxury retailers worldwide |
| Edited | London | 2009 | Real-time retail market intelligence, scrapes product, price, and assortment data | Stella McCartney, Levi's, Gap |
| Trendalytics | New York | 2014 | Combines search, social, and sales signals; tracks Gen Z signals | Targets and specialty retailers |
| First Insight | Pittsburgh | 2007 | Predictive analytics using consumer voting data and machine learning | Vans, BJ's Wholesale, Kohl's |
| Stylumia | Bangalore | 2015 | True Demand forecasting filters noise from search and social signals | Aditya Birla Fashion, Reliance Retail, Levi Strauss |
Heuritech, founded in 2013 by two PhDs in machine learning, became one of the most cited examples after Louis Vuitton commissioned a bespoke system in 2015. The platform reports forecast accuracy above 90 percent for specific item categories. Stylumia, based in Bangalore, has argued that filtering out social media "noise" (memes, bots, brand promotions that do not translate into purchases) improves prediction accuracy by 20 to 40 percent compared with raw social listening.
Generative models have given designers a way to iterate on sketches, prints, and silhouettes without a sample room. CALA, founded in 2016 by Andrew Wyatt, formerly of Louis Vuitton, integrated DALL-E 2 in October 2022 and was one of the first enterprise deployments of the API. A user describes an item in plain English ("oversized bomber jacket with floral embroidery on the back") or uploads reference images, and CALA returns six concept renders that can be edited and pushed through to manufacturing on its supply chain platform.
Maison Meta, founded in Paris by Cyril Foiret in 2022, brands itself as the first generative AI agency for fashion and beauty. Its proprietary tool, Seeed, includes AI Twin training, virtual try-on, faceswap, and inpainting. The agency has produced campaigns for Moncler, Pangaia, and Revolve, and it runs the annual AI Fashion Week show in New York, which debuted in April 2023. Other generative platforms include Resleeve, Designovel, Off/Script, and Lookai.
The Dutch designer Iris van Herpen sits at the boundary between fashion design and computational art. She introduced what is widely cited as the first 3D printed couture collection in 2010 and has since worked with parametric design software, generative algorithms, and biomaterials. Her 2017 "Cellchemy" jewellery pieces were generated from a 3D facial scan and printed on a multi-material printer in collaboration with Delft University of Technology. Van Herpen does not describe her practice as AI design, but her use of generative geometry and high-resolution simulation prefigured many of the workflows now appearing in mainstream brands.
Pure Solutions and Marchesa's 2016 "cognitive dress" with IBM Watson, mentioned above, is sometimes called the first AI-designed garment, though it is more accurate to say the dress was designed by humans informed by Watson's color recommendations and powered by Watson sentiment analysis during the gala itself.
AI-generated models offer brands cheaper imagery, faster turnaround, and a way to display the same item on bodies of multiple sizes and skin tones without organising additional photo shoots. The category became mainstream after the Levi Strauss announcement in March 2023.
On 22 March 2023, Levi Strauss and Co. published a press release titled "LS&Co. Partners with Lalaland.ai" announcing a partnership with the Amsterdam-based startup Lalaland.ai, founded in 2019 by Michael Musandu (born in Zimbabwe and raised in South Africa) and Ugnius Rimsa. Levi's framed the pilot as a way to "increase the number and diversity of our models for our products in a sustainable way."
Within a week the announcement triggered a wave of criticism. Models, photographers, and DEI commentators accused the company of using synthetic Black, Latino, and Asian models as a cheap substitute for hiring real people of colour, and of conflating algorithmic representation with structural diversity. The story was covered by NBC News, Vogue Business, The Drum, and the trade press. On 28 March 2023, Levi's appended an editor's note to the original press release saying the company did not see the pilot as a substitute for hiring diverse human models or for the company's broader DEI commitments, and that live shoots would continue.
The Levi's case became the most-cited example of brand miscalculation around AI imagery and is taught in the Wiley Journal of Contingencies and Crisis Management as a case study in fashion crisis communication. Lalaland.ai was acquired by Israeli 3D fashion software company Browzwear in May 2024 for an undisclosed sum.
H&M announced in March 2025 that it would create digital twins of 30 of its human models, working with Stockholm-based AI agency Uncut. Models would retain rights to their digital twins, be compensated when those twins were used, and could license them to other brands. Images would carry watermarks identifying them as AI-generated. The first images released, a denim campaign shot against fashion-capital backdrops, drew criticism from working models and a measured response from some of the models featured.
Hugo Boss, Mango, Marks and Spencer, and several other brands have run smaller AI imagery experiments. The Shudu Gram and Imma accounts, while not branded by any one retailer, continue to land sponsorship campaigns. After early criticism that Wilson, a white man, was profiting from a Black virtual model, The Diigitals adjusted its practice to credit and compensate the Black model Alexsandrah, who has appeared as Shudu in major campaigns.
Virtual try-on, the rendering of a garment on a shopper's photograph or on a model selected to resemble them, has gone through several waves. Early attempts in the 2010s used 3D body scanning and were largely confined to in-store kiosks. The current generation uses image-to-image diffusion models and computer vision to drape, fold, and shadow a garment on a 2D photograph.
Zeekit, an Israeli startup founded in 2014 by CEO Yael Vizel, CTO Alon Kristal, and Nir Appleboim, built a real-time image segmentation system that mapped a person's silhouette and a clothing item into shared coordinates. Walmart announced its acquisition of Zeekit on 13 May 2021 for undisclosed terms, and integrated the technology as "Choose My Model" and "Be Your Own Model" features on Walmart.com.
Google launched a virtual try-on feature for apparel in June 2023 inside Google Shopping, initially for women's tops from Anthropologie, Everlane, H&M, and LOFT, expanding to men's tops by year end. The system, described in a research paper by Google researchers, was trained to render a single garment image on a diverse set of real human models in sizes XXS to 4XL. Google cited internal data showing that 42 percent of online shoppers do not feel represented by site imagery and that 59 percent feel dissatisfied with how a purchase compares to its product page.
Snap operates a large augmented reality shopping business through Snapchat. Brands embed AR "Lenses" that let users try on glasses, sneakers, jewellery, and cosmetics. Partnerships announced or expanded between 2021 and 2024 include Amazon Fashion (eyewear), Gucci, Dior, MAC Cosmetics, Puma, American Eagle, and H&M. Snap reports that more than 250 million users have engaged with AR shopping Lenses more than 5 billion times, and Ulta Beauty has cited $6 million in sales within two weeks of launching a catalog Lens.
Amazon ran an earlier, more hardware-centred experiment with the Echo Look, a $200 Alexa-powered camera and clothing assistant introduced in 2017. The device photographed a user's outfit and rated it through a feature called "Style Check." Amazon discontinued the Echo Look in May 2020, ceasing app and device support on 24 July 2020. Style recommendations were folded into the Amazon shopping app.
Other notable virtual try-on companies include Vue.ai (the consumer brand of Mad Street Den, an Indian AI startup founded in 2013 by Anand Chandrasekaran and Ashwini Asokan), Perfect Corp. for beauty, 3DLOOK, and Drapr.
Apparel supply chains are notoriously hard to forecast because demand depends on weather, social trends, and pricing in short windows. AI-driven demand planning tools attempt to model these signals at SKU-store granularity. Blue Yonder, o9 Solutions, Kinaxis, SAS, and RELEX Solutions compete for enterprise contracts.
Blue Yonder, owned by Panasonic since 2021, sells an "Intelligent Allocation" product targeted at short-lifecycle fashion goods that splits limited stock across stores and digital channels to maximise full-price sell-through. o9 Solutions emphasises external data ingestion, with a "digital brain" architecture that combines internal sales data with market indicators, competitor prices, and social signals.
Warehouse automation, although technically robotics rather than AI in the narrow sense, increasingly relies on computer vision and reinforcement learning. Uniqlo, part of Japan's Fast Retailing, automated its Ariake distribution centre in Tokyo in 2018 in partnership with Daifuku, reportedly replacing about 90 percent of warehouse staff with robots that read RFID tags, sort garments, and prepare orders for shipment. Fast Retailing has since extended the programme to other Japanese sites and worked with French robotics firm Exotec and Tokyo-based Mujin.
Luxury counterfeit goods are a multi-billion-dollar market, and AI-driven authentication has become an established mid-market service.
Entrupy, founded in 2012 by Vidyuth Srinivasan and Lakshminarayanan Subramanian, sells a handheld microscope and an iOS app that captures around 500 microscopic images of an item's surface and compares them against a reference database of authentic and counterfeit examples. The company reports a 99.86 percent stated accuracy rate and offers a financial guarantee on each authentication. Most clients are pawnshops, resale platforms, and consignment stores rather than the original brands themselves; luxury houses generally do not use third-party authentication.
Competitors include Real Authentication, ALPS, and Authenticate First. LVMH led a $11 million Series A in Entrupy in 2019, signalling some level of acceptance of the technology within the luxury sector.
Return rates in online apparel run between 20 and 40 percent in many markets, with sizing the most cited cause. Several companies sell AI sizing services to retailers.
True Fit, founded in 2010 and headquartered in Boston, builds personalised size recommendations from a stated user profile (height, weight, age, favourite brands) and a proprietary dataset of returns, purchases, and product attributes. The company has cited a 65 percent market share in size and fit solutions and 20 years of data covering more than $600 billion in transactions. It launched an agentic AI shopping assistant in 2026.
Fit Analytics, founded in Berlin in 2010, was acquired by Snap Inc. in 2021 and is widely deployed at H&M, Zara, and Asos. 3DLOOK takes a different approach: a user takes two smartphone photos, and the company's computer vision system extracts more than 80 measurements and generates a 3D avatar in under a minute. Bold Metrics, Secret Sauce Partners, and Volumental (for footwear) round out the category.
Chinese-founded fast fashion giant Shein became the central case study for AI-amplified copying. On 11 July 2023, three independent designers, Krista Perry of Massachusetts, Larissa Martinez, and Jay Baron, filed a civil suit in the Central District of California alleging that Shein had copied their designs at scale. The complaint accused Shein of running a "secret algorithm" that ingested trending images from social media and surfaced them for production on a 25-day cycle from idea to delivery, and it invoked the Racketeer Influenced and Corrupt Organisations Act (RICO), an unusual choice in a copyright case more commonly associated with organised crime prosecutions.
On 22 January 2024, Judge Mark Scarsi granted in part and denied in part Shein's motion to dismiss, but on 8 November 2024 he largely preserved the RICO claim, ruling that the plaintiffs had adequately alleged that Shein's copyright infringement could serve as a predicate act under the RICO statute, and that Shein had "willfully committed copyright infringement and purposefully created and employed an algorithm that generates exact or close copies of works it does not own." The case settled in September 2025 on undisclosed terms.
Legal commentators have noted that the "secret algorithm" in the Shein case is not, strictly, a generative AI model. The system most likely scrapes trending images, ranks them by predicted virality, and assigns them to factories. The case nonetheless became the most cited example of how algorithmic systems can scale a copyright infringement problem from a one-off legal dispute into a systemic complaint.
During the 6 May 2024 Met Gala, AI-generated images of Katy Perry and Rihanna on the red carpet went viral despite neither performer attending. The first AI image of Perry, showing her in a floral gown trimmed with moss that matched the gala theme "Sleeping Beauties: The Garden of Time," gathered more than 288,000 likes and 9 million views on X within three hours. A second image showed her in a corseted gown with a vineyard-inspired train and a key breastplate. A circulated AI image of Rihanna depicted her in a mermaid-style gown with an armpiece featuring flora and fauna.
Perry's mother sent her a complimentary text about the look. Perry replied: "lol mom the AI got you too, BEWARE!" and posted the exchange to social media. Perry later said she was at work in a recording studio at the time of the gala. Dua Lipa was also the subject of viral AI gala images. The platform later added community notes to identify the images as AI-generated.
The episode prompted commentary about the use of generative AI to fabricate celebrity images, the difficulty of policing deepfakes in real time on major social platforms, and the way an event like the Met Gala, where elaborate dresses are expected, creates ideal conditions for AI fakery.
AI in fashion has two opposing environmental stories. On one side, demand forecasting and inventory routing can reduce overproduction, which is responsible for a substantial share of the textile industry's carbon footprint. Stylumia's True Demand product, for instance, claims that better forecasting can cut overproduction by 20 to 40 percent. Reformation, the Los Angeles brand known for its sustainability claims, uses AI for inventory and demand planning.
On the other side, AI accelerates fast fashion. Shein's ability to introduce thousands of new SKUs a day depends on a software pipeline that scrapes trends, allocates production, and routes shipments at a speed that traditional retailers cannot match. Critics argue that AI design tools democratise garment creation but also reduce the cost of copying, making it easier for fast fashion brands to clone independent designers' work overnight. The Shein lawsuits sit at the centre of this debate.
The labour debate is parallel. AI-generated models, as the Levi's and H&M cases illustrate, raise the question of whether algorithmic diversity is a substitute for hiring diverse human models. Working models, photographers, makeup artists, and stylists have argued that synthetic imagery shifts a structural labour problem onto a software vendor without addressing it. Brands have responded with various compromises: H&M's compensation-and-license model for digital twins, The Diigitals' practice of crediting and paying the human references for its avatars, and Levi's commitment to continue live shoots.
Intellectual property is the third axis. Generative models trained on web-scraped fashion imagery raise the question of whether designers can claim infringement when an AI tool produces an item that resembles theirs. The U.S. Copyright Office has historically been reluctant to grant copyright on garment designs (clothing is generally treated as a useful article, with limited copyright protection for separable graphic elements). That doctrine, combined with the difficulty of proving substantial similarity in AI-generated outputs, has left small designers with limited legal recourse outside of pattern, print, and brand-name protections.