It’s been nearly eight years since Zalando first began experimenting with virtual try-on.
In 2018, the German e-tailer began developing its own software that can generate photos of customers wearing items from its catalogue. But the arrival of open-sourced large language models changed the stakes, allowing Zalando to improve its image-generation capabilities enough to start rolling out virtual try-on to customers. It began running tests with small groups in 2022, gradually adding new features, like a measurement tool that can create an image of a person’s body shape from a picture and 3D rotating imagery. Even in the test phase, the company saw benefits: an April 2023 test led to a 40 percent reduction in returns, said Reza Shirvany, Zalando’s director of applied science.
Now, Zalando is ready to launch its virtual try-on for all customers in 2026.
Generative AI “reduced a lot of the barriers for businesses, including ours, to innovate for customers,” Shirvany said. “The scientific reason for building this type of technology in-house is less and less because the world is investing into LLMs.”
Online retailers have long hoped that virtual try-on would help reduce their return rates, but the technology often couldn’t accurately depict how clothes would appear on a real person. Generative AI, however, has gotten it closer to fulfilling that potential, making try-on imagery more photorealistic and able to display how specific fabrics and cuts are likely to lay on a range of bodies.
Those advancements have given way to a new cadre of AI-powered virtual try-on startups including Doji, Zelig and Stiled. Software giants are also betting big on the technology. Since 2023, Google has been expanding the number of products in its shopping catalogue that it lets users see virtually fit onto digital models from sizes XXS to 4XL. In May, the company released a feature that allows users to upload photos of themselves to see how they’d look in any apparel product from its listings, and this month it introduced a new feature that lets them generate try-on imagery from a selfie. Virtual try-on is now the number one Google Shopping tool that users share on social media, said Lilian Rincon, Google’s vice president of product management.
“Generative AI is mainstream. Everybody’s using it and they’re starting to trust this technology,” said Jen Jones, chief marketing officer at e-commerce software firm Commercetools. “People who would have never ever uploaded a selfie of themselves to something like [Google’s] Gemini two years ago would do it without thinking twice today.”
But there’s still a way to go before mass adoption. An October survey from eMarketer of nearly 1,000 adults aged 18 to 65 found that only 1.4 percent of them regularly use some form of virtual try-on.
To make the technology an essential part of e-commerce, software startups are improving the speed at which images are generated, developing advanced features like 3D imagery and videos and making the images shareable on social media, to convince a larger swath of consumers that the technology is fun and easy to use.
“You’re going to see virtual try-on be more and more great in terms of the fidelity of the image and also personalised,” said Amy Wu Martin, an investor at Menlo Ventures. “It’s about how well that then translates to the user wanting to buy clothes.”
How Generative AI Changed Virtual Try-On
Historically, virtual try-on often required users to scan their full body to render an image, or create an avatar that served as an approximation of what they actually look like. Generative AI models, on the other hand, allow users to upload images of themselves and instantly see “try on” clothes, while accounting for important features like “the nuance of the fabric,” Google’s Rincon said.
Those improvements have helped companies ensure that garment details such as graphics on a T-shirt or the lapel shape on a blazer are accurately displayed. That also means that retailers are able to use try-on software on harder-to-render categories such as denim, which can carry up to 100 sizes.
“We are no longer in a world where this is being done through augmented reality. Gen AI is creating something that feels like you in real life,” said Juan Pellerano-Rendón, chief marketing officer at software firm Swap. “That’s what’s going to create more demand.”
Gen AI is also helping virtual try-on startups reduce in-house development costs. In March, four-year-old Zelig, which creates try-on technology for retailers like Revolve, started mixing open-sourced LLMs with its proprietary software that depicts the natural movement of fabrics — think the bunching at the top of a pair of pants. Because they were able to create those images by blending their existing model with generative AI, the company has saved millions, said Sandy Sholl, Zelig’s founder and chief executive.
“It wasn’t scalable the way we were doing it,” Sholl said.
Software providers also use LLMs to experiment with complex features like rotating images and videos. When Swap launched its service for brands to create their own AI agents in September, it included a try-on tool that generates both still images and videos that show users turning around in their clothes, Pellerano-Rendón said.
“We’re making sure that if you’re turning or walking, the way that the clothes fit on you, and your natural body movements, are true to how you would try something on,” he added.
The Future of Virtual Try-On
All these improvements, however, don’t mean much if people don’t actually use virtual try-on in practice.
To attract consumers, some startups are reframing it as a styling tool rather than a return solution. For instance, at Doji, which launched in 2024, the app’s most-used feature lets users shuffle outfits and receive style and product recommendations, said Dorian Dargan, Doji’s co-founder and chief executive.
“Virtual try-on allows people to dip their toe in the water and really build an emotional relationship with the brand,” Dargan said. “If you’re primarily just focused on returns and conversion, you might miss that opportunity around inspiration and exploration.”
Instead of waiting to partner with brands directly, some virtual try-on providers are expanding their reach through browser extensions. In November, Doji launched its extension on Apple’s Safari to let users generate images with products from a larger number of brands across the internet. AI-powered shopping app Stiled overlays products from any brand or retailer’s site onto images of its users. The hope is that Stiled can use those results to convince brands to sign on as partners, said Will Hanson, the company’s founder and chief executive.
“You’re basically figuring out how to make [virtual try-on] a daily habit,” Wu Martin said. “It comes down to just seeing it more.”
The next challenge is proving to brands that virtual try-on is unequivocally a money maker. Zelig’s Sholl says it will be critical for software providers to partner with brands and organise data to directly attribute any bump in sales to try-on software.
“The real shift comes when everybody knows that this technology makes everybody a lot of money,” Sholl said.
