Virtual try-on in 2026: generative, 3D and AR approaches compared

Virtual try-on is not one technology but three. Generative try-on uses AI image models to place a garment on a photo of the shopper; 3D try-on builds an avatar from body measurements and simulates garments in different sizes; augmented reality (AR) overlays a 3D product on a live camera view. Generative tools scale fastest, 3D tools are the ones with published evidence on fit returns, and AR remains strongest for shoes, eyewear and accessories.

What is virtual try-on and why are there different approaches?

Every virtual try-on tries to answer a shopper's uncertainty before purchase, but the uncertainty differs. Some shoppers want to know whether a colour or silhouette suits them. Others want to know which size fits. A third group wants to see a product, such as a trainer or a pair of sunglasses, in place on their own body in real time. Each question demands different data, which is why the market has split into distinct technical families rather than converging on one.

How does generative virtual try-on work?

Generative try-on uses an image model to synthesise a picture of the shopper wearing a garment. Google is the most prominent example. At its I/O conference in May 2025 it announced a try-on experiment in which US shoppers upload a full-length photo of themselves and see apparel listings on their own body. Google described a custom image generation model for fashion that accounts for how materials fold, stretch and drape on different bodies, and said the feature works across billions of listings in its Shopping Graph. In October 2025 TechCrunch reported that Google had extended the feature to Australia, Canada and Japan and added shoes, and noted that Amazon and Walmart had introduced similar features.

The strengths are scale and low friction. Because the model works from ordinary product images, a platform can enable try-on across a huge catalogue without building a 3D model of each garment. The weakness is precision. A generated image is a plausible rendering, not a measurement. It cannot reliably show that size 38 will pull across the shoulders while size 40 will not, because the model does not know the garment's pattern dimensions or the shopper's body measurements.

Read also

PIM explained: why product data is a fashion brand's hidden asset

How does 3D virtual try-on differ?

3D try-on starts from measurements. The shopper's body is represented as an avatar built from measured or predicted dimensions, and the garment is represented with its real size specification and, ideally, fabric properties. The system can then show how different sizes of the same item sit on that body.

Zalando has published the most detail on this approach in fashion e-commerce. In a June 2026 corporate article it said its virtual fitting room uses body measurements to create a 3D avatar so customers can compare how different sizes of the same item might look. The retailer pointed out that a single jeans style can come in more than 30 size combinations and that around 21% of its customers buy jeans each year. It reported that recent pilots reduced return rates by up to 40%, and said its 2026 goal is to roll the experience out to all customers and expand the assortment. The body measurements behind it come from a phone-based tool using two photos or a video, which Zalando says more than 1.5 million customers have tried.

The cost of the 3D approach is data. Each style needs accurate measurements for each size, and convincing drape needs fabric information. That is why 3D try-on is typically rolled out category by category, often starting with items such as jeans where fit drives most returns.

Where does augmented reality try-on fit?

AR try-on uses the phone camera to place a 3D model of a product onto the shopper's body in real time. It is most effective for rigid or semi-rigid products whose shape does not change much when worn. Amazon's virtual try-on for shoes, launched in June 2022 in its iOS shopping app in the US and Canada, is a well-known example: shoppers point the camera at their feet to see the shoe in place, switch colours and share a photo. TechCrunch reported that brands including New Balance, Adidas, Reebok, Puma and Asics were supported at launch.

AR is less suited to soft garments. Simulating how a dress drapes over a moving body in real time on a phone is far harder than tracking a shoe on a foot, and results for clothing tend to look like stickers rather than garments. AR's role in fashion is therefore strongest in footwear, eyewear, watches, jewellery and bags.

Which virtual try-on approach reduces returns?

Published evidence on returns currently favours measurement-based approaches. Zalando's reported results come from tools that combine body measurements, product data and return information, not from image generation alone. Google has not published data on whether its generative try-on changes return rates, and its own Doppl app carried the caveat that fit and clothing details may not always be accurate. That does not mean generative try-on has no effect; it means the effect is unproven in public data.

• If the problem is size and fit returns, measurement-based size advice and 3D comparison have the stronger evidence base.
• If the problem is low confidence about style or colour, generative try-on may help, though its commercial impact is not yet documented publicly.
• If the products are rigid accessories or footwear, AR offers a mature, real-time experience.
• If the catalogue is large and fast-changing, approaches that work from existing images scale more easily than per-style 3D modelling.

Read also

Agent payments and checkout protocols: what fashion retailers need to know in 2026

What should brands consider before choosing?

1. Identify the main reason for returns in each category: if size and fit dominate, start with measurement data rather than imagery.
2. Check what product data exists: graded size specifications and fabric properties are prerequisites for credible 3D fit.
3. Consider where try-on will happen: on the brand's own site, or on platforms such as search engines and marketplaces that use the brand's images.
4. Plan for consent and privacy, since all three approaches process photos or body data of shoppers.
5. Measure outcomes by cohort, comparing return rates and conversion for shoppers who use try-on with those who do not.

The approaches are converging at the edges: 3D systems add generative rendering for realism, and generative systems will likely draw on more structured size data over time. For now, brands get the best results by matching the method to the specific shopper question they need to answer.