Virtual Try-On in 2025: Something Has Actually Changed

I've been watching virtual try-on with the resigned patience of someone who has heard many times that fusion energy is only twenty years away. The demos are always promising. The consumer adoption numbers are always, on closer inspection, less so.

Something has actually shifted in 2025. I want to be honest about what has and hasn't changed, because the technology is genuinely better and the marketing is still doing its thing.

What's New

Google updated its virtual try-on feature in July, adding brands and categories. By December it had launched a selfie-based version: upload a photograph, and a full-body digital model is generated from your face using the Gemini 2.5 Flash Image model. You try on clothing on that model rather than selecting from a preset roster. Google also brought the tool to the UK and India before the year was out, which matters for anyone thinking this is just a US story.

The participating brand list has grown: Anthropologie, Everlane, H&M, LOFT, Levi's, Abercrombie & Fitch, Adidas. Not universally available for all products, but no longer experimental in the way it felt a year ago.

Snap has been building out its ARES platform for fashion retailers. The distinction worth understanding: layering a garment image onto a generic body shape looks like a compositing job. ARES uses 3D body mesh and cloth simulation to drape garments over the user's actual proportions in real time. That is a harder problem, and it looks more like what it is supposed to simulate.

The consumer engagement data is also better than it has been. Google's own data, reported via eMarketer, found that virtual try-on listings in Google Search receive 60% more high-quality views than standard listings. Shoppers try on an average of four different models per product and are more likely to visit a brand's website afterwards. Those numbers are directionally encouraging. Bear in mind they come from Google and should be read as such.

What Hasn't Changed

The fundamental tension is that fit is mostly about more than appearance. Whether something looks good on a generated version of you is a much easier problem than whether it fits correctly, feels comfortable, and behaves the way you expect. The visual representation is useful. It doesn't replace trying the thing on.

This matters most for the categories where returns are highest: tailored clothing, fitted knitwear, anything where cut and drape vary significantly across garments that share a size label. These are also the categories where the economic case for reducing returns is strongest. The technology is most needed exactly where it's most limited.

There is also the data question. A selfie-based try-on tool requires uploading a personal photograph to a platform. For consumers, comfort levels vary. For fashion retailers building their own version, the data governance obligations under UK GDPR are not trivial: facial images processed to identify individuals attract special category status and require explicit legal basis and a data protection impact assessment.

What Retailers Should Actually Do

For retailers already participating in Google's Shopping Graph, the practical action is clear: make sure your products are properly indexed for virtual try-on, check that your product imagery meets the quality requirements for the feature to work well, and monitor whether it is affecting engagement metrics and return rates for the relevant products. The lift is low and the data is real.

The case for building a standalone virtual try-on experience is harder. Development cost is meaningful. Consumer adoption requires friction-free onboarding that most bespoke implementations do not achieve. The technology gap between what Google or Snap can offer at scale and what a mid-market retailer can build in-house is large.

The honest answer for most retailers: participate in the platforms where try-on is available, monitor the data, and revisit the standalone case in 12 to 18 months when the self-service tooling improves.

The market projections are as bullish as they always are. The global market was valued at around $5.8 billion in 2024 and is projected to reach $27.7 billion by 2031, according to Valuates. Apply the usual discount to optimistic market forecasting. The direction is credible even if the magnitude isn't.

Something has changed. It's just not yet everything that needs to change.