Adobe Commerce development
Storefront design
Performance & SEO
AI & agentic commerce
Release tracker
AI virtual try-on lets shoppers see a product on themselves, or on a model that resembles them, before they buy. It uses either generative AI that renders a garment onto a photo or augmented reality that tracks a face, hand or room through the camera in real time. It pays off in categories where fit and appearance drive purchase decisions and returns, such as eyewear, makeup, apparel and furniture, and much less for products where specifications matter more than looks.
This guide explains how the main approaches work, which product categories benefit, and the questions to ask a vendor before you sign.
How AI virtual try-on works
Generative image try-on
The shopper uploads a photo, or chooses a model with a similar body type, and an image model generates a picture of them wearing the item. The model has to preserve the garment's details (pattern, logo, drape) and the person's pose and proportions. Google, for example, offers a try-on feature in Google Shopping that works from a shopper's own full-length photo; its guide (updated November 2025) covers shirts, pants, dresses and shoes. Availability by country and supported categories change, so check Google's current pages for your markets.
Real-time augmented reality
The phone or laptop camera tracks a face, wrist, foot or room, and a 3D model of the product is overlaid live. This is mature for eyewear, makeup, jewelry and watches, and for placing furniture in a room. It needs accurate 3D assets for each product, which is usually the largest cost driver.
Fit and size recommendation
Some tools do not show an image at all. They estimate fit from body measurements, a short questionnaire or past purchases and returns, then recommend a size. For apparel, this can matter more to returns than a visual preview, because a garment can look right and still fit badly.
Category fit: which products benefit
The table below is a starting point for deciding whether try-on is worth evaluating. "Fit" is about how much the technology addresses the shopper's real uncertainty, not about any guaranteed result.
| Category | Main shopper doubt | Best-suited approach | Fit |
|---|---|---|---|
| Eyewear and sunglasses | How frames look on my face | Real-time AR face tracking | Strong |
| Makeup and hair color | Shade match with my skin or hair | Real-time AR | Strong |
| Apparel (tops, dresses, outerwear) | Look, drape and size | Generative try-on plus size recommendation | Good for look; size needs fit data |
| Footwear | Style on foot; size and width | AR or generative for look; fit tool for size | Moderate |
| Jewelry and watches | Scale on wrist, ear or neck | Real-time AR | Good when 3D models are accurate |
| Furniture and decor | Size and style in my room | AR room placement | Strong for large items |
| Hats and accessories | How it looks on me | AR or generative | Moderate |
| Technical and spec-led goods | Compatibility and performance | Not try-on; better specs and comparison tools | Weak |
Two practical filters help narrow the list further. First, look at your return reasons: if "doesn't suit me" or "looks different from the photos" is common in a category, try-on targets a real problem. Second, look at catalog size and churn: a try-on that needs a 3D model per SKU is easier to sustain for a stable range than for fast-fashion collections that change every few weeks.
When it pays off, and when it does not
Try-on tends to make sense when the category scores well above, the average order value can absorb the asset and license costs, and a meaningful share of traffic is on mobile, where camera-based features feel natural. It tends not to pay off when assets cannot keep up with the catalog, when rendering quality is poor enough to mislead (which can increase returns instead of reducing them), or when the feature is buried where few shoppers find it.
Measure it properly. Compare conversion and return rates for shoppers who used try-on against a holdout group that did not see it, by category. Users of try-on are often more engaged shoppers already, so a simple before/after or user/non-user comparison will overstate the effect.
Try-on also connects to wider personalization: a saved photo or size profile is a strong personalization signal, which brings both opportunity and privacy obligations. See our guide to ecommerce personalization for how that data fits into a broader approach.
Privacy: photos and face data
Try-on often processes photos of faces and bodies. Under the GDPR, biometric data processed to uniquely identify a person is special category data, and the UK ICO's biometric guidance stresses that the purpose of processing determines whether that applies. Many try-on tools analyze a face to fit a product rather than to identify someone, but you should confirm exactly what the vendor does with images, how long they are kept, and whether they are used to train models. Some US states also have biometric privacy laws with consent requirements.
This is general information, not legal advice; take advice on your specific markets before launch.
Vendor evaluation questions
Use these questions in demos and in writing. Vague answers are a signal in themselves.
Quality and coverage
- Which of our categories do you support today, and which are on a roadmap?
- Can we test with our own products and a diverse set of skin tones, body types and lighting conditions before signing?
- How do you handle patterns, logos, transparent materials and reflective surfaces?
Assets and operations
- What assets do you need per SKU (standard photos, flat lays, 3D models), and who creates them?
- How long does it take to onboard a new product, and can it be automated from our catalog?
- What happens to try-on when a product goes out of stock or is discontinued?
Integration and performance
- How does the widget load, and what is its effect on page speed and Core Web Vitals?
- Do you integrate with Adobe Commerce or Magento Open Source through an extension, a script, or an API we build against?
- Which events do you send to our analytics, and can we run a holdout test?
Data and privacy
- Where are images processed and stored, and for how long?
- Are shopper images used to train your models? Can that be switched off contractually?
- Do you process biometric identifiers, and what consent flow do you provide?
- Which sub-processors do you use, and in which regions?
Commercials
- Is pricing per SKU, per try-on session, per order or flat?
- What exit terms apply, and do we keep any 3D assets created for us?
Implementation notes for Magento stores
Most try-on tools are added as a script or extension on product pages, with the product identifier passed to the vendor. Load the widget only when the shopper opens it, not on every page view, to protect page speed. Keep the asset mapping (SKU to 3D model or try-on asset) in your catalog so that it survives a vendor change. Send try-on events to your analytics so you can compare cohorts.
For the wider picture of where try-on sits among AI features, see AI in ecommerce.
FAQ
Do I need 3D models for every product?
Only for real-time AR approaches. Generative image try-on usually works from standard product photography, though quality depends on how clean those images are.
Does Google's try-on replace an on-site tool?
No. It works within Google's own shopping surfaces. An on-site tool serves shoppers who are already on your store.
Next steps
Mistri can help you assess which categories are worth testing, prepare product imagery and data through our AI product content services, and integrate and measure a try-on vendor as part of our AI for ecommerce services.
Sources
Google: How to virtually try on clothes using your own image; ICO: Biometric data guidance; Regulation (EU) 2016/679 (GDPR), EUR-Lex.
