Virtual try-on (VTON) is a class of image-generation technology that renders a specific garment onto a photograph of a specific person, producing a still image of that person wearing that garment. In fashion e-commerce it answers the one question a product page has never been able to answer: not what does this look like, but what does this look like on me.
The term covers several quite different approaches. What they share is the output: a photorealistic still image of a real person in a garment they do not own. What separates them is how they get there, and that difference decides what a retailer has to supply, what it costs to run across a catalogue, and how the result looks.
An image-based try-on pipeline runs in four stages. Each one is a distinct computer-vision problem, and output quality is set by the weakest of them.
Viewrobe runs this pipeline with proprietary generative AI models, trained to handle 17 fabric types, and outputs at up to 4K resolution in PNG, WebP or JPEG.
Almost every visible failure in virtual try-on comes down to one of two things, and it is worth knowing which you are looking at when you evaluate a system.
Identity. The person in the output has to be recognisably the person in the input. Their face, their build, their skin tone and their hair should be unchanged. A system that quietly slims the body or regularises the face has not shown the shopper themselves in the garment, it has shown them a model, and the whole purpose is lost.
Fidelity to the product. The garment in the output has to be the garment that will arrive in the box. The colour has to match, the print has to be positioned and scaled correctly, the neckline and sleeve length have to be the real ones. A try-on that renders a flattering generic version of the product creates exactly the expectation gap that drives returns.
| Approach | What the retailer supplies | What the shopper sees | Cost at catalogue scale |
|---|---|---|---|
| On-model photoshoot | Samples, models, studio, crew, retouching | A model wearing the garment, not themselves | Per style and per colourway; weeks of lead time |
| AR try-on filter | A tuned 3D asset per style | An overlay tracking their live camera, in motion | Asset build per style; needs camera permission |
| 3D / CGI garment simulation | A digital pattern and garment model per style | A rendered avatar, or themselves if scanned | Highest build cost; very high accuracy once built |
| Size recommendation | Measurements and returns data | A suggested size, no image | Low; answers fit, not appearance |
| Image-based VTON | Existing product photography | Themselves in the garment, as a still image | Per render; no asset build per style |
These are not mutually exclusive. Size recommendation and virtual try-on in particular solve adjacent halves of the same problem: one tells a shopper which size to order, the other shows them what it will look like.
Most demonstrations use flattering inputs: a straight-on photo of a slim person in tight clothing, against a plain wall, holding still. Those conditions hide nearly every weakness. If you are assessing a system, test it the way your customers will use it.
Virtual try-on earns its place next to the gallery, as an option on the product the shopper is already looking at, not as a separate experience they have to go and find. The shopper is deciding right there, and that is where the doubt is.
Viewrobe attaches to an existing product page with a single asynchronous script tag, on any e-commerce platform, with no plugin to maintain and no replatforming. Headless storefronts can call the REST API directly. The practical detail is in the integration overview.
It is not a fit or sizing engine. It shows a shopper what a garment looks like on them; it does not measure them, and it will not tell them whether the medium or the large is correct. Pair it with a size tool if that is the problem you are solving.
It does not invent what the photograph never showed. If the source image does not reveal how a garment is constructed at the back, no system can render the back of it faithfully.
And it is not a substitute for a physical fitting where the stakes justify one. In bridal and in bespoke tailoring, a try-on is the step that gets a shopper to the appointment with a shortlist, not the step that replaces the appointment.
Written by Viewrobe. Published 6 October 2026, last reviewed 7 October 2026.
No. An augmented-reality filter tracks a live camera feed and draws a garment over it in motion, which is why AR output usually looks like an overlay rather than a photograph. Virtual try-on generates a still image: one photo of the shopper, one garment, one rendered result that can be shown on a product page, saved, or sent in an email. The two solve different problems, and still imagery is what a product page actually needs.
Not with an image-based approach. Viewrobe works from the flat-lay or on-mannequin photography a retailer already has, so there is no 3D asset to commission for each style. Systems built on 3D garment simulation need a garment model per style, which is accurate but expensive to produce at catalogue scale.
No. The shopper uploads a single photo. There is no app to download, no account to create, and no 3D body scan.
With Viewrobe it is used only to generate the try-on the shopper asked for, it is permanently deleted within 24 hours, and it is never used to train AI models. The full detail is on the security and data handling page.
Garments whose shape is legible in the source photograph. Tops, dresses, outerwear, suiting and bridal all render well because the silhouette and the fabric behaviour are visible in a flat-lay or mannequin shot. Viewrobe's models cover 17 fabric types. Items whose appearance depends on how they are worn rather than how they are cut, such as loosely draped scarves, are harder for any system of this kind.
It replaces a particular kind of shoot: the repetitive on-model work of putting every colourway and every size of an existing style in front of a camera. It does not replace a campaign shoot, where the point is art direction rather than coverage.
