AI try-on vs AR filters vs the changing-room mirror

These three things get lumped together as 'virtual try-on' and they have almost nothing in common. They use different technology, they fail in different ways, and they answer genuinely different questions. Picking the wrong one for the question you have is why people conclude that try-on 'doesn't work'.

AR filters: real-time, flat, tracked

An AR try-on runs on your live camera feed. It detects landmarks — the corners of your eyes, the bridge of your nose, the outline of a wrist — and draws a 3D or 2D asset positioned relative to them, thirty times a second.

This is genuinely excellent for a specific class of product: rigid objects that sit in a fixed relationship to a body landmark. Glasses on a nose bridge. A watch on a wrist. Lipstick on a mouth. Earrings. The object does not deform, its position is fully determined by a landmark the tracker can find, and being live is a real advantage because you can turn your head and see the side.

It is equally, structurally bad at clothing. Cloth is not rigid. It hangs, folds, catches light, and behaves according to its own weight and the shape of the body underneath — none of which a tracked overlay models. So an AR garment slides against you as you move, sits at a constant distance from the camera regardless of your build, and carries the lighting of whoever photographed the asset.

AR is a good answer to 'where does this object sit on me'. It has never been an answer to 'how does this fall on me'.

AI try-on: still, generated, photoreal

Generative try-on does not track anything. It produces a new photograph from scratch, conditioned on your body and on the garment simultaneously, so the fabric and the folds and the shadows are generated together under one lighting model rather than layered on top of one another.

The cost is that it is not live. A generation takes something like thirty to sixty seconds, and what comes back is a single still image rather than something you can turn around in. The gain is everything that makes cloth look like cloth. For a full account of the mechanism, see how AI virtual try-on actually works.

The still-image constraint matters less than it sounds, because a still is what you were going to judge from anyway. You are comparing it against a product photograph, not against standing in front of a mirror.

The changing room: everything AI can't do

Worth being honest about. A physical fitting tells you the things that are not visual at all: how heavy a coat is on your shoulders, whether a waistband digs in when you sit, whether a fabric is scratchy, whether a lining slides or clings, whether a garment is warm.

No image-based method touches any of that, and no amount of model improvement will. What has changed is that the visual questions — the majority of them, and the ones that cause most returns — no longer require the trip.

Side by side

Choosing between them

• Buying glasses or a watch? AR. It is the right tool and it is instant.
• Deciding whether a coat suits you? AI try-on. Volume and drape are the entire question and AR cannot represent either.
• Wondering where a dress hem lands on your height? AI try-on, with a full-length photo — the reason that photo matters so much.
• Wondering whether a wool coat is too heavy to wear daily? Nothing here helps. Go and pick one up.
• Deciding between two colours of the same piece? Either, though AI try-on shows you the colour against your actual skin under coherent lighting, which AR does not.

The mirror-selfie baseline

One more comparison nobody makes: the thing most people actually do, which is buy the item, photograph themselves in it, and ask a group chat. It works, and it is worth noticing what it costs — the purchase has already happened, the packaging is open, and returning it is now a task with a deadline.

The interesting property of AI try-on is not that it beats a mirror. It is that it moves the group chat before the purchase instead of after it. You can send four friends a picture of you in the dress and get an honest verdict while the decision is still free.

Try the thing on before the box arrives, not after.

Get the app