How Virtual Try-On Actually Works
A body photo plus a garment photo, and an AI model that has seen enough clothes to know how they fall. That is the whole trick — and it is most of what you need.
Virtual try-on used to be a gimmick. A flat image of a shirt pasted onto a flat image of a person, with cartoon seams and shoulders in the wrong place. Nobody bought anything from it, and everyone forgot it existed.
The current generation is different. It is good enough that you stop thinking about the technology and start thinking about the clothes.
What happens when you press the button
Two inputs go in. A photo of you, taken in neutral light. A photo of a garment — usually a clean product image from a retailer. The system sends both to an image model that has been trained on millions of pairs of "body, garment → body wearing garment".
The model does not paste. It re-renders. It infers how the fabric would fall across your specific silhouette, how the collar sits on your shoulders, where the hem lands on your frame.
The output is a new image — you, wearing that piece — produced in about four to seven seconds.
Why it works now and not before
Three things shifted.
Larger diffusion models. The same family of image models that draws photorealistic scenes from text can also edit one image conditioned on another. Given "this body" and "this garment," a modern model can generate a plausible result without needing a 3D mesh of your body or a 3D scan of the clothes.
Garment-aware training. Research pipelines specifically for fashion — FASHN, Kling, Google's Gemini — are trained on paired data where the same garment appears on different bodies. The model learns how a wool coat drapes versus how a linen shirt drapes, not just how pixels should move.
Speed. A few years ago this kind of render took minutes. Now it happens in the time it takes to tap a second tile.
Where it still struggles
Virtual try-on is not fitting. The output tells you how a garment looks on you, not how it fits you.
Cuts that depend on body mechanics — deep v-necks, high-rise jeans, tailored blazers — still require a real dressing room. Fine details like buttons, zips, and seams sometimes blur. Patterns with repeating logos can warp.
It is best at loose silhouettes, solid colours, and everyday pieces. Those are also most of what people actually buy online.
The output tells you how a garment looks on you, not how it fits you. Those are different questions.
What it is useful for
Three things, in order of impact.
Killing bad ideas fast. You were going to buy a canary-yellow overshirt because it looked great on the model. You render it on yourself, and the colour fights your skin tone. The render saved you a return.
Testing silhouettes you would not risk. Cropped, oversized, pleated, boxy — things you suspect but never try. Seeing yourself in them gives permission, or removes the curiosity cheaply.
Building a shortlist. Before a store visit, render ten candidates. Walk in with three. Save the decision fatigue for the fitting room.
The privacy question
Every good try-on tool sends your body photo to a cloud model. That is unavoidable today — the models are too large to run on device. What varies is what happens after.
Vael stores your body photo in private per-user storage and never uses it to train any model. Try-on results are cached for seven days so repeat try-ons of the same garment return instantly, then deleted. You can wipe everything from the account screen.
Check any try-on product's privacy policy before uploading. If the answer is not clear in one paragraph, do not use it.
Try it
If you want to see what a modern render looks like — on your own body, with any garment from any site — Vael's free demo runs in the browser. No signup. The first render tells you more than reading about it will.