Why your selfie is the future of online shopping (and why that's actually fine)

For years, "virtual try-on" was a gimmick. You'd upload a photo, the app would slap a flat image of a dress onto a vague approximation of your body, and the result looked like a ransom note. Nobody made a real purchase decision based on it. It was a novelty button you pressed once and never again.

That changed quietly over the last two years, and the change matters more than the hype around it suggests. Here's an honest look at where virtual try-on actually is in 2026 — and how to think about the privacy question that comes with it.

What actually changed

The breakthrough wasn't one thing. It was three, arriving at roughly the same time.

Diffusion models learned to dress people. The same family of AI models behind the image generators everyone played with in 2023 turned out to be very good at a narrower, more useful task: taking a photo of a person and a photo of a garment, and producing a realistic image of that person wearing that garment. Models like IDM-VTON crossed a threshold where the output preserves your actual proportions, the fabric drapes plausibly, and — critically — your face still looks like you.

Face restoration got good. Early try-on melted faces. The current generation runs a face-restoration pass that keeps identity intact, so the person in the rendered image is recognizably you, not a uncanny stranger wearing your haircut.

The inputs got smarter. The systems learned what they need: a clear, full-body photo against an uncluttered background, in decent light. Give them that, and the result is genuinely useful. Give them a blurry mirror selfie taken at midnight, and you get a blurry midnight render. Garbage in, garbage out — but good in now produces good out, which wasn't true before.

What it can do today

Realistically, here's the value:

• It answers "does this silhouette work on me?" This is the question that drives most returns. A wide-leg trouser, a cropped jacket, a midi skirt — these read completely differently on different bodies, and a flat product photo on a model tells you almost nothing. Seeing it on your own frame is the actual signal.
• It shows you colour against your skin. A shade that looks gorgeous on the model can wash you out entirely. Try-on closes that gap.
• It reduces the "order three sizes, return two" cycle. Not perfectly — fit prediction is still unsolved — but enough to cut the obvious mistakes.

What it still can't do

In the interest of not overselling:

• It can't predict your exact size in a new brand. That's a different, harder problem involving proprietary fit data most brands won't share. Try-on shows you the look, not the fit at size M vs L.
• It can't judge fabric quality. It can show that something is "knitwear." It can't tell you the knit is scratchy acrylic versus buttery merino. Product copy still matters.
• It needs a decent photo. The single biggest determinant of output quality is input quality. A good full-body shot in even light is worth more than any amount of model cleverness.

The privacy question, addressed honestly

Handing a photo of yourself to an app is a reasonable thing to be cautious about. So here's how a responsible product should handle it — and how Mirrelle does.

Your selfie shouldn't be stored as a free-floating image in a system that advertisers can touch. In Mirrelle's case, the photo is processed into a numeric twin — a mathematical representation of facial landmarks, skin tone, and body proportions — used to personalize styling and power try-on. That representation is encrypted at rest. It isn't sold. It isn't shared with advertisers. And it's deletable in one tap, along with the original photo, in line with GDPR and CCPA.

The right test for any try-on app is simple: Can I delete everything? Do they sell my data? Is it encrypted? If the answers are yes, no, yes, the trade — a photo for a genuinely better shopping decision — is a fair one. If a company can't answer those three questions clearly, that tells you something.

Where this goes

The likely future isn't "everyone uses try-on for everything." It's that "show me this on me" becomes a default expectation — the way "show me a photo of the actual room" became standard in real-estate listings around 2010. Some people will use it constantly. Others will skip it and just take the recommendation. Both are fine. The technology finally works well enough to be a tool rather than a toy, and a tool you can ignore is a tool done right.

That's the bar Mirrelle builds to: try-on is there when you want it, skippable when you don't, and built so your photo stays yours.