Your screen already works. Give it eyes.
A smart player and a depth camera turn a display you already own into a try-on mirror. The same sizing model runs on your mobile site, so a customer gets one answer whether they're on the floor or on the sofa.
Zero-latency AR means the garment overlay is computed on the device in the store, not in a data centre — so the image tracks the body within a single frame instead of lagging behind it. In practice: 30 frames per second or better, under 100 ms motion-to-photon, no round trip to a cloud server. Anything slower reads as broken — the garment slides off the shoulder as the customer turns, and they stop believing what they see.
Cost to convert a screen you own: from $1,290. No new display required.
Same screen. Same wall. Drag to see the difference.
Nothing about the display changes. A player goes behind it and a camera goes above it, and a sign becomes a fitting room.
Drag the handle, or use the arrow keys
Watch what 300 milliseconds looks like.
Both figures are moving identically. The only difference is where the overlay is computed.
On-device — Retailr
Inference and render in the same loop on the player. The garment never separates from the shoulder.
Cloud-rendered
Frames travel to a server and back. On retail broadband the garment visibly drags, and shoppers stop trusting it.
Illustrative animation. Real-world cloud latency varies with connection quality, but the failure mode is consistent — the overlay lags the movement.
What the words actually mean
This category is full of terms used loosely. Here is how Retailr uses them, so you can hold us to it.
Zero-latency AR
An augmented-reality overlay computed entirely on the local device, with no network round trip in the render loop. The practical threshold is 30 fps sustained and under 100 ms motion-to-photon latency — the delay between a person moving and the screen showing that movement with the garment correctly placed.
Above roughly 100 ms the human visual system registers the lag, the garment appears to drag behind the body, and the illusion fails. Cloud-rendered try-on cannot reliably hold that threshold on retail broadband, which is why Retailr runs inference on the player rather than in a data centre.
- 3D body scanning
- Deriving body measurements — shoulder, chest, waist, hip, inseam — from depth-camera frames or, on mobile, from two photographs. No tape measure, no manual entry.
- Size matching AI
- Comparing those measurements against a specific garment's actual spec sheet, not a generic S/M/L table, and returning a recommended size with a confidence level.
- Omnichannel try-on
- One size profile per customer, used identically by the in-store mirror, the kiosk and the mobile web app. A recommendation made in store still applies online a week later.
Why a spec sheet beats a size label
Pick a size and watch what the model compares. It isn't matching "M" to "M" — it's matching real centimetres against the garment's measured dimensions and how much the fabric gives.
This is the part most try-on projects underestimate. Scanning a body is largely solved. Knowing what your garment actually measures, by size, with its real ease and stretch, is the work.
Linen overshirt
Body: 96cm chestIllustrative demo using sample garment data.
From camera frame to size recommendation
Four stages. All but one of them on the device in your store.
Capture
Depth frames at 30 fps in store. On mobile, two photographs taken by the customer.
Body model
Pose and silhouette inference builds a parametric model. Under 400 ms, on the player's GPU.
Garment fit
Matched against the garment's measured spec — dimensions, stretch, intended ease.
Render
Cloth simulation at 30 fps or better, tracking continuously. No network call in this path.
Depth frames at 30 fps, or two phone photos
Parametric model + measurements, on-device, under 400 ms
Matched against the garment's real spec, not an S/M/L table
Overlay at 30 fps, no network call in the loop
The only cloud interaction is catalogue sync and, where the customer opts in, saving their size profile for later. Neither sits in the render path.
Converting a screen you already own
The cheapest route into virtual try-on isn't buying a mirror. It's adding a player and a depth camera to a display that's already mounted and already powered.
| Route | What you add | Hardware | Software | Install |
|---|---|---|---|---|
| Convert existing screen Lowest cost |
Smart player with on-device inference, depth camera, mount | from $1,290 | $15 per screen / mo | $150–$300 |
| Purpose-built mirror | Mirror-finish display with camera and player integrated | $4,500–$7,900 | $15 per screen / mo | $300–$600 |
| Mobile web only | Nothing in store — runs on the customer's own phone | $0 | Store plan | — |
What's in the kit
Smart player with an on-board GPU for local inference, depth camera, mounting bracket and cabling. Ships configured — it pairs to your account on first boot.
What your screen needs
An HDMI input, a spare power outlet, and an orientation that suits the department. 43" or larger works best for full-body; smaller is fine for eyewear and accessories.
What it doesn't need
No mirror glass, no construction, no dedicated circuit, no in-store server. The player handles inference locally; your network only carries catalogue data.
Pricing is indicative and quoted per deployment. Depth camera choice, screen size, orientation and the number of garments to digitise all move the figure. Hardware prices in CAD, current October 2026. A written quote follows a scoping call — we'd rather give you a real number than a brochure one.
One size profile, three places it gets used
In-store mirror
On the sales floor, outside the fitting room. Shows colourways the customer hasn't picked up and calls an associate when a size is needed from stock.
Interactive kiosk
Extends the range beyond what's on the floor. A customer sizes an item the door doesn't stock and has it shipped — a lost sale becomes an order.
Mobile web
No app install. Two photos in the browser produce the same body model and the same recommendation the in-store mirror would give.
The profile belongs to the customer, not the channel. A size measured at the kiosk on Saturday is the size the website uses on Wednesday — if they chose to save it. Opt-in, and deletable by them at any time.
What the published data actually shows
Virtual try-on marketing is full of unsourced percentages. Here is what we can point at a primary source for, and what we can't.
Reduction in size-related returns over a six-month deployment at swimwear brand TA3, bringing its return rate to 9%.
Source: 3DLOOK YourFit case study, reported in Textile World, October 2022. Vendor-supplied figure for a competing product, single brand, not independently audited.
Share of shoppers using that virtual fitting room who went on to purchase. No pre-deployment baseline was published, so this is not a measured lift.
Source: same 3DLOOK case study. We include the caveat because the number is widely quoted without it.
What we won't tell you: that virtual try-on reliably cuts returns by a specific percentage across all apparel. The commonly circulated "20–30% return reduction" figures trace back to vendor marketing rather than published research, and effect sizes vary enormously by category — eyewear and footwear behave nothing like outerwear. Anyone quoting you a precise number before seeing your return data is guessing.
What we'd measure in your pilot instead
Rather than promising a figure, we instrument the things that actually move and report them monthly:
- Size-related return rate, split from all other return reasons — the only return metric try-on can plausibly affect
- Try-on to purchase rate, against a baseline measured in the four weeks before go-live
- Size-swap rate in store, which usually falls before returns do and gives you an earlier signal
- Fit confidence by style, which feeds back into merchandising and buying
Baseline measurement before go-live is what makes any of these numbers meaningful. We'll push for it even when it delays launch by a month.
Body scanning and privacy
Measuring bodies is sensitive, and several jurisdictions regulate biometric data specifically. Our position:
- No cameras in fitting rooms. The mirror sits on the sales floor. Fitting-room data, where we collect it at all, comes from RFID garment tags.
- Processing is local. Depth frames are processed on the player and discarded. No video or imagery leaves the store.
- Measurements, not images. What can be retained is a numeric measurement set, and only if the customer opts in to save a profile.
- Anonymous by default. Walk-up use stores nothing. A profile exists only when a customer deliberately creates one.
- Deletable on request, by the customer directly, without going through you.
- Visible notice at the mirror, which several jurisdictions require and all of them make sensible.
Body measurement may constitute biometric data under Illinois BIPA, Texas CUBI, Quebec's Law 25 and GDPR, among others. Requirements differ on consent, notice and retention. We'll document exactly what a configuration collects; your legal team should confirm it meets your obligations in each market.
Frequently asked questions
What is zero-latency AR?
An augmented-reality overlay computed entirely on the local device with no network round trip in the render loop, sustaining 30 frames per second or better with under 100 milliseconds of motion-to-photon latency. Above roughly 100 ms the delay becomes perceptible, the garment appears to lag behind the body, and the effect stops being convincing.
Can I use my existing screen for virtual try-on?
Yes, in most cases. Adding a smart player and a depth camera to a display you already own costs from $1,290 in hardware plus $150–$300 install, compared with $4,500–$7,900 for a purpose-built mirror unit. The screen needs an HDMI input, a power outlet and ideally a 43-inch or larger panel for full-body garments.
Does virtual try-on reduce returns?
Published case studies report meaningful reductions in size-related returns — one six-month deployment at swimwear brand TA3 reported a 47% drop — but these are vendor-supplied figures for single brands rather than independent research, and effect sizes vary widely by category.
Eyewear and footwear behave very differently from outerwear. We measure size-related returns against a pre-launch baseline rather than quoting a percentage before seeing your data.
How accurate is the size recommendation?
Accuracy depends far more on your garment data than on the scanning. Matching a body model against a garment's measured spec sheet — actual dimensions, fabric stretch, intended ease — produces a usable recommendation. Matching against a generic S/M/L table does not, regardless of how good the scan is. Digitising garment specs is the real work in any deployment.
Do customers need to install an app?
No. In store, the mirror and kiosk need nothing from the customer. On mobile, try-on runs in the browser from two photographs. An app install is the fastest way to lose most of your potential users.
Is body scan data stored?
Only if the customer opts in to save a size profile, and then only as a numeric measurement set — never imagery. Depth frames are processed on the player and discarded. Walk-up use stores nothing at all. Customers can delete a saved profile themselves.
What garment data do you need from us?
Per style: measured dimensions by size, fabric composition and stretch, and a flat or on-model image. Brands with a 3D design pipeline such as CLO or Browzwear can supply assets directly. Brands without one will need a digitisation pass, which is the main project cost to scope.
Send us a photo of the screen you want to convert.
We'll tell you whether it'll work, what the kit costs for your size and orientation, and what digitising your garment data would involve. Thirty minutes, written quote after.