01
Gesture
Computer vision · 1080p/4K camera
The mirror shows a live feed. A wave opens the garment picker — nothing to touch, which matters in high-traffic or hygiene-sensitive stores.
Smart Mirror · Virtual Try-On
Try it on without taking anything off.
A shopper stands in front of the mirror, picks a garment, and sees a photorealistic image of themselves wearing it in a few seconds. The AI runs on the box in the store — no cloud, no upload, no internet required.
Runs offline on an NVIDIA RTX edge box · 2–8 seconds per try-on
Rendering — 2.4s
Nothing leaves the store
Four steps, start to finish.
01
Stand
The shopper steps in front of the mirror. The integrated camera picks them up.
02
Select
They browse the catalogue by touch, by gesture, or from their own phone.
03
Render
Retailr VITON generates a photorealistic image of them in that garment — 2 to 8 seconds.
04
Decide
The result fills the screen. They try the next colour, save it, or buy.
Everything happens inside the mirror.
The camera feed, the body mapping and the render all run on a local GPU. No customer image is uploaded anywhere, and the mirror keeps working when the store's internet doesn't.
No internet needed
Full try-on operation with the network down. Sync catches up later.
Images stay local
Customer photos are processed on the box and not sent to a cloud service.
Fast enough to browse
2–3 seconds on the premium tier, so shoppers try five things, not one.
Unlimited catalogue
Garment count is limited by disk, not by a per-item licence.
One GPU box. Any screen you already own.
The intelligence isn't in the glass — it's in the NVIDIA box behind it. Plug it into a display you already have and that screen becomes a try-on mirror.
Any display
Existing signage screens, portrait totems, video walls or a standard 4K panel. HDMI in, try-on out.
Mirror or not
Run it behind two-way glass for a true mirror, or on a plain screen as a try-on kiosk.
Retrofit a store
Keep the fixtures and the mounts you've already paid for. Only the compute is new.
Scales by GPU
Render speed follows the card — RTX 3060 to 4090 — so you buy performance, not a new mirror.
Audience-aware content
The screen sells before anyone touches it.
Presence and vision sensors tell the screen whether anyone is there and roughly who they are, so it can lead with something relevant instead of a generic loop.
Empty room — plays your signage, promotions and brand content
Someone within 2m — switches to mirror mode with the picker ready
Audience signals — on-device vision estimates category (menswear, womenswear, age bracket) and leads with that range
Dwell time — how long people stop and watch, measured per screen
Nothing recorded — signals are read on the box and discarded; no faces stored or sent anywhere
Audience estimation is optional and configurable — you choose which signals drive content, or turn it off entirely.
Three ways it wakes up.
The mirror plays signage until someone approaches. How it hands over to try-on is your choice — and it works with touch or entirely without.
02
QR & session link
Mobile integration
The shopper scans a code on the mirror. Their browser session links across, loading the wishlist and recently viewed items from your e-commerce site.
03
Proximity
Presence sensor + AI
Default state is digital signage. When someone comes within two metres it switches to mirror mode with relevant suggestions ready.
Why it looks real.
A cut-out pasted on a photo fools nobody. Retailr VITON maps the garment to the body and re-renders it with the fabric behaving the way fabric behaves.
Semantic correspondence learning — maps garments to body structure rather than overlaying them
Latent diffusion — generates real wrinkles, folds and shadows
Detail preservation — patterns, textures and colours stay accurate
Natural draping — simulates how fabric falls on different body types
High-resolution output — detailed enough for a 55-inch screen
Connected to the rest of your stack.
The mirror renders locally, but it isn't a data island. When it's online it talks to the systems you already run.
SKU mapping — garment IDs match your existing SKUs, with live stock checked through the POS API
Endless aisle — out of stock in-store becomes an order shipped to their home
CRM push — items tried and session length flow to your CRM for abandoned try-on follow-ups
Shared assets — the same garment data drives the in-store mirror and web try-on
And it still does the fitting-room job.
RFID picks up what the shopper carried in. The mirror shows sizes, colours, fabric and care, and calls a staff member without anyone getting half-dressed to go looking.
Automatic garment detection by RFID tag
Request another size or colour from the mirror
Alert lands on the nearest associate's device with the room number
Styling suggestions based on what's already in the room
What you learn from it.
Try-on data answers the question a POS never can: what did people want, and not buy?
Tried vs bought
Engagement-to-conversion per garment, so you can see what looks good on the rail and fails in the mirror.
Top tried items
What draws attention, which sizes get picked, which pairings repeat.
Dead stock signals
Items tried often but never bought point at fit or price, not demand.
Brand dashboards
Custom BI on a dedicated cloud layer, kept separate so mirror performance never suffers.
Advanced BI dashboards are a separate module with their own cloud instance and development scope — deliberately, so heavy reporting never slows the render.
Questions we get asked.
Does it really work without internet?
Yes. Rendering, body mapping and the garment catalogue are all local. Connectivity is only used for optional syncing with your POS, CRM or e-commerce catalogue.
Where do customer images go?
They're processed on the local machine. Nothing is sent to a cloud rendering service. We'll walk through exactly what is and isn't retained during the demo, so you can post the right in-store notice.
How long does one try-on take?
Between 2 and 8 seconds, depending on the hardware tier. The flagship build renders in 2–3 seconds.
How many garments can we load?
The catalogue is limited by local storage, not by licence. Each garment is prepared once and reusable indefinitely.
Does it need touch?
No. It works in touch and touchless modes — gesture activation and proximity switching both work without contact.
What about accessories rather than clothing?
The same pipeline handles worn items like eyewear and accessories; talk to us about your specific category and we'll show you a sample render.
Send us one garment. We'll send back the try-on.
The fastest way to judge this is to see your own product rendered. Book a demo and we'll do exactly that.