What is On-device camera processing?
On-device camera processing means the analysis of a camera feed happens entirely in the device browser or app, without sending video frames to a server. The video never leaves the device.
The longer answer
On-device processing runs computer-vision models (face landmarks, hand tracking, segmentation) locally using WebAssembly and WebGL or WebGPU. Only the model output — a handful of coordinates or a label — is used by the application. The raw camera frames stay in browser memory and are discarded each frame.
Why it matters
Three reasons. Privacy: the shopper video is never seen by your servers, by a third party, or by a person. Latency: round-trip to a server would add 100 to 300 milliseconds per frame, breaking real-time interaction. Compliance: regulations around biometric data are simpler when no biometric data is collected.
How it works
The browser exposes the camera via getUserMedia, the model runs via the MediaPipe Tasks Web API or TensorFlow.js, and the rendering engine reads landmark coordinates each frame. The video MediaStream stays within the browser sandbox; nothing is uploaded.
Where this fits in WearRoom
WearRoom processes 100 percent of the camera feed on the shopper device. We never see, store, or relay video. Our servers only ingest anonymized funnel events — try-on opened, completed, added to cart. This is the architecture from day one, not a setting.
Questions
How can I verify that no video is uploaded?
Open the browser developer tools, switch to the Network tab, and watch traffic during a try-on session. You will see model assets and analytics pings, but no video or image uploads.
Is this affected by biometric regulations?
When no biometric data is collected or stored, biometric-data regulations like Illinois BIPA generally do not apply. Always consult counsel for your specific situation.