Browser ML · ONNX Runtime · Privacy-first
BG Remover AI — Browser-Native ML for Background Removal
BG Remover AI is a single-page web app that strips backgrounds from photos using an ONNX segmentation model — running entirely in the visitor's browser. No upload. No server roundtrip. No quiet retention for training. I built it as a worked example to point at when teams ask whether browser-native inference is real (it is).
Why I built it
Most "free background remover" sites on the open web take your photo, run it on a server they control, and retain it for training or marketing. That trade-off is unnecessary in 2026 — modern ONNX Runtime Web is fast enough to run a real segmentation model in WebAssembly on commodity hardware.
I wanted a worked example I could point engineers at when the question came up. "Yes, you can ship the model to the client. Here's what it looks like."
How it works
The app loads an ONNX model into ONNX Runtime Web, lazy-fetched on first use. Image preprocessing happens in a Web Worker so the UI thread doesn't stutter. The model emits a mask, which is composited back over the source. The whole flow is client-side; the server's only job is to serve static assets.
Built with React, TypeScript, and Vite. Streams the model in chunks so first paint stays fast even on slow connections.
What I take from it
Browser-native inference is production-ready for the right model sizes. For privacy-sensitive AI features in web products, shipping the model to the client is now a real architectural option — not a research curiosity. That changes the architecture of a lot of AI-enabled web development going forward, and I suspect WebGPU will widen the window over the next 12–18 months.