This demo showcases adding virtual backgrounds to a live video stream directly in the browser. It implements and compares three different ML-based person segmentation models: BodyPix (TensorFlow.js), MediaPipe Meet Segmentation (TFLite/WASM), and ML Kit Selfie Segmentation. The project supports multiple backends including Canvas 2D with CPU processing, WebGL 2 shaders, and WebAssembly for inference acceleration.