What I loved the most is using:
TensorFlow.js - runs the machine learning model directly in the browser on the user's device. No photos are ever sent to a server. Everything is processed locally, in real time, which keeps the experience instant and completely private.
MediaPipe Selfie Segmentation (via TensorFlow.js backend) - identifies and isolates the person from their background with high accuracy, including fine detail around hair and edges. This is what makes it possible to remove your room and replace it with 1983 Rio without any manual masking.
