Hacker News Reader

The experience is as if you're looking directly at a person sitting across from you. There is no camera offset.

The Q is "done how"

IIRC depth sensors, you're not actually looking at the original image from a camera, they recreate the person as a 3D model.