It allows you to set up a 3D scene, then take what appears in it’s camera and put it into a 2D sprite texture. A lot like if you had an actual camera monitoring a room and you are watching it on a television.
For one example, it can be used for things such as taking a 3D character and projecting them into a 2D game.
Do you know what kind of impact this has on performance, compared to rendering the sprites as planes/billboards directly into the 3d scene?