Fascinating feature! Is it known how they do it?

Is it just an image transformation or a full blown AI model using Gaussian Splats or something along those lines?