Back when I was learning how 3d vtuber models worked. I took my WIP omni-man model and partially rigged it. Had this working with a Pixel 5 as the camera source.

Tested out a Blender addon that allowed me to use a Google Pixel and a vtubing app for face tracking. In theory you could just greenscreen this out and use Blender directly rather than having to use vtubestudio or vrchat.