would love to see some video chat providers work on an open source implementation

TL;DR detect face, mask notface, then render face over the mask