FaceID is more complicated: it uses an infrared camera and projects an array of dots on your face so the problem wouldn't just be generating a realistic video of a face but more along the lines of constructing a mask which would have similar 3D structure including how it reflects infrared light.

Sounds likely. When you train FaceID you are filmed. What's to stop someone from using film of you giving a speech?

At best the biometric locks are like locks on your house. Stops most people but not someone really determined.

It uses an IR camera as a proximity sensor. A video isn't good enough. You would need AI to construct an accurate 3D model of the face based on it.

Yes that's what I meant, someone could construct the necessary biometrics from a video and some ML.