Creating a unique hash per face leveraging face recognition techniques

Question

Using face recognition and related techniques I would like to create a unique hash for each face. The idea is that same person's face always returns the same hash, irrespective of lighting conditions, make-up, angle and rotation and similar. Obviously, identical twins might potentially result in the same hash, but that's okay. Also, I am not aiming at perfection, this is not security relevant, other factors such as ability to execute on a client device are more relevant.

Also, I don't need to actually determine to whom the face belongs, it is sufficient to just distinguish it uniquely from every other face and deterministically re-create the same hash each time I see the same face again.

Ideally, I'd like to use some JavaScript library that can run in a browser like face-api.js, pico.js or similar.

However, I am a bit confused how to approach this from a conceptual perspective. For example, if I determine the landmarks of a face, are these landmarks staying the same for each face, so I could use them to create the hash? Also, how should I handle the situation when the landmarks are just slightly different (e.g. due to lighting conditions) so this would result in a different hash due to small deviations? Is there any approach that is robust to such deviations?