Fixing faces by training on closeups

Google Photos recognize people between photos. What if Topaz video had a similar feature? It would look through the entire video to find the part there the person is closest to the camera and use that information to restore the other parts. Some manual pairing may be reqired. “Are these two faces the same person?”. The LoRA should ideally be securely stored and only available to that user.