you would have to train a own model with pictures of the face you want.

after that you can merge it to the other model, but its very time consumption