Where does the sound on these AI videos come from, is that AI generated in the same pass, a second model, or is it added by a human after the fact?