Begs the question given this is diffusion based how much of the "ipadapter/faceid/controlnet" tech can be brought over, what would a audio-faceid or audio-ipadapter look like for something like this.