Generates realistic talking face animations from a portrait image and audio using the CVPR 2025 Sonic model