With the computational efficiency of Gaussian splatters, this could be ground-breaking for photorealistic avatars, possible driven by LLMs and generative audio.