Openspline: Host Realtime Talking Avatars Locally at 90+ FPS

A developer has introduced Openspline, an open-source avatar layer that allows users to generate real-time, lip-synced talking avatars from a single image locally on a GPU. The lightweight model achieves over 90 frames per second on an NVIDIA RTX 4090 graphics card. This project enables developers to easily integrate visual talking heads into existing voice AI agents using just a few lines of code. By running locally on commercial hardware with high performance, it lowers the barrier for interactive, real-time generative AI applications. Openspline integrates with existing voice AI agents in only 5 lines of code and provides both a PyPI package and a web demo. Although still in a very early stage of development, the project source code is available on GitHub.