Data to be translated: Meta AI research team has successfully addressed the challenge of high-fidelity relighting of dynamic 3D avatars and introduced a light-reconfigurable Gaussian encoding avatar. This method employs a geometry model based on 3D Gaussian functions and a learnable radiance transfer appearance model, capable of capturing sub-millimeter level details in dynamic facial sequences more clearly. Additionally, this method enables dynamic generation and relighting, allowing the avatar to be flexibly controlled based on expressions, gaze, viewing angles, and lighting, thereby creating dynamic, interactive content and achieving seamless driven avatars.