SadTalker is a project that produces realistic talking head videos using a single portrait image and audio. It involves training a model to learn 3D motion coefficients for stylized audio-driven single-image talking face animation. The project provides several new modes, such as still, reference and resize modes, for better and custom applications. It also provides offline patches, pre-trained models on Google Drive, and gradio web demos for local and online use. The project is not an official product of Tencent and is only intended for personal, research, and non-commercial use.