LivePortrait is a human animation generation model based on an implicit keypoint framework. It synthesizes realistic videos by using a single source image as a reference for appearance and acquiring actions (e.g., facial expressions and head poses) from driving videos, audio, text, or generation. The model effectively balances computational efficiency and controllability, and significantly improves the generation quality and generalization capability through expanded training data, a hybrid image-video training strategy, upgraded network architecture, and the design of better motion transfer and optimization objectives.