Identity Preservation

Identity preservation is a critical requirement for production lip sync: the person in the output video must still look like themselves. AI lip sync models can inadvertently alter facial features, skin tone, or subtle identity markers when modifying the mouth region, especially across different ethnicities, ages, or facial structures.

Techniques for strong identity preservation include using identity-specific loss functions during training, leveraging face recognition embeddings to constrain the output, and architectures that separate identity information from motion information. Sync (sync.so) prioritizes identity preservation as a core quality metric to ensure that dubbed videos look authentic.

Identity Preservation relates to several other concepts in the AI lip sync pipeline: Face Segmentation , and Neural Rendering .