Latentsync

LatentSync is a cutting-edge, open-source lip-synchronization framework powered by Audio-Conditioned Latent Diffusion Models. By integrating Whisper audio embeddings with advanced temporal alignment (TREPA), it transforms arbitrary audio and video inputs into photorealistic, high-resolution (512x512) talking head videos. Designed for creators, researchers, and developers, LatentSync eliminates the "blurry mouth" artifacts of legacy models, delivering cinema-grade synchronization with superior temporal stability and visual fidelity.