LatentSync

About LatentSync

LatentSync is a web tool for synchronizing lip movements in video with a new audio track. Users supply an audio file and a video, and the service generates a lip-synced result. It is aimed at film and TV dubbing teams, creators repurposing short-form content, virtual avatar builders and educators localizing training material.

The approach uses audio-conditioned latent diffusion models rather than intermediate motion representations, and the site says it works across multiple languages, including optimized support for Chinese content. Output is described at 512x512 resolution with attention to temporal consistency, so mouth movements stay smooth across frames.

The site also lists an open source ecosystem with inference code and checkpoints, plus a Gradio app and a command line interface for running it locally. Paid one-time credit packs are listed for the hosted service, with no watermark and commercial use included.