LatentSync
LatentSync is an end-to-end audio-conditioned lip-sync framework that generates perfectly synchronized talking-head videos from any source video and target audio.
ByteDance 
WaveSpeed Audio-conditioned lip sync via latent diffusion
LatentSync is an end-to-end lip-synchronization model developed by ByteDance that takes a source talking-head video and a target audio track as inputs and produces a new video with mouth movements precisely aligned to the provided speech. It is built on audio-conditioned latent diffusion, meaning it operates directly in the latent space rather than relying on 3D meshes or 2D facial landmarks. The model supports multiple languages and accents, works with diverse speakers and recording conditions, and handles both real people and stylized characters such as anime.
A key technical feature of LatentSync is Temporal REPresentation Alignment (TREPA), a technique designed to reduce flicker, jitter, and frame-to-frame artifacts, keeping head pose and lip motion stable across longer sequences. The model is recommended for use with high-resolution source videos — 720p, 1080p, or 4K — paired with clean audio recordings. It is well suited for workflows involving lip-syncing, audio dubbing, digital human creation, and video-audio alignment.
What LatentSync supports
Lip Synchronization
Aligns mouth movements in a source video to a target audio track end-to-end, without relying on 3D meshes or 2D facial landmarks.
Audio Input Processing
Accepts an audio URL as a conditioning input, supporting multiple languages and accents across diverse speakers and recording conditions.
Video Input Processing
Takes a source talking-head video URL as input and preserves the subject's identity, pose, background, and scene structure in the output.
Temporal Consistency (TREPA)
Uses Temporal REPresentation Alignment to reduce flicker, jitter, and frame-to-frame artifacts, keeping motion stable over long sequences.
High-Resolution Output
Supports source videos at 720p, 1080p, or 4K resolution, delivering sharp facial detail for both real people and stylized characters.
Ready to build with LatentSync?
Get Started FreeCommon questions about LatentSync
What inputs does LatentSync require?
LatentSync requires two inputs: a video URL pointing to a source talking-head video and an audio URL pointing to the target audio track you want the subject to appear to speak.
What video resolution is recommended for best results?
The model documentation recommends using source videos at 720p, 1080p, or 4K resolution, paired with clean, dry audio recordings for optimal output quality.
Does LatentSync support languages other than English?
Yes, LatentSync is described as multilingual and supports multiple languages and accents, adapting to diverse speakers and recording conditions.
What is the context window for this model?
The model has a context window of 50,000 tokens as listed in its metadata.
Does LatentSync use facial landmark detection or 3D meshes internally?
No. LatentSync operates via audio-conditioned latent diffusion and does not rely on 3D meshes or 2D facial landmarks to generate lip-synced output.
What is the training data cutoff for LatentSync?
According to the model metadata, the training date is listed as August 2025.
Parameters & options
Audio to be synchronized.
Video to be synchronized.
Explore similar models
ByteDance Omni Human 1.5
ByteDance's OmniHuman 1.5 transforms static images into emotionally expressive, audio-driven digital humans using cognitive simulation and multimodal AI.
ByteDance Seedream 5.0 Lite
ByteDance's lightweight Seedream 5.0 image editing model, delivering fast, high-fidelity transformations with exceptional face preservation and multi-reference support.
ByteDance DreamActor V2
ByteDance DreamActor V2 is a motion transfer model that animates any character — human, cartoon, or creature — by applying movements from a reference video to a static image.
MeiGen Infinitetalk
An audio-driven avatar generation model that transforms a single photo or silent video into a lifelike talking or singing video with precise lip sync, natural body movement, and support for videos up to 10 minutes long.
Kling AI Avatar Standard
Kling AI Avatar transforms a single portrait photo into a natural talking-head video driven by any audio track, with precise lip-sync and stable identity preservation.
HeyGen HeyGen Video Translate
HeyGen Video Translate is an AI-powered tool that automatically translates and lip-syncs videos into multiple languages with a user-friendly interface.
Start building with LatentSync
No API keys required. Create AI-powered workflows with LatentSync in minutes — free.