bytedance/LatentSync

Taming Stable Diffusion for Lip Sync!

51
/ 100
Established

Audio-conditioned latent diffusion operating directly in Stable Diffusion's latent space, using Whisper-extracted audio embeddings injected via U-Net cross-attention layers. Trains with TREPA, LPIPS, and SyncNet losses in pixel space while working on compressed latents, avoiding intermediate motion representations. Supports multi-resolution training (256×256 to 512×512) with configurable efficiency modes, ranging from 20–55 GB VRAM depending on stage and resolution.

5,506 stars. No commits in the last 6 months.

Stale 6m No Package No Dependents
Maintenance 2 / 25
Adoption 10 / 25
Maturity 16 / 25
Community 23 / 25

How are scores calculated?

Stars

5,506

Forks

899

Language

Python

License

Apache-2.0

Last pushed

Jun 20, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/bytedance/LatentSync"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.