pipeline_tag: image-to-video
library_name: diffusers
HighSync: High-Quality Lip Synchronization via Latent Diffusion Models
HighSync is an end-to-end diffusion-based framework for high-fidelity lip synchronization that generates photorealistic talking-face videos aligned with arbitrary input audio. It is the first lip sync model to operate natively at 512x512 resolution, positioning it as a viable solution for professional production environments.
- Paper: HighSync: High-Quality Lip Synchronization via Latent Diffusion Models
- GitHub: saeed5959/high_sync
Abstract
We present HighSync, an end-to-end diffusion-based framework for high-fidelity lip synchronization that generates photorealistic talking-face videos aligned with arbitrary input audio. Existing approaches consistently struggle to reconcile image quality with synchronization accuracy, producing either visually degraded outputs or temporally inconsistent lip movements. HighSync addresses both challenges simultaneously and, to our knowledge, is the first lip sync model to operate natively at 512x512 resolution. Central to our approach is the identification and systematic elimination of a data leakage phenomenon that has silently undermined temporal modeling in prior work, preventing models from developing a genuine dependence on the audio signal.
⚒️ Installation
Environment
Ubuntu 20 or 22
Setup
git clone https://github.com/saeed5959/high_sync
cd high_sync
pip install -r requirements.txt
apt-get install ffmpeg
Download Pretrained Weights
git lfs install
git clone https://huggingface.co/saeed-5959/high_sync pretrained_weights
🚀 Usage
First, convert your source video to 25 FPS:
ffmpeg -i input.mp4 -r 25 out_25.mp4
Then run the inference script:
python -m inference --source_video "video_path.mp4" --driving_audio "audio_path.wav" --output "save_path.mp4"
Citation
@article{daghigh2024highsync,
title={HighSync: High-Quality Lip Synchronization via Latent Diffusion Models},
author={Saeed Firouzi Daghigh and Majid Iranpour Mobarekeh and Mostafa Alavi and Mehdi Bagheri},
journal={arXiv preprint arXiv:2605.16918},
year={2026}
}
🙏 Acknowledgements
This work is mainly based on EchoMimic. We would also like to thank the contributors to the AnimateDiff, Moore-AnimateAnyone, and MuseTalk repositories.