1. HunyuanVideo: A Systematic Framework For Large Video Generative Models
- Author
-
Kong, Weijie, Tian, Qi, Zhang, Zijian, Min, Rox, Dai, Zuozhuo, Zhou, Jin, Xiong, Jiangfeng, Li, Xin, Wu, Bo, Zhang, Jianwei, Wu, Kathrina, Lin, Qin, Yuan, Junkun, Long, Yanxin, Wang, Aladdin, Wang, Andong, Li, Changlin, Huang, Duojun, Yang, Fang, Tan, Hao, Wang, Hongmei, Song, Jacob, Bai, Jiawang, Wu, Jianbing, Xue, Jinbao, Wang, Joey, Wang, Kai, Liu, Mengyang, Li, Pengyu, Li, Shuai, Wang, Weiyan, Yu, Wenqing, Deng, Xinchi, Li, Yang, Chen, Yi, Cui, Yutao, Peng, Yuanbo, Yu, Zhentao, He, Zhiyu, Xu, Zhiyong, Zhou, Zixiang, Xu, Zunnan, Tao, Yangyu, Lu, Qinglin, Liu, Songtao, Zhou, Dax, Wang, Hongfa, Yang, Yong, Wang, Di, Liu, Yuhong, Jiang, Jie, and Zhong, Caesar
- Subjects
Computer Science - Computer Vision and Pattern Recognition - Abstract
Recent advancements in video generation have significantly impacted daily life for both individuals and industries. However, the leading video generation models remain closed-source, resulting in a notable performance gap between industry capabilities and those available to the public. In this report, we introduce HunyuanVideo, an innovative open-source video foundation model that demonstrates performance in video generation comparable to, or even surpassing, that of leading closed-source models. HunyuanVideo encompasses a comprehensive framework that integrates several key elements, including data curation, advanced architectural design, progressive model scaling and training, and an efficient infrastructure tailored for large-scale model training and inference. As a result, we successfully trained a video generative model with over 13 billion parameters, making it the largest among all open-source models. We conducted extensive experiments and implemented a series of targeted designs to ensure high visual quality, motion dynamics, text-video alignment, and advanced filming techniques. According to evaluations by professionals, HunyuanVideo outperforms previous state-of-the-art models, including Runway Gen-3, Luma 1.6, and three top-performing Chinese video generative models. By releasing the code for the foundation model and its applications, we aim to bridge the gap between closed-source and open-source communities. This initiative will empower individuals within the community to experiment with their ideas, fostering a more dynamic and vibrant video generation ecosystem. The code is publicly available at https://github.com/Tencent/HunyuanVideo.
- Published
- 2024