Researchers from Peking University and an AI company in Shenzhen called Rabbitpre have teamed up to start a project called Open-Sora. Their goal is to create a version of OpenAI's text-to-video model, Sora, with the help of the open-source community, that is easier to use and more scalable, reports South China Morning Post. OpenAI's Sora has created a lot of noise in the AI world since it came out, and this project is China's latest move in the AI race. The big question is whether this will be a success.
According to the project's GitHub page, the Open-Sora project has made considerable progress by developing a three-part framework and presenting four demonstrations of restructured videos. These videos vary in resolution and aspect ratios, with durations ranging from three to 24 seconds.
The project's future objectives include refining the technology to produce higher-resolution videos and enhancing training with additional data and more graphics processing units to give the service more computing power. These steps are crucial for improving the model's performance and expanding its capabilities to generate more detailed and longer videos.
The introduction of OpenAI's Sora has caused varied responses within the Chinese business and technology sectors. Some companies are eager to utilize the text-to-video AI model. Others are more concerned about China's competitive edge in this field, mainly because of the United States' tightening export rules that prevent companies like Nvidia from selling high-performance GPUs to China-based entities, which greatly slows down the development of AI technologies in the People's Republic.
China's big tech companies, such as Tencent AI and ByteDance, have also ventured into the text-to-video domain. Tencent AI released VideoCrafter2, an open-source video generation and editing toolbox capable of generating videos from text, albeit limited to two-second videos. ByteDance's MagicVideo-V2 combines various modules to create an end-to-end video generation pipeline.
The Open-Sora initiative is a collaboration between the Peking University Shenzhen Graduate School and Rabbitpre, forming the 'Rabbitpre AIGC Joint Lab.' Established in June 2023, this joint lab is dedicated to researching AI-produced content. The project team comprises 13 members, including Assistant Professor Yuan Li and Professor Tian Yonghong from Peking University, Rabbitpre's founder and CEO Dong Shaoling, and Chief Technology Officer Zhou Xing.
