China announces Sora challenger that can generate videos from text similar to OpenAI tool, but much shorter

China has devised its own text-to-video artificial intelligence (AI) tools, including: OpenAI Sorabut the new model can only create videos no longer than 16 seconds, compared to 60 seconds for the US service.

Vidu, so far the country's biggest hope to catch up with Sola, was launched over the weekend by startup Shengzhou Science and Technology in collaboration with Beijing-based prestigious Tsinghua University.

The company says the model can generate 1080p resolution videos based on simple text prompts.

“Vidu is the latest achievement in autonomous innovation, making breakthroughs in many fields,” said Zhu Jun, principal researcher at Shengshu, who is also deputy director of the Tsinghua AI Research Institute. He announced this model at the Zhongguancun Forum held in the capital of China and said: , according to a Beijing News report.

A screenshot of a demo video released by Shengshu. — Screenshot of the demo video published by Shengshu.

Vidu is “imaginative,” “able to simulate the physical world,” and “able to create 16-second videos with consistent characters, scenes, and timelines,” Zhu said, adding that the model has “Chinese elements.” He added that he could understand.

During the model's unveiling, Shengshu released several demo clips, including one featuring a panda sitting on the grass and playing a guitar, and another featuring a puppy swimming in a pool. Both show vivid detail.

Vidu's debut raises expectations in a country racing to catch up with the world's major generative AI players, including Microsoft-backed OpenAI.

Unlike OpenAI's ChatGPT, which inspired many China-based competitors after its launch in November 2022, Sora video previews published in February have attracted a lot of attention from Chinese big tech companies and startups. It hasn't attracted a similar level of enthusiasm.

Industry experts said one of the factors hindering the advancement of Chinese companies is the lack of sufficient computing power.

Sora needs 8 clips to create a 1 minute clip Nvidia A100 graphics processing unit (GPU) The system can run for more than three hours, said Li Yanwei, a Beijing-based technology consultant working in the field of intelligent computing.

“Sora requires a lot of computing power for inference,” he said.

The United States has tightened export controls on advanced chips made by Nvidia and others, such as the A100 and H100 GPUs, which are hugely popular as training components for AI systems, but are prohibited from being shipped to China.

OpenAI's Sora pours 'cold water' on China's AI dreams

Beijing-based Shengshu was founded in March 2023, and its core team consists mostly of members from the Tsinghua Provincial AI Research Institute, as well as members from Alibaba Group Holding, Tencent Holdings, and ByteDance. Alibaba, the owner of Post Inc., is also working on its own video model.

Shengshu raised hundreds of millions of yuan last month from investors including Qiming Ventures. Shiura AI According to emerging database provider ITjuzi, Baidu Ventures.

Source link