During a recent tech conference, a Chinese startup launched an AI model similar to Sora, attracting significant attention. A Chinese startup, Zhipu AI, has launched a new AI model called Ying, which can create videos from text inputs. This development follows the trend set by OpenAI’s Sora model. The launch highlights the growing competition in the AI video creation space globally.
Reports, including one from The Information, indicate that Zhipu AI’s introduction of Ying reflects a broader trend among Chinese tech companies. These firms are racing to develop text-to-video AI models, similar to those from OpenAI. OpenAI’s services, including its models, are currently unavailable in China due to regional restrictions.
Zhipu AI’s Ying can generate a six-second video from text within 30 seconds. The model is accessible to all users in China through the official website and mobile apps of Zhipu AI’s ChatGLM chatbot. Users can also fine-tune their results with various style options. At the recent World Artificial Intelligence Conference in Shanghai, a demonstration featured an old photograph brought to life using technology similar to Ying.
China’s Push in AI Amid US Sanctions
A Chinese startup launched an AI model similar to Sora by integrating advanced features. With the US imposing sanctions on AI models and chips developed by American companies, Chinese tech firms have intensified their focus on AI advancements. Major players like Alibaba, SenseTime, and Baidu are leading this charge.
Investor and technologist Kai-Fu Lee notes that Chinese companies are actively sharing their AI developments on open-source platforms. This approach fosters faster collaboration and innovation within the global AI community.
Yiran Chen, a professor at Duke University, comments that while Chinese companies excel at replicating and enhancing existing US technologies, they face challenges in creating entirely new innovations. However, their rapid advancements in AI suggest significant progress in the coming years.
Ying Model Capabilities and Features
Zhipu AI has unveiled its new text-to-video model, Ying, which generates six-second video clips from both text and image prompts in approximately 30 seconds. Users can customize the videos with various style options, including 3D animation, cinematic effects, and oil painting aesthetics. Additionally, emotional themes like tense, lively, and lonely can be applied for personalized results.
Ying is available through Zhipu AI’s ChatGLM chatbot on both the official website and mobile apps. At the launch event in Beijing, the company announced that the service is immediately accessible to all users for unlimited use. However, the free version may experience longer wait times during peak usage.
The launch of Ying follows closely on the heels of a similar release by Kuaishou, the short video platform rivaling ByteDance’s Douyin, known as TikTok outside China. Kuaishou’s Kling video model, introduced just two days earlier, allows users to generate six videos per day and offers annual paid plans for up to 60 or 800 monthly video generations, priced at 396 yuan (US$54.63) and 3996 yuan, respectively. This competitive landscape indicates that Chinese tech firms are positioning themselves against industry leader OpenAI in the AI video generation market.
Technology Behind Ying
OpenAI, based in San Francisco, pioneered AI video generation with the announcement of Sora in February. However, the model is still not publicly accessible. Aditya Ramesh, a key developer, explained that the team is focused on ensuring the model cannot be misused to create and spread false information.
To rival OpenAI, a Chinese startup launched an AI model similar to Sora, emphasizing speed and efficiency. Zhipu AI’s Ying utilizes a self-developed text-to-video model named CogVideoX, similar to OpenAI’s Sora, which employs diffusion transformer (DiT) architecture. Ying boasts an improved inferencing speed, enabling faster video generation. Zhipu AI’s CEO, Zhang Peng, mentioned that the development team drew inspiration from Sora’s algorithm design. While OpenAI has shared technical details about Sora, the model is not yet available for public use.
Also Read: Apple’s Artificial Intelligence Features Are To Be Delayed.