Kuaishou, one of TikTok's main rivals to its Chinese sister site Douyin, showed off several new features of its text-to-video model Kling AI at the World Artificial Intelligence Conference (WAIC) in Shanghai last week, including the ability to generate videos of up to 10 seconds in length.
At WAIC, visitors lined up to try out tools like Sora, which is currently available by invitation only: Users generated videos by submitting simple prompts like “panda eating salmon” or “Mona Lisa putting on glasses with her hands,” and the resulting clips demonstrated the Kling AI's ability to render inputs nearly perfectly.
AI-generated videos have since flooded the Chinese internet, with Kling AI being used to create clips of characters from historical films carrying out modern-day missions, spawning multiple memes.
A video of My Fair Princess character Rong Tao feeding Princess Ziwei chicken thighs, which has become a popular internet meme in China, has recently gone viral on social platforms. The AI-generated video is based on the show's most famous scene in which Rong Tao tortures Princess Ziwei by repeatedly stabbing her with needles.
Why is this important: Kuaishou will likely be hoping that its series of large-scale homegrown models, including language model KwaiYii, image-focused Kolors and video-centric Kling, will give it an advantage as it continues to challenge ByteDance's Douyin and TikTok.
detail: More than 500,000 users have signed up to help beta test Kling, and the number of videos generated so far has reached 7 million, Kuaishou's senior vice president Guy Kun revealed at the WAIC forum last weekend. Kling, a rival to Sora, is so popular that English-language posts can be found on X, formerly known as Twitter, instructing users outside of China how to sign up for a trial of Kling AI.
- Kuaishou provided practical tips on screen at the WAIC event, advising users to use simple words and sentence structures and avoid overly complicated language. It also emphasized that the model is not sensitive to numbers, giving the example that if given the prompt “10 puppies on the beach”, the number may not remain consistent in the output.
- A member of Kuaishou's large-scale language model team told TechNode that the data used to train Kling AI cannot be made public, but did indicate that it is open source.
- Meanwhile, TikTok's rival WAIC announced it would open-source its Midjourney-like model Kolors, a move Kuaishou said was aimed at contributing to a richer ecosystem of text-to-image generation communities.
- Kuaishou Group's investment in research and development will increase fourfold in four years, with spending increasing from RMB 2.9 billion in 2019 to RMB 12.3 billion in 2023.
context: Kuaishou, China's second-largest short video producer, has launched an AI strategy in 2023, CEO Cheng Yixiao said, saying generative AI has a “very rich combination of business scenarios and great value potential” for content platforms.
Editor's note: “Landing AI” is a special report series focused on the field of artificial intelligence compiled by TechNode. By exploring the development of AI landing in China and the behind-the-scenes stories of the industry, we will dig deeper into what is possible under the new wave of AI.
