Kuaishou, China's second-largest short video app, has introduced its new artificial intelligence (AI) model, Kling, designed to generate high-quality videos from text prompts (via South China Morning Post).
The move places Kuaishou in direct competition with OpenAI's Sora and other emerging players in AI-powered video generation.
Kling
How Kling Works
Currently in its trial phase, Kling can transform text prompts into video clips up to two minutes long, with a resolution of 1080p. According to the company, Kling supports various aspect ratios and can generate both realistic and imaginative scenes.
Demonstration videos showcased a range of scenarios, including a white cat driving through city streets and a boy eating a cheeseburger.
Kuaishou's Kling is designed to compete with OpenAI's Sora, which was unveiled earlier this year and has already been used by filmmakers at the Tribeca Film Festival.
While Sora is not yet widely available, Kling is accessible through a waitlist. Other Chinese competitors include Shengshu Technology's Vidu and Zhipu AI's upcoming video generation tool.
Kling utilizes a diffusion transformer model, similar to Sora, to create videos with realistic physics and smooth movements. Kuaishou claims Kling can generate videos in 1080p at 30 frames per second and supports a variety of shot types and aspect ratios.
The model also boasts advanced 3D face and body reconstruction, enhancing the accuracy of expressions and limb movements.
Kling joins a growing list of AI innovations from Kuaishou, including the KwaiYii large language model (LLM) and the Kolors text-to-image model. The company also offers an AI Dancer feature and is developing an image-to-video feature based on Kling's capabilities.
According to SCMP, Kuaishou reported a net profit of 4.12 billion yuan (US$575.1 million) in the first quarter of 2024, a significant turnaround from the previous year's loss.
Revenue rose 17% year-on-year, driven by online marketing services and e-commerce. With nearly 400 million daily active users, Kuaishou remains a major player in the short-video app market, second only to ByteDance's Douyin.
It remains unclear whether Kuaishou will release Kling outside of China. The global availability of such advanced AI technologies could accelerate competition and innovation, potentially pushing companies like OpenAI to expedite their releases and enhance their models' safety and cost-efficiency.
In Other News
A study by Epoch AI warns that the publicly available training data for AI language models could be depleted by the early 2030s, potentially stalling AI progress.
The remarkable growth of AI in recent years has been fueled by larger models and expanding datasets, the study notes. However, the internet's supply of high-quality human-generated text, the lifeblood of AI models like ChatGPT, is not infinite.
