In February this year, OpenAI dropped its video-generation model Sora that can create one-minute-long, high-definition videos. Sora has not even become accessible yet, and another text-to-video generation model seems to be making headlines. Kuaishou Technology, a Chinese company known for its short-video platform, has unveiled its text-to-video model named Kling.
Reportedly, Kling uses a technology similar to Sora and can produce 1080p high-definition videos that can last up to two minutes. The model can create large-scale realistic motions which essentially simulate physical world characteristics.
On Thursday, the company released a demo video showcasing its ability to create a video. Reportedly, the model is open for access and in some instances, it can create videos that are much ahead of Sora’s output.
The Kling video model is self-developed by Kuaishou’s LLM team. As of now, it is available on invite to test on the Kuaiying app, a video shooting and editing app from Kuaishou company.
Several AI enthusiasts shared their creations from Kling on X. The model generates videos up to two minutes at 30fps quality. On the other hand, Sora has only been capable of creating one-minute-long videos so far. Based on the samples posted on X, videos generated by Kling seem to be accurately simulating real-world physical properties.
2.
A car driving on the road in the evening with a gorgeous sunset and tranquil scenery reflected in the rearview mirror pic.twitter.com/keWeKjanNf
— Rowan Cheung (@rowancheung) June 6, 2024
It needs to be noted that Kling is not the only text-to-video model from China. In April, Vidu AI, a text-to-video model capable of 16-second videos at 1080p was released in China. Kling reportedly is based on Diffusion Transformer architecture and can translate textual prompts into vivid visuals.
5. A man riding a horse through the Gobi Desert with a beautiful sunset behind him, movie quality.pic.twitter.com/PAerK5ShCT
— Angry Tom (@AngryTomtweets) June 6, 2024
The Kling model uses advanced 3D face and body reconstruction backed by the company’s proprietary technology 3D VAE, allowing for users to create videos in various aspect ratios. This is possible through variable resolution training, allowing full expression and limb movement from a single full-body photo.
It seems China is rapidly advancing its AI model development and Kling seems to be a testament to its rapid growth. It seems the ChatGPT-maker’s Sora is likely to face stiff competition in China.
