Chinese Sora Alternative Blows Everyone’s Mind

While the world is waiting for OpenAI’s Sora, Chinese TikTok competitor company Kuaishou has dropped a Sora-like model which is crazily good. Called Kling, the model is for open access and creates videos even better than Sora in many cases. 

With the prompt - A Chinese man sits at a table and eats noodles with chopsticks; the model generated an almost realistic looking video when compared to Will Smith’s demonic looking noodles video released last year created by Modelscope Text2Video. 

Kling can generate 2 minute videos with a single prompt in 1080p quality at 30fps and accurately simulates real-world physical properties. 

Leveraging Diffusion Transformer architecture, KLING translates rich textual prompts into vivid scenes. With proprietary 3D VAE and support for various aspect ratios through variable resolution training, KLING has advanced 3D face and body reconstruction technology that allows for full expression and limb movement drive from a single full-body photo.

It is clear that China is increasingly getting ahead when it comes to building AI models. Kling, currently in open access, gives just a preview of what the country is building. 

OpenAI has said that it is planning to release Sora by the end of this year, but it might be too late for the company to catch up with China’s text-to-video models. The only thing that it is keeping this at bay is that China might not release the model for world wide access.

Interestingly, Kling isn’t the first video generation model from China. Released in April, Vidu AI was the first chinese version of Sora which was able to create 16 seconds long with 1080p resolution.