AI that creates hyper-realistic Full HD videos from just Chinese text confronts Sora

Kuaishou, China's second largest short video sharing platform, has launched a text-based video creation AI called Kling with a series of features that helped OpenAI's Sora.

This attracts attention such as improving motion quality, sticking to descriptions and being able to create different frame sequences.

AI that creates hyper-realistic Full HD videos from just Chinese text confronts Sora Picture 1 AI that creates hyper-realistic Full HD videos from just Chinese text confronts Sora Picture 1

With just one description, Kling can create a 2-minute video with 1080p resolution and 30 frames per second, Kuaishou said, accurately simulating physical effects like the real world. Something that most AI models currently cannot do.

Kling can also enhance the movements and expressions of characters in videos with the ability to build real-life 3D faces and bodies.

Although some videos created by Kling still show the blurring commonly seen in AI-generated images, the level is much smaller.

Kling also uses the diffusion transformer model like Sora, but the difference is that China's AI owns a proprietary model, supporting many different frame rates and rotation types.

Currently, Kling has opened a waiting list for users to experience, but it is unclear whether it can be released outside of China or not. While OpenAI's Sora is not yet widely available.

Kling promises to be a formidable competitor to OpenAI's Sora, especially in the Chinese market and shows that China is making rapid progress in AI model development.