Sora Who? Kling Is the New AI Video Generator Everyone Is Talking About

While OpenAI’s forthcoming Sora AI video generation tool has creatives both tantalized and horrified, a new player has emerged that has many AI enthusiasts crowing about its capabilities. Developed by Kuaishou, a Chinese tech giant and competitor to TikTok, Kling is also already available—with some caveats.

Kling's capabilities surpass the competition in several significant ways. Kling can generate videos up to two minutes long in 1080p resolution at 30 frames per second. This is a substantial improvement over Pika and RunwayML, which can generate only a few seconds each—and it’s twice as much as what OpenAI said Sora would handle.

Kling appears on the scene shortly after the release of Vidu, another Chinese AI video creation tool developed by Shengshu Technology and Tsinghua University. Vidu generates shorter videos, but can create highly realistic and detailed videos with Universal Vision Transformer (U-ViT) technology, which has already proven superior to Runway and Pika.

 Artificial Intelligence

Midjourney Leaps into AI Video Creation

Midjourney, the generative image creation tool perhaps best known for running inside a Discord server, is spreading its AI wings. The creators of Midjourney announced on Tuesday that they plan to introduce a “text to video” model in the next few months. The company will begin training its video models starting in January, CEO David Holz said during an "Office Hour" Discord session. This move represents a natural progression for the platform, building upon a mature image model to stir the competi...

For its part, Kling uses advanced 3D face and body reconstruction technology to generate realistic movements and limb movements based on a single, full-body picture. It also avoids other common problems that some AI video generators have when depicting people, such as extra or impossibly bent limbs.

Like Sora, Kling also has a better understanding of real-world physics, allowing for more accurate simulation of physical interactions between objects. It’s also designed to accurately follow prompts and create sequences of shots with multiple views, allowing for more complex and dynamic videos.

Kuaishou released a series of demo videos showing how Kling would render the scenes OpenAI demonstrated with Sora. The videos seemed very realistic, with good scene compositions and movements. The quality was superior to Runway and Pika in terms of realism and consistency, and was on par with Sora—even surpassing it in some generations.

Some users have shared other creations on social media, and they seem convincing enough to justify the hype.

Kling is currently available as a public demo in China through a waitlist, but Kuaishou is promising a wider global release.

The Kling website is in Mandarin and appears to require a Chinese phone number in order to register. You will also need to download the Kwaicut app and follow the steps shown on the official Kling-AI website before securing a place in line.

Until then, for those looking to generate AI videos, the best choices are Pika Labs, RunwayML and—for local generation—Stable Video Diffusion.

Edited by Ryan Ozawa.