China's Unveils An AI Video Generator

When OpenAI announced Sora, an AI-powered tool that transforms text descriptions into videos, it quickly became a global sensation because of its mind-blowing results. The closest competitors, Pika Labs and RunwayML were significantly lagging in terms of quality and coherence in the text prompt.

Perhaps one reason is that AI video generators require a significant amount of computing power to run. Sora requires eight NVIDIA A100 graphics processing units (GPUs) running for over three hours to produce a one-minute clip. One NVIDIA A100 costs over 10,000 USD.

But today, Sora might finally have a real competitor—China has just unveiled its text-to-video AI tool, Vidu.

What is Vidu?

Vidu is a text-to-video AI model developed by Chinese AI startup ShengShu Technology and Tsinghua University. It was announced on April 27, 2024, and is designed to generate high-definition, 16-second videos in 1080p resolution with a single click.

According to the chief scientist at Shengshu, Zhu Jun,

“It is imaginative, can simulate the physical world, and produces 16-second videos with consistent characters, scenes, and timeline.”

Check out some examples from the demo video below.

I am not 100% sure, though, if the examples in the demo video were really generated by Vidu and not manipulated in any way.

The Vidu AI model is built on a proprietary visual transformation model architecture called the Universal Vision Transformer (U-ViT), which integrates two text-to-video AI models: the Diffusion and the Transformer.

This architecture enables the creation of really good videos with dynamic camera movements, intricate facial expressions, and authentic lighting and shadow effects

Is it better than Sora?

It’s funny because the demo video includes clips that resemble some of OpenAI’s demo videos for Sora. Take a look at the side-by-side example below:

Which one is better? Well, to be honest, I like both videos. But the softer and warmer tones of the video generated by Vidu make it look more realistic and natural.

Also, it’s worth noting that Sora can generate up to 60-second videos, while Vidu can only render up to 16-second videos.

How to get access

Vidu is currently not available to the general public. However, they have opened up a waitlist form to get early access:

1. Go to www.shengshu-ai.com
2. Click the blue button in the top right of the page
3. Fill out the form to request access

They are also planning to integrate the video generation capability into an AI tool called PixWeaver.

Final Thoughts

The demo videos from Vidu look super impressive, but we should take them with a grain of salt until we can check out the quality and realness of the videos ourselves.

China showing off Vidu is a big deal because it shows they’ve got the tech and resources to compete with the best in the world when it comes to AI. I can’t wait to get my hands on Vidu and see how it stacks up against Sora.

‍