This technique can create AI videos in just seconds!
What's the story
Researchers from Tsinghua University, Shengshu Technology, and the University of California, Berkeley have developed a new technique called TurboDiffusion. The method can speed up artificial intelligence (AI) video generation by up to 200 times without compromising image quality. The team tested TurboDiffusion on a consumer-grade system with NVIDIA's RTX 5090 graphics processing unit. It found that it could reduce the time taken to generate a standard-definition five-second video clip from over three minutes to just 1.9 seconds.
Enhanced efficiency
TurboDiffusion's impact on HD video generation
The TurboDiffusion technique also proved its efficiency in generating high-definition videos. On the same machine, it cut down the time to generate a five-second HD video from almost 80 minutes to just 24 seconds. This is an improvement of about 200 times over traditional methods. Currently, platforms like Shengshu's Vidu take around three and five minutes for a standard five-second video, while OpenAI's text-to-video model Sora needs several minutes for short clips.
Technological advancement
Significance in AI video generation
The development of TurboDiffusion highlights the rapid progress made by Chinese researchers in improving AI video-generation technology. The new technique could revolutionize the industry by eliminating bottlenecks in production processes, as noted by industry analyst Kyon Xu. This means that video-generation models would no longer slow down content creation, paving the way for faster and more efficient production workflows.