Researchers have developed a new artificial intelligence technology for creating videos, called “TurboDiffusion,” which is characterized by a speed up to 200 times faster than similar technologies, while maintaining image quality and detail accuracy.
This innovative tool is the result of a collaboration between “ShengShu Technology,” Tsinghua University in China, and a group of researchers from the University of California in the United States, with the aim of reducing the time and costs associated with producing videos using artificial intelligence, which previously required a long time and enormous computing capabilities.
The researchers reported that they tested the new technology on a desktop computer equipped with a modern “Nvidia RTX 5090” graphics card, and the results showed that the time required to produce a 5-second video with standard definition (“SD”) was reduced from more than 3 minutes to only 1.9 seconds.
Similarly, the video generation time for high definition (“HD”) videos of the same duration decreased from about 80 minutes to only 24 seconds, representing a significant improvement in processing speed of about 200 times compared to current methods.
While well-known tools such as “Vidu” from “ShengShu” and “Sora” technology that “OpenAI” is developing require several minutes to produce short clips, the “TurboDiffusion” tool represents a qualitative leap, as it enables users to create visual content in a matter of seconds or minutes instead of hours.
The “TurboDiffusion” technology for creating videos does not rely on a single mechanism to accelerate performance, but combines 4 advanced technologies specifically designed to speed up and facilitate the process in an unprecedented manner, while maintaining image quality and scene stability, by reducing the amount of data processed inside the device, without compromising the quality of the results.
The technology relies on an algorithm known as “SageAttention” that allows it to perform operations at high speed via specialized electronic chips, and also uses an approach called “Sparse Attention”, which focuses only on the important information within each image or scene, reducing the number of calculations and saving time and effort.
Instead of going through many stages to create each video, “TurboDiffusion” only uses 3 or 4 steps, thanks to a modern technology known as distillation, making it much faster than traditional methods.
This technology also works to reduce the size of the numbers and data used within the system (to what is called 8-bit), which reduces the pressure on the device’s memory and speeds up the processing process significantly.
When these four technologies are combined, “TurboDiffusion” technology can create high-quality videos at high speed, making it ideal for content creators, producers and designers who need instant and effective results.
Major companies have already started adopting this technology, such as “NVIDIA”, which integrated it into its famous “TensorRT” engine, in addition to other platforms such as “Huawei Ascend” and “Moore Threads”.
