Chinese Researchers Launch Pyramid Flow AI Model for Five-Second Video Generation in 56 Seconds

Oct 10, 2024, 04:21 PM

Chinese researchers have launched a new AI model called Pyramid Flow, which uses a technique known as pyramidal flow matching to generate high-quality videos. This model can create five-second videos at 384p resolution in just 56 seconds. Additionally, Pyramid Flow SD3, an open-source text-to-video model with an MIT license, can generate 10-second videos at 768p resolution and 24 frames per second within 20.7k A100 GPU training hours. The development of these models signifies a significant advancement in the field of AI video generation, with multiple companies like Runway, Kling, and Luma also contributing to this rapidly evolving technology.