Diffusion models, which have shown success in generating images, are now being applied to video generation. This task is more challenging than image synthesis because it requires temporal consistency across frames and more world knowledge. Collecting large amounts of high-quality video data is also difficult compared to text or images. --- Why it matters: This development matters for researchers working on AI models that can generate realistic videos, which has applications in areas like film, animation, and virtual reality. Source: https://lilianweng.github.io/posts/2024-04-12-diffusion-video/