We develop Video Latent Diffusion Models (Video LDMs) for computationally efficient high-resolution video synthesis. We first pre-train an LDM on images only; then, we turn the image generator
into a video generator by introducing a temporal…