VideoLDM by Nvidia

VideoLDM by Nvidia is a latent diffusion model designed for high-quality video generation, editing, and understanding. Built on top of Stable Diffusion, it extends image diffusion capabilities into the temporal domain, enabling consistent, coherent video sequences rather than isolated frames. The model operates in a compressed latent space, making it more computationally efficient while preserving visual fidelity and temporal smoothness.

Key capabilities include text-to-video generation, where users can synthesize short video clips from natural language prompts, and image-to-video generation, which animates a single image according to a described motion or scene evolution. VideoLDM also supports video-to-video transformations, such as style transfer, appearance changes, or content modification while maintaining the original motion structure. Its architecture incorporates temporal attention and conditioning mechanisms to handle motion dynamics and enforce frame-to-frame consistency.