Lance Unifies Image and Video Tasks in a Single Efficient Model

Lance demonstrates that unified multimodal capabilities don't require massive models—achieving strong performance across image generation, editing, and video understanding with only 3B parameters. This efficiency breakthrough is significant for making advanced generative AI more practical and deployable at scale. The staged multi-task learning approach shows a new paradigm for training models that excel simultaneously across diverse generative and understanding tasks.