Lance – image/video generation and understanding in one model

Lance is a unified 3B-parameter model from ByteDance that performs both image/video generation and understanding within a single framework, eliminating the need for separate models. Designed for AI researchers and developers, it streamlines multimodal tasks like text-to-image synthesis, visual question answering, and video comprehension. Its interest lies in achieving competitive performance on both generation and understanding benchmarks with a relatively compact architecture, offering a more efficient and integrated approach to multimodal AI.