Show HN: Lance – image/video generation and understanding in one model

ByteDance released Lance, a 3B active parameter unified model handling image/video generation and understanding in a single architecture. The multimodal unification of generative and perceptual tasks in one compact model is notable; paper at arxiv 2605.18678 and weights on HuggingFace. Practitioners building vision pipelines should evaluate it as a lightweight alternative to separate gen/understanding models.