vidi

We introduce Vidi, a family of Large Multimodal Models (LMMs) for a wide range of video understanding and editing (VUE) scenarios. The models focus on temporal retrieval, spatio-temporal grounding, and maintaining robust open-ended video QA performance.