InternVideo2 Scaling Video Foundation Models for Multimodal Video Understanding We introduce InternVideo2, a new video foundation model (ViFM) that achieves the state-of-the-art performance in action recognition, video-text tasks, and video-centric dialogue. Our approach