OmniMotion is an AI tool that enables tracking of all points in a video across all frames, even through occlusions. Unlike traditional methods that operate within limited temporal windows, OmniMotion provides accurate, full-length motion estimation of every pixel in a video. It achieves this by representing the video using a quasi-3D canonical volume and performing pixel-wise tracking through bijections between local and canonical space. This representation ensures global consistency, allows tracking through occlusions, and models both camera and object motion. OmniMotion outperforms existing methods in terms of accuracy and quality, as demonstrated through evaluations on benchmark datasets and real-world footage. The tool also provides visualizations of correspondences and pseudo-depth, although it may struggle with rapid and non-rigid motion or thin structures. Overall, OmniMotion is a powerful and comprehensive tool for motion tracking in videos, with potential applications in areas such as computer vision, video analysis, and augmented reality.