1 paper
Lee Hsin-Ying, Hanwen Jiang, Yiqun Mei +3
Current motion-controlled image-to-video generation models rigidly follow user-provided trajectories that are often sparse, imprecise, and causally incomplete. Such reliance often…