SG-I2V: Self-Guided Trajectory Control in Image-to-Video Generation

Koichi Namekata1   Sherwin Bahmani1,2   Ziyi Wu1,2   Yash Kant1,2   Igor Gilitschenski1,2   David B. Lindell1,2  
1University of Toronto 2Vector Institute

arXiv 2024

Project Page

Comparison with Supervised Baselines (Paper Fig. 4)

Input Trajectory

Ours

DragNUWA

DragAnything