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

Koichi Namekata¹ Sherwin Bahmani^1,2 Ziyi Wu^1,2 Yash Kant^1,2 Igor Gilitschenski^1,2 David B. Lindell^1,2

¹University of Toronto ²Vector Institute

arXiv 2024

Comparison with Supervised Baselines

Input Trajectory

Ours

DragNUWA

DragAnything