Map2Video grounds AI video generation in real-world locations. It lets creators annotate character trajectories on maps and directly manipulate cameras and character masks in geodetic coordinates. In a study with twelve filmmakers, it improved controllability, reduced cognitive effort, and strengthened spatial consistency across shots compared with an image-to-video baseline.
Hye-Young Jo, Mose Sakashita, Aditi Mishra, Ryo Suzuki, Koichiro Niinuma, and Aakar Gupta. 2026. Map2Video: Guiding Real-World-Grounded AI Video Generation. Proceedings of the 39th Annual ACM Symposium on User Interface Software and Technology. ACM.
coming soon