1 paper
Zhijian Qiao, Xinjiang Wang, Jiajie Chen +5
In camera-controlled video generation, geometry-aware positional encodings condition tokens on camera extrinsics and per-token viewing rays. Existing schemes, however, have a scale…