1 paper
Zixun Fang, Yawen Shao, Kai Zhu +7
We introduce \textbf{CoDeR}, a new paradigm for world modeling. Unlike existing video world models that implicitly represent world dynamics through visual observations, our system…