1 paper · 1 filter
Lingyu Liu, Yaxiong Wang, Li Zhu +1
We introduce a training-free framework specifically designed to bring real-world static paintings to life through image-to-video (I2V) synthesis, addressing the persistent challeng…