Animating General Image with Large Visual Motion Model

摘要

We present the pioneering Large Visual Motion Model (LVMM) meticulously engineered to analyze the intrinsic dynamics encapsulated within real-world imagery. Our model fortified with a wealth of prior knowledge extracted from billions of image pairs demonstrates promising results in predicting a diverse spectrum of scene dynamics. As a result it can infuse any generic image with authentic dynamic effects enhancing its visual allure.