L4GM is a 4D large-scale reconstruction model that can quickly generate animated objects from single-view video input. It utilizes a novel dataset comprising multi-view videos showcasing animated objects rendered in Objaverse. This dataset includes 44K distinct objects and 110K animations, rendered from 48 viewpoints, generating 12M videos encompassing a total of 300M frames. L4GM is built upon the pre-trained 3D large reconstruction model LGM, which can output 3D Gaussian ellipsoids from multi-view image input. L4GM outputs a 3D Gaussian Splatting representation for each frame, subsequently upsampled to a higher frame rate for temporal smoothing. Moreover, L4GM incorporates time-self attention layers to assist in learning temporal consistency and utilizes multi-view rendering loss at each timestep for model training.