This repository provides the PyTorch implementation of a CVPR 2023 paper on conditional image-to-video generation. The model uses latent flow diffusion to synthesize realistic videos from single images, incorporating optical flow estimation to model motion patterns. The implementation includes training code and pretrained models for multiple datasets (MUG, MHAD, NATOPS) and supports cross-domain applications such as animating faces from the FaceForensics dataset.