FantasyPortrait is a video generation model based on diffusion transformers that animates portraits by augmenting expressions. It supports animation of single and multiple characters simultaneously. The model is published with weights on Hugging Face and ModelScope, and includes inference code and datasets for training.