OmniLottie
OmniLottie is an open-source, end-to-end multimodal instructed vector animation generator that produces Lottie JSON outputs. Standing as a CVPR 2026 publication, it is the first system of its kind to leverage pre-trained Vision-Language Models to generate complex animations from diverse inputs including text, images, and videos. The project introduces MMLottie-2M, a large-scale dataset containing two million richly annotated Lottie animations, to support research and development in this domain. Additionally, it provides MMLottieBench, a standardized benchmark for evaluating multi-modal vector animation generation capabilities. The software includes model weights for a 4B parameter version, inference code, and an online demo hosted on Hugging Face. It supports integration into ComfyUI via community plugins, enabling users to generate animated vector graphics directly within node-based workflows. OmniLottie is designed for developers, designers, and researchers seeking to automate the creation of scalable, ligh