OpenVGLab/OmniLottie

Leverages Vision-Language Models to generate Lottie animations from multimodal inputs (text, images, videos) using parameterized Lottie tokens, enabling generation of complex vector animations in a single end-to-end pipeline. Includes the MMLottie-2M dataset (2M annotated animations) and MMLottieBench evaluation protocol for standardized benchmarking of multimodal vector animation generation. Supports both original PyTorch and HuggingFace safetensors formats with seamless integration via `from_pretrained()` API, plus community-contributed ComfyUI plugin support.

503 stars. Actively maintained with 12 commits in the last 30 days.