OmniLottie

Why we included this project

Motion designers and front-end teams already shipping Lottie files will find this useful. You can prompt it with text, an image, or a video clip, and it returns an editable vector animation as a Lottie JSON rather than a raster video, so the result feeds straight into the animation pipeline they already use for web and mobile. The repo includes inference code, pretrained weights, a Gradio demo, and a benchmark, so you can run your own prompts and see the output quality before adopting it. For teams automating early animation drafts for UI motion or product illustrations, it's a reasonable starting point, though training code is still on the roadmap and the 4B model needs a capable GPU.