Pose Animator – An open source tool to bring SVG characters to life (2020) (blog.tensorflow.org)

Google Creative Technologist Shan Huang released Pose Animator, an open‑source tool that maps real‑time pose and facial keypoints from TensorFlow.js PoseNet and MediaPipe FaceMesh (webcam) models onto SVG characters to produce live vector animation in the browser. The project provides a full‑body rig (90 keypoints, 78 bones) and a system that drives SVG bezier curves and bones from model outputs, with design‑tool integration (e.g., Illustrator) so illustrators can embed rigs directly in artwork. Small details—like sharing weights for collinear curve handles—help keep vector geometry stable as characters move. For the AI/ML community this is significant because it bridges research-grade pose estimation and practical creative tooling: real‑time, client‑side ML (TF.js + MediaPipe) enables low‑latency, privacy‑preserving animation pipelines without server inference. Technically, Pose Animator translates noisy pose/facemesh outputs into smoothed bone transforms and bezier control adjustments, democratizing character rigging and opening use cases in interactive web experiences, virtual avatars, education, and rapid prototyping. It also highlights how model keypoints can be turned into robust geometry controllers, pointing to further hybrid systems that combine lightweight on‑device vision models with expressive procedural animation.