Show HN: Auto-Rig Web – Zero-click 3D humanoid rigging in a Web Worker (github.com)

A developer has introduced Auto-Rig Web, a groundbreaking tool that enables zero-click, browser-local 3D humanoid rigging of static .gltf meshes, transforming how users create custom avatars for interactive 3D platforms like metaverses and gaming. This innovation eliminates the need for manual rigging, which often requires extensive time in tools like Blender or reliance on slow cloud services. Instead, Auto-Rig Web processes rigging entirely on the client side using a lightweight ONNX vision model, completing the task in milliseconds. The significance of this development lies in its efficient, client-side architecture, leveraging WebGPU for accelerated AI inference and Three.js for 3D processing. The tool not only predicts joint positions and constructs skeletal hierarchies but also calculates skin weights using advanced procedural algorithms, allowing for real-time dynamic posing through an integrated Cyclic Coordinate Descent Inverse Kinematics (CCDIK) solver. This promise of a streamlined, high-performance rigging process is poised to empower creators in the AI/ML community to enhance user-generated content while maintaining a high level of flexibility and creativity. The project is open-source, inviting further development and contributions from the community.