facebookresearch/ai4animationpy

AI4AnimationPy is a Python framework for creating AI-driven character animation, meaning it trains neural networks on motion capture data (recordings of real human or animal movement) and then uses those networks to generate smooth, realistic character movement in real time. The core problem it solves is workflow fragmentation: traditionally, AI animation research required training models in Python but then switching to a separate 3D visualization tool to see results, with awkward data pipelines connecting the two. This framework puts everything, training, inference (running the model), and real-time visualization, into one unified environment built on NumPy and PyTorch (standard Python numerical computing libraries). The architecture borrows ideas from game engines: it has an Entity-Component System (a design pattern where objects are made of reusable parts called components), update loops, and a real-time renderer with features like shadow mapping and post-processing effects. Motion capture files can be imported from GLB, FBX, and BVH formats, which are standard 3D animation file types. Interactive demos included with the project show a walking biped character, a dog with gait transitions, real-time inverse kinematics (a technique for making limbs reach toward targets naturally), and motion editing tools. You would use this if you're a researcher or developer working on AI character animation, for example, generating locomotion controllers, training on public motion capture datasets, or building physics-based character simulations. It's written in Python and licensed for non-commercial use. The full README is longer than what was provided.