Skip to content

AI4AnimationPy

AI4AnimationPy is a Python framework for AI-driven character animation using neural networks, developed by Paul Starke and Sebastian Starke.

It provides motion capture processing, training & inference, animation engineering, and optional real-time rendering — all in Python via NumPy/PyTorch. It replaces the Unity dependency of AI4Animation by unifying data processing, feature extraction, neural network training/inference, and visualization into a single Python framework.

Key Features

  • Entity-Component-System (ECS) architecture with game-engine-style lifecycle (Update / Draw / GUI)
  • Vectorized math library — quaternions, axis-angle, matrices, Euler, mirroring
  • Neural network architectures — MLP, Autoencoder, Flow Matching, Codebook Matching
  • Optional real-time rendering via Raylib
  • Inverse kinematics solver
  • Motion feature modules — root trajectory, joint contacts, etc.
  • GLB, FBX, BVH import pipeline with NPZ serialization
  • Three execution modes — Standalone, Headless, Manual

Technology Stack

Layer Technology
Language Python 3.12+
Math Backend NumPy (default), PyTorch (switchable)
Neural Networks PyTorch, ONNX Runtime
Rendering Raylib (optional)
Motion Formats GLB, FBX, BVH, NPZ