GIRAFFE is a research project that learns to synthesize and control 3D scenes as compositional generative neural feature fields. It extends NeRF-style implicit representations with a Generative Adversarial Network framework to enable disentangled control over object shape, appearance, and camera viewpoint. The model generates novel scenes by composing multiple objects with different attributes into a unified neural radiance field, producing image animations with controllable camera trajectories and object translations.