fog framework lets AI code express motion and emotion with 68% accuracy

Vivian Liu and Lydia Chilton from Columbia University present fog, a function composition framework that uses AI-generated code to express motion and emotion in animations. The framework builds an open-ended motion vocabulary by composing functions for verbs, adverbs, gestures, and emotions. Paired with an animation editor that allows direct manipulation and dynamically generated UI, fog enables users to create Heider-Simmel style animations where abstract shapes convey complex social behaviors. The approach treats motion as modular, composable code, making expressive animation accessible even to non-programmers.

Evaluation results are compelling: in a perceptual study with 452 fog-generated animations, viewers correctly identified intended semantic meanings 68% of the time—a 2.68x improvement over chance baselines. A mixed-methods user study with both professionals and novices found that fog's interface accelerates iteration speed and expands creative exploration while maintaining fine-grained control. This work bridges AI code generation and human-computer interaction, offering a new paradigm for generating emotionally intelligent animated content without requiring deep artistic or programming expertise.