COrigami: AI pipeline that designs flat-foldable origami from text

COrigami tackles the challenge of generating physical art that satisfies both strict geometric constraints and subjective aesthetics. The pipeline starts by converting a natural language description into a semantic stick figure, which represents the core structure of the desired origami shape. Next, it computes a base packing—a set of paper regions that map to specific parts of the subject—and then solves for a flat-foldable crease pattern using mathematical equations that guarantee the design can fold flat without tearing. Finally, the system shapes the folded pattern and refines it through reinforcement learning driven by an autonomous aesthetic evaluation loop, which judges visual recognizability and appeal.

This work demonstrates how AI can co-create with humans in a mathematically rigid domain. The system generates structural starting points that human artists can further modify and expand, effectively serving as a collaborative assistant. By integrating algorithmic optimisation with multi-objective constraints—both physical foldability and visual aesthetics—COrigami pushes the boundaries of generative AI in applied art. The research, published on arXiv (cs.AI), highlights a path toward reliable, mathematically grounded co-creativity where machines help artists overcome technical barriers to explore complex designs.