COrigami

Combines reinforcement learning with Gemini to design origami crease patterns, using a semantic representation and visual feedback loop to fold arbitrary target shapes.

Paper ↗Crease Patterns ↓

Designed with COrigami, shaped and folded by Brandon Wong.

Origami is a uniquely constrained art form: every design must obey the mathematics of flat-foldability while still looking like the thing it depicts. COrigami is an end-to-end AI pipeline that tackles both at once, turning a natural-language prompt into a crease pattern that genuinely folds.

By pairing algorithmic optimisation with an autonomous aesthetic critic and optimising it with RL, the system acts as a collaborative assistant — it produces mathematically sound starting points that human artists can expand and shape into a finished, physical model.

How it works

1
Semantic stick figure
Gemini turns a natural-language prompt into an abstract, structured description of the subject — its parts and proportions.
2
Base packing
The stick figure is packed into a base that allocates a region of paper to each limb, flap, and appendage.
3
Flat-foldable solving
A solver searches for a crease pattern that provably folds flat, satisfying origami's rigid geometric constraints.
4
Shaping
The flat-folded base is shaped into a recognisable three-dimensional form.
5
Aesthetic RL loop
A reinforcement-learning loop refines the design, guided by an autonomous critic that scores how good the result looks.