3D Software Synthesis Driven by Constraint-Expressive Intermediate Representation
Scenethesis is a method that converts natural language descriptions of 3D scenes into executable 3D software. It uses ScenethesisLang, a special language, to formally define spatial constraints. The process breaks down into four stages: translating text to specs, synthesizing individual objects, solving complex spatial constraints iteratively using a Rubik-inspired solver, and finally generating Unity-compatible code.