SceneChat
A modular LLM-based pipeline for converting natural language queries into 3D indoor scenes in Blender.
SceneChat explores how natural language and user feedback can be used to construct 3D indoor scenes. The system uses a modular LLM-based pipeline to parse user intent, refine scene descriptions, and generate structured outputs that can be rendered in Blender.
The project was built around the idea that scene generation should be interactive: users should be able to steer and correct the system through natural language rather than through low-level scene editing commands.
Topics: LLMs, structured output generation, 3D scenes, Blender, human-in-the-loop systems.