I built a tool that generates 3D objects assembled with separate, logical parts (e.g. it generated a microwave in the video with complete internal assembly and a door that swings open)

Standard AI 3D generators (like Meshy or Tripo) are limited. They produce solid, monolithic 3D objects that look good but are practically useless, because: \- Want to rig or animate it for a game? Can't easily do that, because it’s a dead, monolithic blob instead of a functional, modular asset. \- Want to change the arm of a robot you generated? Regenerate the entire asset. \- Want to edit something manually? The whole thing collapses because it's not actually structured. Free github project here: [https://github.com/RareSense/Nova3D](https://github.com/RareSense/Nova3D) But you'll need to bring your own API Key (BYOK) **Under the hood (if you're interested):** It uses an LLM as a structured code compiler, instead of an image generator. It writes native Blender Python (bpy) code blocks that target specific nodes in the scene graph. *The trick is that everything compiles through Blender's actual scene graph structures instead of pixel or point-cloud diffusion.* Final export is a clean multi-part GLB with transform nodes and working pivot axes preserved.