A new research paper by Meta and Oxford University scientists outlines a powerful AI-driven technique for generating scalable 3D models. Text or 2D image prompts are all it needs. An online demo is available.