Creating three-dimensional objects and scenes from text or photographs. Used in games, product visualisation, film and architecture. Output quality has improved quickly and usually still needs cleaning up by hand before it can be used in a real production pipeline.
Technical view
Representations include meshes, neural radiance fields and Gaussian splatting, the last two reconstructing scenes from multiple photographs with high visual fidelity and awkward integration into standard content pipelines. Text-to-3D approaches distil two-dimensional diffusion priors into three-dimensional representations, with multi-view consistency the central difficulty. Practical adoption depends on producing clean topology and usable materials, not just plausible appearance.
Related
Part of AI Terminology Index, section: Modalities and Applications.