3D scene generation glossary
Plain-language definitions for AI 3D creation: gaussian splatting, NeRF, photogrammetry, PBR materials, skybox, HDRI, OpenUSD, procedural generation, and more.
- Gaussian splatting
- A 3D rendering and reconstruction technique that represents a scene as millions of tiny, colored, semi-transparent 3D 'blobs' (Gaussians), enabling photorealistic scenes rendered in real time from ordinary photos or video. [source]
- NeRF (Neural Radiance Field)
- A neural-network representation of a 3D scene learned from 2D images, which can render the scene from new viewpoints. NeRFs preceded and inspired many current capture-to-3D tools. [source]
- Text-to-3D
- Generating a 3D model or scene directly from a written description, typically producing a mesh with textures that can be edited or exported to game engines. [source]
- Mesh (polygon mesh)
- The standard 3D model format: a collection of vertices, edges, and faces (usually triangles or quads) that defines an object's shape and is what most game engines and 3D software consume. [source]
- PBR materials
- Physically based rendering materials describe surfaces with maps such as albedo, roughness, metallic, and normal so they respond to light realistically and look consistent across engines. [source]
- Skybox
- A panoramic backdrop that wraps around a 3D scene to depict distant sky and scenery; AI skybox generators produce these as seamless 360° equirectangular images. [source]
- HDRI
- A high-dynamic-range image storing a wider range of brightness values than a normal photo; 360° HDRIs are used to light 3D scenes with realistic environmental illumination. [source]
- Photogrammetry
- Reconstructing accurate 3D geometry and textures of real objects or places from many overlapping photographs, the basis of most capture-to-3D scanning tools. [source]
- LiDAR
- A sensing method that measures distance with laser pulses; phone and drone LiDAR sensors give 3D scanning apps direct depth measurements to speed up and stabilize reconstruction. [source]
- Procedural generation
- Creating content algorithmically from rules and parameters instead of authoring it by hand; often combined with AI to build terrain, levels, and large game worlds. [source]
- Digital twin
- A digital replica of a real-world object, facility, or process kept in sync with its physical counterpart, used for simulation, monitoring, and planning. [source]
- USD / OpenUSD
- Universal Scene Description, an open framework originally developed by Pixar for composing and interchanging complex 3D scenes across different tools and pipelines. [source]
- World model
- An AI model that learns an internal simulation of an environment so it can predict what happens next; generative world models produce interactive, navigable video or 3D worlds from prompts. [source]