3D graphics is the branch of computer graphics concerned with representing, manipulating and rendering objects and environments using three-dimensional spatial information. A 3D scene normally represents position along three spatial axes and can contain geometry, cameras, lights, materials and other scene information. The final result can be displayed as a two-dimensional image or stereoscopic output even though the underlying scene has three-dimensional structure. 3D graphics is used in games, animation, visual effects, architecture, engineering, computer-aided design, scientific visualisation, simulation, virtual reality and product design. The essential distinction from ordinary 2D graphics is that 3D graphics explicitly models spatial relationships such as depth, viewpoint, surface orientation and three-dimensional transformations.
A typical 3D object is represented by geometric data such as vertices, edges and faces. Polygonal meshes are particularly common, with triangles being the fundamental primitive supported efficiently by most real-time graphics hardware. More specialised representations include subdivision surfaces, implicit surfaces, point clouds, voxels, constructive solid geometry and parametric models. The appropriate representation depends on the application. CAD systems may require precise geometric surfaces and solids, while games commonly use polygonal meshes optimised for real-time rendering. A mesh can also contain information such as vertex normals, texture coordinates, skin weights and material assignments. Geometry is only one part of a complete 3D scene because appearance and spatial context require additional information.
Three-dimensional graphics relies on coordinate systems and transformations. An object can be defined in local coordinates and then transformed into world coordinates by translation, rotation and scaling. A camera defines the viewpoint from which the scene is observed. The rendering system transforms objects through camera or view space and then projects them onto an image plane. Perspective projection represents depth by making distant objects smaller, creating the visual effect associated with human viewing and conventional cameras. Orthographic projection does not apply the same perspective scaling and is common for technical drawings and certain strategy or interface views. Matrix mathematics provides a standard way to combine these transformations efficiently.
Materials and lighting determine how three-dimensional surfaces appear. A physically based material model can describe properties including base colour, metallic response, roughness and transmission. Surface normals determine how lighting interacts with geometry, while texture coordinates map two-dimensional image data onto three-dimensional surfaces. Normal maps and related techniques can create the visual appearance of small surface details without adding equivalent geometric complexity. Lighting can include direct sources, environment maps, global illumination and other approximations. The visual result depends on the interaction of geometry, materials, lighting, camera position and rendering algorithms rather than on the mesh alone.
Visibility is an important problem in 3D graphics because multiple objects can occupy the same projected screen location while only the nearest visible surface should normally appear. Depth buffers provide a common real-time solution by storing depth information for rendered fragments and comparing new fragments against existing depth values. Back-face culling can discard surfaces facing away from the camera when they are known to be invisible for the particular material and geometry. More sophisticated rendering methods can handle transparency, reflections and indirect visibility. These techniques are necessary because simply projecting every polygon onto the screen would not correctly reproduce the spatial ordering of surfaces.
3D graphics can be rendered in real time or offline. Real-time rendering is designed to generate images quickly enough for interactive applications, such as games and simulators. Offline rendering can take considerably longer and is often used for film-quality visual effects, animation and architectural imagery where image quality is more important than immediate interaction. Real-time systems use techniques such as level of detail, culling, texture compression, baked lighting and temporal reconstruction to achieve acceptable quality within limited frame budgets. Offline systems can use computationally expensive global illumination, path tracing and high-quality sampling. The distinction is primarily one of performance requirements and rendering workflow rather than a fundamentally different form of 3D geometry.
3D graphics is closely related to 3D modelling, animation and simulation but each describes a different activity. Modelling creates or modifies the geometric representation of an object. Rendering converts a scene representation into an image or other visual output. Animation changes scene state over time. Simulation computes behaviour according to physical or logical rules. A game character can therefore be modelled as a mesh, animated through a skeleton, simulated for collisions and rendered through a graphics pipeline. Confusing these concepts can obscure the structure of a graphics application because each is a separate component of the broader 3D production and runtime process.
Modern 3D graphics supports increasingly realistic and interactive environments through programmable shaders, physically based rendering, GPU acceleration, real-time ray tracing and advanced image reconstruction. Virtual and augmented reality add requirements for stereoscopic rendering, low latency and multiple viewpoints. Digital content can also be generated procedurally or through machine-learning techniques. Despite increasing hardware capability, 3D graphics remains constrained by geometric complexity, memory bandwidth, shader workload, lighting calculations and display resolution. Efficient 3D graphics therefore depends on choosing suitable representations and algorithms and on balancing visual fidelity against computational cost.
