Loading field map
Computer graphics has always been pulled between two opposing demands: the desire to create images that are indistinguishable from reality, and the need to produce them fast enough to be useful. A third pressure—the need to communicate clearly rather than realistically—adds a further dimension. The ten frameworks that make up the field's history each represent a different answer to these tensions, and the relationships among them—supersession, competition, convergence, and coexistence—tell the real story of how computer graphics became what it is today.
The earliest computer displays drew images as collections of lines. Vector Graphics (1963–1985) stored pictures as mathematical curves and instructions for a beam to trace them on a cathode-ray tube. Ivan Sutherland's Sketchpad (1963) demonstrated that a user could manipulate geometric shapes directly on a screen, but the images were wireframes—no filled surfaces, no shading, no depth. Vector graphics was elegant and memory-efficient, but it could not produce the continuous tones of a photograph.
Raster Graphics (1969–Present) replaced the line-drawing model with a rectangular grid of pixels, each storing a color value. The shift was not merely a technical improvement; it was a change in what an image was. A raster image could represent any continuous-tone picture, but it required far more memory and processing power. As memory costs dropped and display resolution improved through the 1970s and 1980s, raster graphics superseded vector graphics for almost all applications. Vector graphics did not disappear entirely—it survives in typography, CAD wireframes, and SVG—but it became a specialized tool rather than the dominant paradigm.
Two frameworks that are not rendering methods in their own right nonetheless provide the substrate on which all other frameworks depend. Animation (1964–Present) is the study of how to represent and compute change over time. Early work, such as the 1964 film The Story of a Computer by Edward Zajac, showed that computers could generate moving images by interpolating between keyframes. Over decades, animation developed techniques for skeletal rigging, inverse kinematics, particle systems, and physics-based motion. Today, animation is not a single method but a collection of approaches that coexist: keyframe animation for authored motion, motion capture for recorded performance, and physics simulation for emergent behavior. The sibling subfield of Physical Simulation has increasingly fed into animation, providing methods for cloth, fluids, and deformable bodies that would be impossible to keyframe by hand.
Geometric Modeling (1964–Present) addresses the representation of three-dimensional shape. Early work used polygonal meshes and parametric surfaces such as Bézier patches and NURBS. Later developments added subdivision surfaces, implicit surfaces, and signed distance fields. Every rendering framework—whether it uses rasterization, ray tracing, or volume rendering—depends on a geometric representation to define what exists in the scene. Geometric modeling is not a competitor to rendering frameworks; it is the infrastructure that makes them possible. The choice of representation (polygons vs. NURBS vs. voxels) directly constrains what a rendering algorithm can do.
The most visible progression in computer graphics is the sequence of frameworks that model how light interacts with surfaces. Local Illumination (1971–1990) computed the color of a surface point using only the direct light arriving from each light source and the surface's orientation. Gouraud shading (1971) interpolated vertex colors across polygon faces, and Phong shading (1975) interpolated surface normals for smoother highlights. These methods were fast and produced plausible results, but they ignored a fundamental physical fact: light bounces. A red wall casts a red tint onto a nearby white object, and shadows are not just absences of light but structured patterns of occlusion. Local illumination could not capture any of this.
Global Illumination (1984–Present) addressed the missing physics by modeling the full transport of light through a scene. James Kajiya's rendering equation (1986) formalized this as an integral over all directions from which light can arrive at a surface point, including light that has bounced off other surfaces. Radiosity (1984) solved a simplified version for diffuse interreflection, and path tracing (1986) used Monte Carlo integration to sample the full integral. Global illumination produced images of unprecedented realism, but at a staggering computational cost. A single frame could take hours or days to render.
Physically-Based Rendering (PBR) (2004–Present) did not replace global illumination so much as standardize and systematize it. PBR is a framework that commits to physically accurate material models—microfacet BRDFs, energy-conserving reflectance, and measured or simulated surface properties—combined with physically based light transport. The landmark textbook Physically Based Rendering: From Theory to Implementation (2004) by Pharr, Humphreys, and Jakob provided both the theory and an open-source implementation (pbrt) that became the de facto reference. PBR derives directly from global illumination, narrowing the earlier framework's broad space of possible algorithms to a specific set of physically grounded choices. It also absorbed techniques from Volumetric Rendering to handle participating media such as smoke, fog, and subsurface scattering.
While global illumination and PBR pursued physical accuracy, a separate framework pursued speed. Real-Time Rendering (1980–Present) is defined by a hard constraint: the image must be generated at interactive frame rates, typically 30 or 60 frames per second. For most of its history, real-time rendering relied on rasterization—the process of projecting triangles onto the screen and shading each pixel with local illumination. The result was fast but visibly less realistic than offline global illumination.
For decades, Real-Time Rendering and Global Illumination were in direct competition. Real-time systems sacrificed physical accuracy for speed; offline systems sacrificed speed for accuracy. The boundary began to blur in the 2010s as GPU hardware became powerful enough to run simplified path tracing in real time. NVIDIA's RTX architecture (2018) introduced dedicated ray-tracing cores, and game engines such as Unreal Engine 5 adopted real-time global illumination techniques like Lumen. Today, the two frameworks are converging: real-time rendering increasingly incorporates PBR material models and global illumination approximations, while offline rendering has become fast enough for interactive preview. The competition has not ended—trade-offs remain—but the sharp divide has softened into a continuum.
Two frameworks explicitly rejected core assumptions of the mainstream lineage and found durable specialized roles. Volumetric Rendering (1988–Present) rejected the assumption that a scene consists of surfaces. Medical imaging, scientific visualization, and special effects all require the display of data that has no clear surface—clouds, smoke, flames, CT scans. Marc Levoy's 1988 paper "Display of Surfaces from Volume Data" showed how to render volumetric data directly by accumulating color and opacity along rays through a 3D grid. Volumetric rendering coexists with surface-based rendering rather than replacing it; each is suited to different data. Over time, PBR absorbed volumetric techniques for participating media, but pure volume rendering remains essential for scientific and medical visualization.
Non-Photorealistic Rendering (NPR) (1990–Present) rejected the goal of photorealism itself. Why make an image look like a photograph when a line drawing, a watercolor, or a cartoon can communicate more clearly? Takafumi Saito's 1990 paper "Comprehensible Rendering of 3-D Shapes" demonstrated that line drawings could convey shape more effectively than shaded renderings for technical illustration. NPR and PBR are explicit competitors: both produce final images, but they pursue opposite fidelity goals. PBR aims to match the physical behavior of light; NPR aims to match the communicative conventions of human artists. The field sustains both because they serve different purposes. A medical textbook needs NPR-style cutaway drawings; a product visualization needs PBR's material accuracy. NPR has also influenced PBR indirectly—stylized rendering in games often blends physically based materials with non-photorealistic outlines or cel shading.
The leading frameworks today are PBR and Real-Time Rendering, which are in the process of merging into a single real-time physically based pipeline. Global Illumination remains the theoretical foundation, but its algorithms are increasingly deployed in real-time form. Animation and Geometric Modeling continue as essential infrastructure, evolving alongside rendering rather than competing with it. Volumetric Rendering holds a stable niche in visualization and effects. NPR persists as a living alternative, maintained by a community that values communicative clarity over physical accuracy.
The major agreement across active frameworks is that physically based material models are the right foundation—even NPR often starts from PBR materials and then stylizes the output. The major disagreement is about what the image is for: PBR and real-time rendering optimize for perceptual realism, while NPR optimizes for information transfer or artistic expression. A second tension runs between generality and specialization: PBR aims to handle all materials with one unified model, while NPR embraces a toolkit of specialized techniques for different visual styles. These disagreements are not signs of fragmentation; they are the productive tensions that keep the field moving.