Rendering (3D): Generating Images From 3D Models, From Rasterization to Real-Time Path Tracing

In 2026, Capcom shipped two games that run full path tracing in real time. Resident Evil Requiem and PRAGMATA both use Capcom's RE ENGINE with a custom path tracer that handles direct lighting, indirect lighting, and reflections through a single unified pipeline. This is the same rendering technique that Pixar uses for feature films, where each frame takes minutes or hours to compute. Capcom's version runs at interactive frame rates on consumer GPUs. The development took approximately 18 months and required rebuilding the engine's lighting pipeline from the ground up. At GDC 2026, Capcom and NVIDIA presented a technical deep dive into the implementation, describing how they validated their reference path tracer against offline DCC tools, then scaled it down to a game-optimized version using NVIDIA's DLSS Ray Reconstruction for denoising. The presentation is available on the NVIDIA Game Developer YouTube channel. For a rendering technique that was considered too expensive for real-time use just five years ago, path tracing is now shipping in consumer games. That transition is the latest chapter in a 50-year history of making computers draw images from 3D data.

Rendering is the process of generating a two-dimensional image from a three-dimensional model. A 3D model is a mathematical description of objects in virtual space: their geometry, their surface properties, their position relative to a camera. Rendering takes that description and computes what the camera would see. It simulates how light interacts with the objects, how surfaces reflect or absorb light, how shadows form, how colors appear under different lighting conditions. The output is a 2D image, typically a grid of pixels, that represents the 3D scene as viewed from the camera's perspective.

This entry covers the history of 3D rendering from the first rasterized images to real-time path tracing, the researchers who defined the field, and how 2026 developments are collapsing the gap between offline and real-time rendering quality.

What Is Rendering and How Does It Work?

There are two fundamental approaches to 3D rendering: rasterization and ray tracing. Rasterization converts 3D geometry into 2D pixels by projecting triangles onto the screen and filling them. For each triangle, the renderer determines which pixels it covers, computes the color of each pixel based on the triangle's material and the scene's lighting, and writes the result to the framebuffer. Rasterization is fast because it works triangle by triangle, and GPUs are highly optimized for this workflow. It is the rendering method used in virtually all real-time applications, from games to interactive 3D web content. The limitation of rasterization is that it does not naturally simulate complex light behavior. Shadows, reflections, and refractions require separate techniques (shadow maps, reflection probes, screen-space effects) that approximate rather than simulate physical light.

Ray tracing simulates light by tracing rays through the scene. For each pixel in the output image, the renderer shoots a ray from the camera through the pixel and into the 3D scene. When the ray hits a surface, it computes the surface's color based on the material and the light sources. To compute shadows, it shoots additional rays from the surface toward each light source to check for occluders. To compute reflections, it bounces the ray off the surface and continues tracing. To compute refractions (for glass or water), it bends the ray according to Snell's law. Ray tracing produces physically accurate images because it follows the actual behavior of light, but it is computationally expensive because each pixel may require dozens or hundreds of rays.

Path tracing is a variant of ray tracing that fully simulates global illumination. Instead of shooting separate rays for shadows, reflections, and refractions, path tracing shoots a single ray from the camera and lets it bounce through the scene randomly, accumulating light at each bounce. The process is repeated many times per pixel, and the results are averaged. With enough samples, the image converges to a physically correct result that includes indirect lighting, color bleeding, soft shadows, and caustics. Path tracing is the gold standard for image quality. It is what offline renderers like Pixar's RenderMan, Arnold, and Blender's Cycles use. The cost is time: a single frame of a Pixar film may take hours to render.

Origins and History

The Foundations (1960s-1980s)

The first computer-rendered 3D images were rasterized wireframes. Ed Catmull's PhD work at the University of Utah (1974) introduced the Z-buffer, which solved the hidden surface problem by storing the depth of each pixel and keeping only the nearest surface. This made rasterization of solid 3D objects practical for the first time. Catmull also introduced texture mapping, which applies 2D images to 3D surfaces, and subdivision surfaces, which render smooth curved surfaces as polygon meshes.

Ray tracing was introduced by Arthur Appel in 1968, who described a method for casting rays from the camera through pixels to determine visible surfaces and compute shadows. Turner Whitted, at Bell Labs, extended this in 1980 with recursive ray tracing, which bounced rays off reflective and refractive surfaces. Whitted's paper, "An Improved Illumination Model for Shaded Display," is one of the most cited papers in computer graphics. His images of glass spheres reflecting and refracting each other demonstrated that ray tracing could produce images that rasterization could not match.

Path tracing was introduced by James Kajiya in 1986 in a paper titled "The Rendering Equation." Kajiya formulated the rendering of 3D scenes as a single integral equation that describes all light transport, and he showed that path tracing (Monte Carlo sampling of light paths) could solve it. Kajiya's rendering equation is the mathematical foundation of all physically based rendering. Every modern path tracer, from RenderMan to Cycles to the RE ENGINE path tracer, solves a version of Kajiya's equation.

The Film Era (1980s-2000s)

Pixar's RenderMan, first released in 1989, became the standard renderer for CGI film production. RenderMan used a Reyes (Render Everything You Ever Saw) algorithm for rasterization, combined with ray tracing for reflections and refractions. Pixar films from Toy Story (1995) onward were rendered with RenderMan. The renderer evolved over decades, adding path tracing support in RenderMan 19 (2015).

Arnold, developed by Marcos Fajardo in the late 1990s and commercialized by Solid Angle in 2010, was designed as a brute-force path tracer from the start. Arnold became the standard renderer for visual effects studios including ILM, Sony Pictures Imageworks, and Weta Digital. Films including Gravity (2013), Arrival (2016), and Dune (2021) were rendered with Arnold.

The Real-Time Era (2000s-2026)

Real-time rendering used rasterization exclusively until 2018, when NVIDIA introduced RTX hardware that accelerated ray tracing on consumer GPUs. The RTX platform made it practical to trace rays in real time for reflections, shadows, and global illumination, though these were still selective additions to a rasterized base. Games like Control (2019), Cyberpunk 2077 (2020), and Minecraft (with RTX, 2020) used RTX ray tracing for specific effects.

Full path tracing in real time became practical in the 2020s with the combination of RTX hardware and AI-based denoising. NVIDIA's DLSS Ray Reconstruction, introduced with DLSS 3.5 in 2023, uses a neural network to denoise path-traced images in real time, allowing fewer samples per pixel while maintaining image quality. This is what made Capcom's RE ENGINE path tracer possible.

Key Researchers and Pioneers

Ed Catmull (b. 1945)

Catmull's 1974 PhD thesis at the University of Utah introduced the Z-buffer, texture mapping, and subdivision surfaces. These three innovations are used in every real-time renderer today. Catmull co-founded Pixar and oversaw the development of RenderMan. He received the Turing Award in 2019.

Turner Whitted (b. 1950)

Whitted's 1980 paper on recursive ray tracing established that ray tracing could simulate reflections and refractions by bouncing rays through a scene. His images of glass spheres were the first to show that computer graphics could produce images indistinguishable from photographs for certain materials. Whitted worked at Bell Labs and later at NVIDIA Research.

James Kajiya (b. 1952)

Kajiya's 1986 paper "The Rendering Equation" formulated all light transport as a single integral equation and showed that path tracing could solve it. This paper is the theoretical foundation of physically based rendering. Every path tracer in use today, from film renderers to game engines, traces its lineage to Kajiya's equation.

Marcos Fajardo (b. 1972)

Fajardo developed Arnold, the first widely adopted brute-force path tracer for film production. Arnold's design philosophy was simple: trace enough rays and the image will be correct. This was considered impractical in the late 1990s, but as computing power increased, Arnold's approach became the industry standard. Fajardo's work proved that brute-force physical simulation could replace the clever but approximate techniques that dominated film rendering.

Real-Time Path Tracing in 2026

The most significant development in 3D rendering in 2026 is the arrival of full path tracing in shipping games. Capcom's RE ENGINE path tracer, developed over approximately 18 months by a team led by Kenta Nakamoto and Kosuke Nabata, handles both direct and indirect lighting through a single path-tracing pipeline. The system uses NVIDIA's RTX Kit for ray tracing acceleration and DLSS Ray Reconstruction for denoising. Capcom also implemented a custom ReSTIR GI technique to stabilize indirect lighting by reusing path samples across frames. The results were presented at GDC 2026. Read more at NVIDIA Developer Blog.

At SIGGRAPH 2026, NVIDIA presented DLSS 5, the next generation of its neural rendering technology. Unlike earlier DLSS versions that upscaled images, DLSS 5 uses a compressed one-step pixel-space diffusion model to enhance rendered frames in real time. The model takes the engine's rendered frame as input and produces a more realistic output, using motion vectors from the engine for temporal stability. NVIDIA demonstrated 4K 60+ FPS with path tracing on a single GPU. The key innovation is that DLSS 5 preserves artistic intention by working from the engine's own output rather than generating images from scratch. Read more at WCCFTech.

Also at SIGGRAPH 2026, researchers presented Gaussian Point Splatting, a stochastic rendering technique that scales to hundreds of millions of Gaussians in real time. Instead of blending each Gaussian across multiple pixels, the method samples a single pixel-sized opaque point from each Gaussian and splats it using 64-bit atomic operations. This reduces per-Gaussian work to a single write, making rendering cost nearly independent of Gaussian size or overlap. The paper includes a Shadertoy demo for opacity correction. Read more in ACM Transactions on Graphics.

Disney Research Studios presented a Neural Render Proxy at EGSR 2026 that enables interactive relighting of static scenes. The system decouples rendering into path sampling and emission computation, training a scene-specific neural network that learns how light is transported from any scene location to any image pixel. This allows lighting changes at 30 to 60 Hz while closely approximating path-traced quality. Read more at Disney Research.

Reallusion launched the RTX Render open beta in July 2026, bringing NVIDIA RTX Path Tracing to iClone 8 and Character Creator 5. The system includes physically accurate lighting, an RTX material system with transmission, coat, sheen, and hair controls, real f-stop depth of field, and a dynamic procedural sky. Read more at Reallusion Magazine.

Related Terms

Rendering is the second step in the 3D pipeline, after 3D modeling. The appearance of rendered surfaces is determined by shaders, which are programs that compute how light interacts with materials. Rendering connects to digital art as the process that makes 3D art visible. It relates to procedural generation when rendering is used to visualize procedurally generated content. For more on digital art, read our post on the history of digital art from pixel to prompt, or explore our guide to digital art.

See Rendering in Practice

Every CGI film and 3D game is a rendering demonstration. To see the difference between rasterization and path tracing, compare a game running with ray tracing off and on. The reflections, shadows, and indirect lighting change dramatically. For the highest quality offline rendering, look at recent Pixar films like Inside Out 2 (2024) or any film rendered with Arnold, such as Dune: Part Two (2024). The SIGGRAPH conference, held annually, is the premier venue for new rendering research, and the SIGGRAPH 2026 technical papers sessions include the latest work on real-time path tracing, neural rendering, and Gaussian splatting.

For more on the 3D pipeline, read our entries on 3D modeling and shaders, or explore our post on the history of digital art.