Level of Detail

Level of detail is the practice of rendering an object at a fidelity matched to how much it actually contributes to the final image, rather than at its maximum quality regardless of context. A tree a few meters from the camera gets a full polygonal mesh with individual leaves; the same tree a kilometer away gets a few hundred triangles or a flat billboard, because no viewer can tell the difference at that distance and resolution.

The simplest scheme is discrete LOD: an artist or a decimation tool prebakes several mesh versions at descending triangle counts, and the engine swaps between them at distance thresholds. The classic failure mode is popping — a visible snap as the mesh suddenly changes silhouette or shading at the swap boundary — mitigated by cross-fading between levels or by adding hysteresis, switching at a slightly different distance going out than coming back, so an object hovering near the threshold doesn't flicker between levels every frame.

Continuous LOD avoids discrete swaps by progressively simplifying or refining a single mesh, collapsing or splitting edges smoothly as distance changes — harder to author but eliminates popping entirely. Modern engines increasingly use virtualized geometry (Unreal's Nanite is the best-known example), which streams and rasterizes only the mesh detail a pixel can actually resolve, blurring the line between LOD and continuous streaming.

The same idea generalizes past meshes: impostors replace a whole complex object with a flat, camera-facing billboard rendered from a precomputed image, and audio, physics, and AI systems apply their own versions of level of detail — a distant NPC gets a cruder simulation because no one is close enough to notice the difference.