A History of Voxel Art: From Comanche to Minecraft and MagicaVoxel

The word “voxel” is formed from volume element, by analogy with “pixel.” A voxel can store density, material, colour, occupancy, or another value at a position in a 3D grid.

CT and MRI workflows are natural examples: a series of sampled slices forms a volume, and each cell can hold a measurement. A voxel is not inherently a visible cube. The cube is a convenient way to display or edit a cell.

NovaLogic's Comanche: Maximum Overkill became a well-known early use of “voxel” terrain in a commercial game. Its Voxel Space approach combined a height map, a colour map, and view-dependent sampling to render rolling terrain efficiently on contemporary CPUs.

This was not a fully general 3D grid in which arbitrary blocks could be stacked at every coordinate. It was closer to a 2.5D height field, but its smooth valleys and hills looked very different from the sparse polygonal landscapes of many early-1990s games.

As consumer GPUs accelerated polygon rendering, height-field voxel engines became less prominent. The idea returned in a different form: worlds made from discrete cells that players could modify.

Minecraft, first released publicly in 2009, made placing and removing blocks the central interaction. Its worlds are divided into chunks so that terrain can be generated, stored, meshed, and rendered in manageable regions. The blocks are usually converted into polygon meshes for the GPU; calling the visual style “voxel” describes the grid-based world model rather than a requirement to render raw volume samples.

Microsoft announced its agreement to acquire Mojang and the Minecraft franchise for US$2.5 billion in September 2014, illustrating how far the block-world idea had moved from its independent origins.

Games including 3D Dot Game Heroes, Cube World, Crossy Road, and Trove helped establish cube-based modelling as a deliberate art style rather than merely a rendering shortcut.

The constraint is productive: artists build readable silhouettes, colour groups, and lighting from a limited grid, much as pixel artists work with a limited 2D raster. The results can be recognisable at small sizes while remaining easy to rotate and animate in 3D.

MagicaVoxel by ephtracy provides a focused editor and renderer for voxel art. Its documented .vox format uses a chunked binary structure. Common chunks include:

VOX header
MAIN
├─ SIZE  model dimensions
├─ XYZI  voxel positions and palette indices
└─ RGBA  optional 256-colour palette

In an XYZI entry, x, y, z, and the colour index each occupy one byte. Later format versions also support scene graphs, transforms, groups, materials, and other chunks. Implementers should follow the current format document rather than assuming every file contains only one SIZE/XYZI pair.

Drawing one complete cube for every occupied cell wastes work: interior faces can never be seen, and adjacent coplanar faces can often be merged.

Greedy meshing scans the visible faces along each axis and combines compatible neighbours into larger rectangles. One rectangle can be emitted as two triangles. The result depends strongly on the model, but large flat surfaces can be represented with far fewer quads than the naive one-cube-per-voxel method.

1. Discard faces between two occupied neighbouring cells.
2. Build a 2D mask for the remaining faces on each slice.
3. Merge compatible cells into maximal rectangles.
4. Emit each rectangle as a quad with the correct material and orientation.

A voxel editor can export a mesh to formats such as OBJ or glTF/GLB. Khronos defines glTF as a compact, runtime-neutral 3D asset transmission format for native and web engines. It can represent complete scenes with nodes, transforms, meshes, materials, cameras, and animation.

A common workflow is to create a model in a voxel editor, generate an optimised surface mesh, export GLB, and load it into Blender, a game engine, or a WebGL application. The voxel grid remains the authoring model even when the delivered asset is a conventional triangle mesh.

• Voxels began as cells in sampled 3D volume data, not as an art style.
• Comanche popularised Voxel Space height-field terrain in 1992.
• Minecraft made editable block worlds a global visual language.
• MagicaVoxel offers a focused editor and a widely supported chunked .vox format.
• Greedy meshing removes hidden faces and merges visible cells into larger quads.
• glTF/GLB provides an interoperable way to deliver the resulting 3D assets.

Try building a model in NanToo's online Voxel Maker, which can save projects and export VOX, OBJ, GLB, or an animated GIF in the browser.