Voronoi cells

How it works

Scatter points over a surface and give every spot to the point nearest to it. The regions that come out form a Voronoi diagram: irregular, convex cells that tile the plane with no gaps, like dried mud, giraffe skin, foam, cracked paint or the panes of a church window.

Cellular noise computes it per pixel. It places one point in each grid cell, at a random offset from a hash, and searches the pixel’s cell and its eight neighbors. That search returns three useful things, and each Voronoi look is a choice among them:

• F1, the distance to the nearest point: round bumps, scales and bubbles.
• F2 − F1, the second-nearest distance minus the nearest: zero on the borders, so a smoothstep on it draws walls, cracks and lead lines.
• The cell id, which point won: a flat value per region, for random colors, heights or materials per cell.

The pattern works in UV space on a quad or a mesh, and the same search in three dimensions (27 cells) works on world position, so a mesh without good UVs still gets seamless cells.

Every engine that ships Voronoi computes something a little different (another jitter, another distance metric, another range for the result), so a look tuned in one doesn’t carry over by copying numbers. The platform guides say what each one gives you.