Loading Blender-made assets...
Controls
Try: preset "Crystal growth" then "Single cell"; "Clouds" on a dense seed; switch Torus/Walls; drag to orbit (drag right moves the scene right).
Live data
Population history (last 200 generations)
The science

A 3D cellular automaton lives on a cubic lattice. Each cell has a state s in {0..C-1}; state 1 is "alive", states 2..C-1 are decay stages. The neighbour count n is the number of alive cells among the 26 Moore neighbours ((2r+1)^3 - 1, r=1) or the 6 von Neumann neighbours.

Update (Bays-style "S/B/C/N" notation): s'=1 if s=0 and n in B; s'=1 if s=1 and n in S; otherwise a live cell moves to state 2 (or 0 if C=2) and decay states advance s'=(s+1) mod C. All cells update simultaneously.

The 2D Life rule is B3/S23. Carter Bays found 3D analogues such as 4555 (S4-5, B5); unlike in 2D, there is no single canonical 3D Life. "445" is S4/B4 with 5 states; "Clouds" is S13-26/B13-14,17-19; "Amoeba" is S9-26/B5-7,12-13,15 (rule strings follow the community lists collected by Softology).

What is checked. The update function is exact integer logic, so the checks are exact: neighbourhood sizes, the single-seed birth count, the handshake identity (the sum of all neighbour counts equals 26 x the number of live cells on a torus), shift invariance, a still life (the 2x2x2 block under 6-7/5-6), and above all the dense typed-array engine is compared cell-for-cell with a second, independent sparse hash-map engine for 12 generations, for your current rule and for a 10-state decaying rule.

Honest simplifications: finite lattice (torus or dead walls), no continuous physics. The rules are explored empirically; interesting regimes are found by search, not derived. Cell colours are a visual choice. The cells are Blender-modelled gel cubes with baked textures, lit by a Blender-rendered studio environment; the lighting is cosmetic and does not enter the rules.

Made with 🟠 Blender · Three.js