worley: Cellular Distance Patterns

Grid-based cellular texture generation

Worley noise creates cellular patterns by calculating distances to randomly distributed points within a grid structure. The algorithm generates organic cell-like formations by evaluating proximity to nearest neighbors.

The function operates through nearest-neighbor analysis across a 3×3 (2D) or 3×3×3 (3D) grid:

$$W(p) = 1 - \min_{i \in N} d(p, s_i + r_i)$$

where:

• $p$ represents input coordinates
• $N$ denotes neighboring grid cells
• $s_i$ are grid cell positions
• $r_i$ are random offsets within each cell
• $d(\cdot, \cdot)$ calculates distance (default: Euclidean)

The algorithm maintains two distances: $F_1$ (nearest) and $F_2$ (second nearest), enabling various cellular effects through their combination.

Function Variations

Function	Purpose	Output
worley2Vec2(p)	2D distances	vec2(F1, F2) nearest distances
worleyVec2(p)	2D cellular	float inverted F1 distance
worley2Vec3(p)	3D distances	vec2(F1, F2) nearest distances
worleyVec3(p)	3D cellular	float inverted F1 distance

Distance Metrics

The implementation uses Euclidean distance by default but can accommodate other metrics:

Euclidean: $d = \sqrt{(x_2-x_1)^2 + (y_2-y_1)^2}$ Manhattan: $d = |x_2-x_1| + |y_2-y_1|$ Chebyshev: $d = \max(|x_2-x_1|, |y_2-y_1|)$

Live Editor

const fragment = () => {
      const coord = uv.mul(8)
      const cells = worleyVec2(coord)
      const edges = worley2Vec2(coord)
      const outline = edges.y.sub(edges.x).mul(8)
      return vec4(cells.mul(vec3(1, 0.8, 0.6)).add(outline.mul(0.3)), 1)
}

Result

Loading...

Live Editor

const fragment = () => {
      const scale = uv.mul(6)
      const w1 = worley2Vec2(scale)
      const w2 = worley2Vec2(scale.add(vec2(0.5)))
      const interference = w1.x.sub(w2.x).abs().mul(4)
      const gradient = vec3(interference, interference.mul(0.7), interference.mul(0.4))
      return vec4(gradient, 1)
}

Result

Loading...

Algorithm Structure

The Worley noise algorithm follows this computational pattern:

1. Grid Cell Identification: $n = \lfloor p \rfloor$, $f = fract(p)$
2. Neighbor Evaluation: Iterate through adjacent cells
3. Random Offset Generation: $o = random(n + g) \times jitter$
4. Distance Calculation: $d = distance(g + o, f)$
5. Nearest Point Tracking: Update F1, F2 distances
6. Result Inversion: Return $1 - F_1$ for cellular appearance

Jitter Control

The random offset multiplier (jitter) controls cell regularity:

• jitter = 0: Perfect grid arrangement
• jitter = 1: Full random displacement (default)
• jitter > 1: Exaggerated irregularity

Mathematical Properties

Worley noise exhibits several characteristics:

Bounded Output: Values range $[0, 1]$ for standard configurations Cell Continuity: Adjacent cells share boundary conditions Scale Invariance: Pattern maintains structure across scales Grid Alignment: Base structure follows integer grid spacing

Distance Combinations

Different F1/F2 combinations create varied effects:

Expression	Visual Effect
1 - F1	Solid cells
F2 - F1	Cell boundaries
F1 + F2	Inverted cells
F2 / F1	Edge emphasis

Implementation Details

The TSL implementation uses nested loops for neighbor evaluation:

• 2D: 9 cells (3×3 grid)
• 3D: 27 cells (3×3×3 cube)

Distance tracking maintains two variables:

• distF1: Nearest point distance
• distF2: Second nearest point distance

Random offsets employ consistent hashing for reproducible patterns while maintaining spatial coherence across grid boundaries.