Each piece starts from a figure assembled out of a few plain shapes: circles, boxes, capsules and triangles, merged with soft unions, bitten by subtractions or hollowed into shells. A signed distance field measures, at every point of the sheet, how far it lies from that figure’s outline. The series draws the field’s level lines in a fixed rhythm (a few lines, then a band) so that the figure echoes outwards to the edges of the paper and inwards to its core. The bands are left blank, ruled with hatching that turns where it crosses the outline, or filled flat. Then a few events, chosen by the hash, disturb the field locally: a seam where the lines step by a phase, a fold that creases them, a ripple that makes them wave. Press New seed for a fresh hash, then share the URL to show someone that exact piece.
Traits
The weights were set before any drawing code was written. They were then checked against a census of 2,000 seeds.
| Trait | Values (weight) |
|---|---|
| Palette | Graphite 22 · Riso 20 · Signal 18 · Ochre 14 · Verdigris 12 · Night 14 |
| Figure | Emblem 26 · Pair 20 · Constellation 24 · Lattice 15 · Frame 15 |
| Banding | Bare 20 · Hatched 35 · Cross-hatched 15 · Solid 30 |
| Event | Calm 10 · Seam 25 · Fold 25 · Ripple 25 · Mixed 15 |
| Spacing | Fine 30 · Medium 45 · Wide 25 |
Credits
- Figures: signed distance functions for the circle, box, rounded box, segment and equilateral triangle, the "onion" shell and smooth combinations, after Inigo Quilez’s articles 2D distance functions and smooth minimum, written here in TypeScript.
- Level lines: marching squares, the 2D case of Lorensen & Cline’s marching cubes (1987), with linear interpolation along cell edges and saddles resolved by the cell average. The bands are filled pixel by pixel from the same grid, antialiased by the local gradient.
- Randomness: sfc32 by Chris Doty-Humphrey. The hash seeding, named sub-streams and verification follow Camille Roux’s genart skill (MIT).
- The palettes are original. The code was written with an AI assistant. It draws everything itself: no AI-generated images.