Each flower starts from a single petal shape, round, pointed, notched, fringed or long like a ribbon, repeated three to twelve times around a centre, in one to four whorls. Nothing here is drawn with a brush: every petal is a small polygon whose edges are split again and again, each new point nudged by a random amount, and the result is filled dozens of times at a few percent opacity, each layer deformed anew. The layers agree in the middle and disagree at the rim, so petals get a dense core and a soft, bleeding edge, and neighbouring petals blend where their layers overlap. Stamens, seeds, leaves, stems, loose petals and the washes on the paper are painted the same way. A piece holds one flower, a few on stems, a bouquet or a printed field of them. 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 | Peony 22 · Iris 18 · Marigold 18 · Camellia 14 · Celadon 14 · Nocturne 14 |
| Symmetry | Three 10 · Five 32 · Six 22 · Eight 20 · Twelve 16 |
| Petal | Round 26 · Pointed 24 · Notched 20 · Fringed 16 · Ribbon 14 |
| Whorls | Single 35 · Double 40 · Full 25 |
| Composition | Solo 34 · Stemmed 24 · Bouquet 24 · Scatter 18 |
Credits
- Layered translucent polygons: the watercolour technique described by Tyler Hobbs in A Guide to Simulating Watercolor Paint with Generative Art (2017): a base shape deformed a few times, then many nearly transparent layers, each deformed again, with more variation where the edge should be soft. The code here is an independent implementation applied to petals, with per-vertex variance growing from the petal base to its tip.
- Deformation: recursive midpoint displacement with roughly Gaussian offsets (a sum of three uniform draws) proportional to the edge length and to a per-vertex variance, after Fournier, Fussell & Carpenter, "Computer Rendering of Stochastic Models" (Communications of the ACM 25, 1982).
- Further reading on the same technique: Sighack’s tutorial Generative Watercolor in Processing.
- Randomness: sfc32 by Chris Doty-Humphrey, seeded from the hash with named sub-streams, following Camille Roux’s genart skill (MIT).
- The flowers and palettes are original. The code was written with an AI assistant. It draws everything itself: no AI-generated images.