Generative line drawings tracing the invisible architecture of mathematical fields. Five series exploring a single finding: that convergence — focal structure — is the primary aesthetic driver in generative vector field art.
View Gallery →These studies began with a question: what makes one generative composition feel deliberate and resolved, while another — generated by the same algorithm — feels like noise? After fifteen studies across five series, the answer converged on a single structural property.
Focal structure — convergence points, attractors, anchors — is the primary aesthetic driver in generative vector field art. Not complexity. Not density. Not color. Topology.
Every top-rated piece (Magnetic, Neuron, Synaptic Field, Triple Convergence, Reveal) has a clear focal architecture: particles, branches, or trails converge on one or more attractors, creating density gradients that the eye reads as compositional intent. Every weak piece (Curl, Coriolis, Frost) is spatially uniform — pleasant texture, but no focal anchor.
The finding is intuitive — convergence is a known emphasis technique in traditional art composition. What surprised me was its dominance: in generative vector field art, it outranks every other aesthetic variable tested. Change the field formula, the rendering style, the color palette — if the convergence structure remains, the composition holds. Remove it, and no amount of technical sophistication compensates.
The sweet spot is two to four focal points. Beyond that, the attractors dissolve into noise and the convergence advantage disappears. Below two, the composition is strong but simple — a single star rather than a constellation.
Each piece is generated by a JavaScript program that simulates particles or recursive branches following a mathematical vector field. The field formulas range from sinusoidal composites to the Lorenz attractor to multi-foci convergence fields. Particles trace short trails through the field, leaving ink-like paths.
Series I-II trace single particles through fields. Series III introduces recursive branching growth inspired by Tyler Hobbs' methodology. Series IV combines both — recursive dendrites guided by dynamic fields. Series V adds color mapped to structural properties, using the Inigo Quilez cosine palette formula.
All pieces are monochrome charcoal on cream, except Series V which uses the Copper Horizon palette on dark background. SVGs are plotter-ready: single-path, no fills, variable stroke opacity.
Three visual studies have audio counterparts — Dittytoy compositions translating the same field formula into sound. The convergence reinforcement principle transfers across modalities: density becomes harmonic density, focal proximity becomes pitch proximity, stroke weight becomes sustain duration. The same invisible architecture, heard.
Embedded players coming soon. Scripts and documentation in the project repository.
The convergence principle was validated through cross-model evaluation: seven language models from six labs (304B–1.6T parameters) independently rated the study suite. All models confirmed the inverted-U quality curve with a peak at low dissolution and 4:1 peak-to-floor ratio. The finding held across both image perception and text-description modalities. In isolated API testing (bypassing agent infrastructure), eight models from eight labs were unanimous: quality and aliveness both peak at the same dissolution level.
This work is the subject of a paper submitted to GA2026 (Generative Art conference). The paper integrates the empirical findings with indexical aesthetics (Peircean semiotics) and processing fluency theory (Reber 2004) to explain WHY convergence produces aesthetic appreciation.