Generative Art

Generative art is art (or texture, or terrain, or a whole city) produced by a system — an algorithm, a set of rules, a random process — rather than drawn stroke by stroke by hand. The artist's job shifts from placing every pixel to designing the rules and then choosing which outputs of the system to keep: tune a parameter, reseed the random number generator, and a genuinely different piece appears. This hub gathers every interactive procedural generation simulation on mysimulator.uk into one guided starting point, so instead of reading about noise functions and grammars in the abstract you can drag a slider and watch the algorithm redraw itself in real time in your browser.

The most widely used building block in generative art is Perlin noise — a smooth, coherent kind of pseudo-randomness invented for the original Tron film to make computer-generated textures look organic instead of static-like. Layering several octaves of Perlin noise at different frequencies (fractional Brownian motion, or fBm) produces the rolling terrain heightmaps, cloud textures and wood-grain patterns behind countless games and digital art pieces. The flow-fields simulation puts that same noise field to work as a vector field, tracing 3000 particles through it to build brushstroke-like patterns reminiscent of Van Gogh's Starry Night, while domain warping takes noise a step further by feeding it back into its own coordinates — f(p + f(p + f(p))) — turning flat noise into the swirling, marbled fields behind most "digital painting" generative art.

A second major family of techniques builds structure through recursive rewriting rather than noise. L-systems (Lindenmayer systems) start from a single symbol and a handful of production rules — replace "F" with "F+F−F−F+F", say — and after a few iterations of rewriting, a turtle-graphics interpreter turns the resulting string into a Koch snowflake, a dragon curve, or a startlingly plant-like fern. Iterated Function Systems (IFS) achieve the same self-similar complexity a different way, applying a small set of affine transformations to a point millions of times until a fractal attractor — the Barnsley fern, the Sierpiński triangle — emerges from what looks at first like scattered noise. The Mandelbrot and Julia set explorers push this idea into the complex plane: iterate z → z² + c and colour each pixel by how quickly it escapes to infinity, and you get a boundary of genuinely infinite detail that keeps revealing new structure no matter how far you zoom in.

A third family generates order from purely local rules applied simultaneously across a grid or a swarm. Rule 30 and the other elementary cellular automata update each cell from its two neighbours according to a fixed lookup table, yet some rules (30 is famously one of them) produce output complex enough that Wolfram has argued it may be genuinely unpredictable — no shortcut exists to know row 10,000 without computing every row before it. Diffusion-limited aggregation and Voronoi growth model physical processes — particles random-walking until they stick to a cluster, crystals nucleating and growing outward — and produce the same branching dendrites and mosaic textures you can find in frost patterns, mineral deposits and bacterial colonies. Phyllotaxis shows the same idea driven by a single "magic" number: place seeds at the golden angle (137.5°) from the centre outward and Fibonacci spiral arms snap into place automatically, exactly as they do in a real sunflower head — no other angle packs seeds as efficiently.

The generative-city simulation and the Voronoi stippling technique combine several of these ideas into complete pipelines: recursive block subdivision lays out roads and zoned parcels the way an L-system lays out branches, then each parcel is filled and extruded procedurally, while weighted Voronoi stippling uses Lloyd's relaxation algorithm to let thousands of points self-organise into pointillist portraits driven by an image's own density function. Follow the learning path below for a suggested order that moves from noise to grammars to complex-plane fractals to a full procedural pipeline, browse the full grid for anything that catches your eye, or jump to the Generative Art category page for the complete list.

What makes these simulations different from a static gallery of generative art is that every one of them is a live algorithm running in your browser, not a pre-rendered image. Change the octave count on the Perlin noise simulation and the terrain genuinely regenerates from different mathematics, not a different picture from a lookup table; change the production rule on the L-system simulation and turtle graphics genuinely re-interpret a different string. That distinction matters for anyone who wants to actually learn how the algorithm works rather than just admire a finished piece — because tuning the parameter and watching the output change is the fastest way to build real intuition for coherent noise, recursive grammars and local update rules alike.