Random-looking is not unrecoverable
Most generative art does not need true physical randomness. It needs a sequence that looks varied but can be recreated. A pseudo-random number generator turns a starting value—the seed—into a deterministic stream of numbers.
Every decision can consume the next number: position, hue, size, rotation, branch direction. Start again with the same seed and the stream repeats, so the image repeats too. The seed becomes a compact address for a much larger work.
artwork = render(algorithm version, seed, parameters)Design a deterministic pipeline
Create one seeded generator and pass it to every part of the composition. Avoid mixing it with Math.random(), the current time, unordered network results, or device-dependent input. Those hidden sources make a piece impossible to replay exactly.
Determinism also depends on call order. Adding one random color choice near the start shifts every later number. For larger projects, separate streams—for layout, palette, and texture—make revisions more stable.
- Normalize the seed into a known integer range.
- Keep algorithm and palette versions.
- Use seeded randomness for every stochastic choice.
- Record canvas size and all user parameters.
Explore a neighborhood of possibilities
Seeds are excellent discovery controls. Step from 120 to 121 to 122 and you sample distinct outcomes while keeping every other design choice fixed. Lock the seed while tuning composition; change only the seed when comparing variation.
A good generator constrains chance. It may select colors from a harmonious palette, keep shapes inside safe bounds, or bias size toward a useful range. Randomness supplies variation; the system supplies taste.
One seed, several visual systems
The same number can drive radically different algorithms. A crystal system may use angles and repeated polygons; an organic system may grow curved paths; a geometric system may divide a grid. The seed identifies an outcome only together with the algorithm.
This is why a trustworthy export should keep metadata nearby: tool name, version, seed, dimensions, and settings. A filename such as seed-art-v1-geometric-54321.png is far more useful than download-7.png.
Why reproducibility matters
For artists, it saves happy accidents. For teachers, every student can begin from the same example. For developers, a failing seed becomes a permanent test case. For collectors and collaborators, it makes authorship and editions easier to discuss.
Reproducibility does not remove surprise. It lets you return to the surprise, study it, and share the path that found it.
Turn the idea into an experiment.
The quickest way to understand a pattern is to change it and watch what happens.