Generative art ideas: build a family of seeds

Variation with a recoverable identity

Generative art does not require an image that changes forever. A useful first project produces a family of still images from one compact recipe. The interesting decision is which qualities belong to every family member and which details can change without dissolving that relationship.

Generate bounded offsets from one seed

Draw a regular field of small discs. Give each disc a random offset inside a limited radius, while keeping its nominal grid position. Set a seed before generating the offsets. Store the seed alongside the spacing and offset amount so you can reconstruct a result later.

Keep a visible note of offset radius while assembling the first version. Keep the ordinary state beside the most extreme intended state. The useful boundary is where the relationship remains understandable, rather than where the tool stops allowing a larger value. Check both ends before filling the middle. A useful acceptance check is whether each seed differs while the underlying field stays recognizable. Judge that at the intended viewing size or working distance, where tiny construction mistakes and exaggerated effects can have very different consequences.

Find accidental sequence restarts

Calling the seed setter repeatedly inside the disc loop can restart the sequence and make every disc behave alike. Set it once before the generation pass. If the image changes after resizing, check whether the number or order of random calls has changed.

Check one local instance of offset radius before assuming that the whole composition needs a different approach. Temporarily show construction points and intermediate values. Reduce repeated elements to a handful and pause anything that changes over time.

Compare six related constellations

Generate six seeds using identical dimensions and spacing. Choose a pair that feels distinct yet related. Explain the relationship in concrete terms such as shared density or bounded displacement; avoid treating novelty itself as the selection criterion.

Test a compact set of alternatives: 0, 6, and 18 pixels. Give each trial the same amount of finishing work so polish does not bias the comparison. Review them once as complete pieces and once while concentrating on the changing feature. If your preference reverses, note which artistic goal each version serves more effectively.

Record the random-call order

A seed identifies a random sequence, not a finished picture by itself. If you add a random choice for color before choosing positions, later positions can change because they consume different parts of that sequence. For a stable series, generate and store the positions first. Apply decorative choices in a separate pass, or save the complete generated data when exact reconstruction matters more than keeping the recipe compact.