Noise versus random in generative art

Two kinds of neighboring change

A jittering fence and a rolling hillside can use the same number of points but communicate very different worlds. The distinction comes from relationships between neighboring values. Independent random choices suit interruption; smoothly related values suit gradual change. Neither quality is automatically more artistic.

Sample two lines within one height band

Place evenly spaced horizontal sample positions across the canvas. For the first line, choose each vertical position independently. For the second, sample a smooth noise function at slowly increasing inputs. Keep the vertical amplitude identical so you are comparing relationships rather than merely different heights.

Separate noise input spacing from the other decisions so that a later revision has an identifiable cause. Create a plain reference version before adding expressive variation. Preserve that reference while revising, because memory is poor at comparing subtle differences across separate sessions. Compare the versions in the same viewing conditions. A useful acceptance check is whether neighboring changes suit the intended texture rather than accidental jitter. Judge that at the intended viewing size or working distance, where tiny construction mistakes and exaggerated effects can have very different consequences.

Separate input scale from visible amplitude

If the smooth line looks flat, increase the distance between input samples or enlarge the vertical range. If it looks agitated, reduce the input spacing. Do not confuse the input scale, which controls how rapidly values vary, with the output scale, which controls visible height.

Treat a disappointing result as evidence about noise input spacing, rather than immediately adding another effect. Temporarily show construction points and intermediate values. Reduce repeated elements to a handful and pause anything that changes over time.

Predict the missing points

Overlay only five sample points on each line and predict the missing shape between them. Then show all the points. This reveals whether your eye is relying on continuity, abrupt contrast, or a mixture of both.

Use 0.01, 0.05, and 0.2 per sample as three deliberately separated trial conditions. Keep the subject, framing, and main materials consistent across those trials. Note what changes first, what remains stable, and where the intended effect becomes excessive. Choose a version for an articulated reason. A useful note names a visible relationship that you could inspect again, rather than simply calling the result better.

Use the same vertical envelope

Put both comparison lines inside a band from y = 100 to y = 260. Convert each generator's result to that shared interval before judging it. Otherwise a line with smaller visible height may seem calmer simply because it occupies less space. Keep the number of sample positions identical too. The experiment should reveal the relationship between neighbors, with amplitude and sample count held deliberately constant.