Flow Field

A closer look

Locally aligned curves or short marks, gradual directional changes, controlled spacing, coherent bundles and deliberate areas of low density; colour reinforces the movement rather than hiding it.

A rule for movement, not a single signature look

In generative art, a flow field is a spatial arrangement of directions that can guide a drawing. Imagine placing a tiny compass at every point of a sheet: the compass tells a moving pen which way to go next. Many pens following the same underlying arrangement can produce a family of related paths. The result may resemble wind, hair, water or topography, but the resemblance does not mean the image measures any of those things.

Flow Field belongs here as a technique and visual approach, rather than a historical movement with one founder, palette or manifesto. A field may be described by a formula or sampled on a grid. The field itself is distinct from the marks used to reveal it. Long curves, tiny dashes, dots and broad bands can make very different pictures from similar directional information. AOI’s accompanying images interpret this visual language with image generation; they are not claimed to be outputs of a recorded vector-field program.

Technical context: tracing and steering

Tyler Hobbs’s 2020 essay Flow Fields explains a practical art-making method: store directions, choose starting positions and extend curves through repeated small steps. His examples show how line length, initial placement and field design change the character of the result. It is a useful artist’s account of the method, not evidence that flow fields began with that essay. The drawings reproduced in the source belong to Hobbs; the studies on this page are new AOI interpretations.

Daniel Shiffman’s The Nature of Code demonstrates a related approach using agents that read a local vector as their desired motion. Steering introduces an additional behaviour: an agent adjusts its velocity instead of necessarily following the local direction immediately. That distinction matters when describing an animation or trying to reproduce it. Direct curve tracing and agent steering are related tools, but their paths need not coincide.

What makes the image readable

Look first for agreement among neighbouring marks. When adjacent lines bend together, the eye reads a current. When their directions conflict without a visible reason, the image reads as tangled decoration. A convincing composition usually establishes one dominant journey across the frame, then interrupts it with a secondary turn, a widening gap or a quiet island. Those changes give the viewer a reason to keep looking after the initial impression of complexity.

Treat spacing as a compositional material. Closely packed lines create a dark mass even when each line is thin; a widening bundle can gradually reveal the background. A narrow pale channel between two dense regions may become the strongest shape in the picture. Before adding another colour, try changing where the marks begin and where they stop. Empty space should feel designed, not like an accidental hole left by an unfinished system. These are editorial design exercises, rather than claims about a universal algorithm.

Build a study one decision at a time

For a coded study, separate the field, the starting-point arrangement and the drawing rule. Hold two of those choices steady while changing the third. A useful first experiment is a uniform field, which exposes how line weight and density behave before any dramatic swirling begins. Then introduce one broad directional change. Keep a record of the program, parameters and any random seed if repeatability is part of the project; a compelling screenshot alone cannot reconstruct the process.

For an image-generation study, describe the visible result instead of pretending to specify executable mathematics. Ask for smooth locally aligned lines, a clear route through the image, a particular density contrast and a restricted palette. Inspect the result for stray crossings, inexplicable endpoints and curves that turn abruptly inside an otherwise smooth bundle. An AI illustration can communicate the aesthetic while failing the stricter test of whether every curve follows one mathematically consistent field. Label the result accordingly.

Palette, scale and useful variations

Begin with a strong value relationship: dark lines on pale paper, or pale lines on a dark ground. Reserve a brighter colour for a small number of paths so it describes emphasis rather than turning every mark into a competing event. Judge the image at its intended output size. Lines that look elegant at full resolution may merge into a flat grey area when reduced; large bundles and deliberate voids should remain legible in a gallery thumbnail.

The AOI studies explore three visual problems: curved turning regions, short directional marks and angular bands. The first emphasizes continuous movement, the second uses orientation to create texture, and the third tests what happens when directional changes become visibly faceted. A change of mark or turning behaviour is more informative than ten palette swaps. Try describing each study without mentioning its colours; if the descriptions are identical, the set probably needs more structural variety.

Applications and the limits of the metaphor

Flow-field imagery can suit abstract prints, editorial covers, motion identities and backgrounds that need a strong directional rhythm. It is particularly useful when an image should suggest coordination without depicting a literal object. Give text a genuinely quiet region, and keep important boundaries away from the busiest part of the field. Decorative complexity should support the communication rather than simply occupy the available pixels.

The common cliché is the same soft swirl repeated without a compositional decision. Another is to call any curving line pattern a simulation. A flow-field aesthetic does not by itself establish fluid dynamics, physical accuracy or data provenance. If an illustration represents measured wind or another real quantity, its directions, scales and source data require a separate verification process. For an abstract artwork, the better question is simpler: does the pattern guide attention, reward close inspection and still make a strong whole?

Pocket-sized facts

A flow field supplies local directions; the drawing rule determines how those directions become marks.

Directly tracing a field and steering an agent toward a field direction are related but distinct procedures.

The AOI images are AI-generated visual interpretations, not verified plots of an implemented field.

Examples worth exploring

Tyler Hobbs’s Flow Fields essay presents varied uses of the method and names the artworks shown; consult the original rather than assigning those titles to generated images.