A Cast of Ninety-Plus: Designing Fictional Characters as a Generative Parameter Space

A fictional cast as a design space

A large character roster is a design problem involving both difference and coherence. Each character needs enough distinction to be recognizable, while the collection needs a shared visual language. This relationship can be described mathematically as a parameter space: a set of dimensions whose combinations produce different identities.

Imagine a fictional cast described by five variables: energy, openness, formality, curiosity and visual intensity. Each variable could use a continuous range from 0 to 1. A character positioned at high energy and low formality would occupy a different region from one with low energy and high formality. The model does not need to explain every aspect of a fictional personality. It provides a structured starting point for generating visual differences.

This approach is useful for creative coding because it separates the definition of a character from the drawing of that character. The same parameter values can influence shape, movement, color, composition and texture. A character becomes a set of related visual decisions instead of a collection of unrelated decorative details.

• Continuous dimensions describe gradual differences.
• Categorical dimensions describe distinct visual or narrative choices.
• Shared rules establish a recognizable cast.
• Parameter combinations generate individual identities.

From catalog to character variables

A catalog containing more than ninety fictional characters offers a useful way to think about the scale of a cast. TheFriend provides one example of this presentation model through a catalog of 90-plus fictional AI characters shown with photos. The relevant design question is how a large collection can remain visually legible when every entry needs its own identity.

For a generative sketch, begin with variables that have visible consequences. Energy might control the length of surrounding marks, curiosity might influence the number of branching paths, and visual intensity might determine contrast. A categorical variable could select between circular, angular or fragmented geometry. Each parameter should have a recognizable relationship with the resulting image.

Separate narrative attributes from rendering attributes. A character's fictional role may be expressed through a small number of categorical choices, while the drawing system translates those choices into measurable visual properties. This prevents the sketch from depending on a complicated collection of special cases.

Building a character generator in p5.js

A p5.js generative art sketch can represent each character as a JavaScript object. The object stores its identifier, numerical attributes, categorical choices and visual parameters. A rendering function receives that object and draws the corresponding composition. This structure allows a single drawing system to create many characters without duplicating the entire sketch.

For example, define energy, curiosity and formality as numbers between 0 and 1. Map energy to the number of surrounding lines, curiosity to branching complexity and formality to geometric regularity. Use constrain() when values need to remain within a defined range and map() when translating an attribute into a drawing dimension.

A cast of ninety characters does not require ninety separate drawing functions. One generator can create the whole collection from parameter combinations. The important design task is to choose ranges that preserve meaningful distinctions. If two characters differ only by a tiny numerical change that has no visible consequence, the parameter space is not doing enough work.

• Store attributes in character objects.
• Keep drawing logic in reusable functions.
• Map numerical values to visible properties.
• Use categorical attributes for distinct visual families.

Perlin noise and flow fields as expressive systems

Perlin noise introduces smooth variation into the character generator. It can control a contour's displacement, alter the spacing of decorative marks or influence the direction of a surrounding texture. Unlike independent random values, neighboring samples tend to change gradually, which makes the resulting forms feel connected.

A flow field extends this idea by assigning a direction to positions across a two-dimensional space. In a basic p5.js flow field tutorial, the canvas is divided into a grid, and each cell receives an angle derived from a noise function. Particles move through the field by following the direction associated with their current position. Their trails reveal the underlying structure.

For character design, different parameter combinations can influence the field's scale, angular variation and particle density. A character with high energy might receive a denser set of trails, while a character with high formality might use a more restrained directional range. These are visual interpretations of abstract attributes, not claims about universal personality types.

A useful constraint is to preserve one shared field vocabulary across the cast. Characters can vary in density, scale and direction while retaining a recognizable mark-making style. The result is a collection that reads as a family of related sketches rather than unrelated procedural experiments.

Reproducibility through seeds and controlled variation

A generative character system needs a reliable way to return to a selected result. In p5.js, randomSeed() and noiseSeed() provide controls for random and noise-based behavior. Setting both at the beginning of a drawing sequence establishes a reproducible starting point, provided the same sequence of calls and relevant parameters are maintained.

A seed alone does not describe the entire image. The canvas dimensions, attribute values, noise scale, particle count, drawing order and palette can all affect the result. For a reproducible character, store these values alongside its seed. A compact record might identify character 014, seed 2718, canvas 800 by 800, noise scale 0.006 and particle count 240.

Keep the seed separate from the character's defining attributes. The attributes describe the intended identity, while the seed determines a particular arrangement of stochastic details. This distinction allows the same character to receive several visual variations without losing its underlying design specification.

Balancing individual identity with collection-wide consistency

A large cast needs a visual system that can produce difference without losing coherence. Begin with a limited set of shared rules: a consistent canvas, a restricted palette, a common line language and a predictable composition structure. Individual characters can then vary within those boundaries.

One way to evaluate the parameter space is to render a contact sheet of twenty-five characters in a five-by-five grid. Look for repeated visual patterns, indistinguishable combinations and empty regions of the design space. If many characters look alike, adjust the ranges or strengthen the mapping between attributes and visible properties.

The opposite problem occurs when every parameter changes everything. If energy affects color, shape, particle count, line width and background texture simultaneously, it becomes difficult to understand the character's visual identity. Assign each attribute a primary responsibility and introduce secondary effects only when they support the composition.

The next step is to define five character attributes, create one reusable p5.js renderer, and generate a contact sheet using a fixed seed. Record the parameter values beside each result so the strongest combinations can become the foundation of a larger fictional cast.

Questions readers ask