Zensibles

The medium

AI-generated illustration names a way images are made. Whether it can be a medium — the way ink, oil, or a camera are media — depends on the practice behind it. This is that practice.

The controls

A generation engine is an instrument. A pass can be thrown wide — one seed idea scattered across many styles and compositions — or stopped down to regenerate a single zone of an image that is otherwise settled. It can hold a subject and sweep the style and texture, hold a style and sweep the subject, or merge two prompts into a composite that inherits from both and looks like neither. Knowing which control to reach for, and when, is the first layer of the craft.

The selection pool

Most engines render in batches, so the count that matters is not how many times the artist asked — it is the total number of bespoke images generated and judged. Those are the selection pool; the final piece is a choice from it. The broken ones take no artist to reject: anyone can spot a hand with the wrong number of fingers. The real work starts with the renders that are good on their own. A capable engine produces pleasing images constantly; the artistic questions start there. How is the composition? What is the color doing? What is the subject, what is it doing, what is the image saying? Until the answers cohere into a gestalt, the image isn't a piece — just competent output, with a good engine's merits and none of an author's.

Refinement

From a render with something to it, the process works like evolution. What is working goes back into the engine with more direction, and what comes out is based on the first but moved the way the words point. Selection, then directed mutation, again and again — the distance settings deciding how wild each generation may run. The line branches: an idea gets tested on neighboring subjects to find where it actually lives. Then it converges — the engine strikes the bones of an image worth finishing, and generation narrows to that image alone. A single piece can pass through hundreds of generations this way.

Finishing

The narrowing ends when the image's intent becomes recognizable — when what it is trying to say can be put in a sentence. Once that sentence exists, the image can be judged against something. Does it say it? If not, why not — something in the frame the sentence doesn't want, or something it needs missing? A piece is ready when what it is saying sufficiently matches what it is trying to say.

The artist then adjusts the prompt and may regenerate only targeted zones until they feel the image is saying what they hoped to find. The choice to stop generating and consider the image done belongs to the artist — who should be able to articulate why they stopped where they did, same as any other medium.

The label

The term has edges. AI-generated illustration, used honestly, means an image created entirely by manipulating a generation engine — every element emerged from the process above. Work that continues in another medium — vectors redrawn by hand, added type — is multimedia, and should be named as such. A medium line only means something if there are things it excludes.