AI as medium, not author

Every piece in the catalog is composed with AI. That is the medium.

The medium is not the author. The author directs the medium. The medium does not direct the author. Most online conversation about AI art assumes the opposite, that the model is making the work and the artist is reviewing the output. That has not described the actual workflow of any serious daily-art account on this side of 2024.

What "AI as medium" actually looks like

There is a piece in your head before there is a prompt. The piece in your head is specific: a kneeling figure in shallow water, a glass crown, wet white cloth, faint blue cracks radiating outward from the knees. The model does not have that piece in its head. The model has the corpus.

The prompt is the first translation from the piece in your head to the corpus. Most prompts do not land on the first call. The corpus answers with the average of what it has seen — generic high-fantasy, generic backlit silhouette, generic queen. The author writes the prompt again, narrowing.

The model is now closer to the piece in the author's head. It is still not the piece in the author's head. The author chooses one generation out of several, edits the regions that did not land, composites elements from a second generation, runs the result through a second model that knows how to do skin better than the first one, drops in a halftone screen by hand, masks the type, and ships.

That is the workflow. The model produces options. The author chooses, edits, composites, and ships. The model has never seen the finished piece — only its parts.

Where direction lives

In the choice of model. Soul_2 for faces. Nano_banana_2 for atmosphere. Marketing_studio_image for posters. The model is the first piece of direction; the prompt is the second.

In the composition. The artist decides what is in the frame — a kneeling figure, not a standing one; the throne offscreen, not in frame; the scythe bisecting the panel, not held by anyone. The model can render these. The model does not pick them.

In the negative space. Half the catalog's force is in what is not in the frame. Crystal Throne has no throne. Beneath the Grain has half a face. Free From Hell has more zine furniture than character. The decision to leave space empty is the most consistently human decision in the workflow.

Where taste lives

In the choice of which generation to keep and which to throw away. Most images produced for any one piece never appear anywhere. The kept generation is selected on a margin that has nothing to do with the prompt. Better hand placement. Better fall of cloth. The slight off-look of an eye that was supposed to be facing left.

The discarded generations are the work, too. They are the steps the author took toward the kept one.

What the medium gives

A daily cadence. The medium makes daily production possible. Without AI, a portrait of the depth of Red Geisha or Crystal Throne would take a week of brushwork. With AI, it takes hours — but only after years of brushwork-level taste have been built up.

A vocabulary at scale. The catalog can include a halftone print, a black-and-white ink splash, a manhwa cover layout, a cyberpunk research file, and a streetwear photograph — all in the same week. No single human medium gives that range. The AI medium does, at the cost of having to maintain consistent taste across all of them.

What the medium does not give

The piece. The piece is upstream of the medium. The author has to have the piece in their head before any model can render anything useful.

A daily cadence is not a daily prompt. It is a daily piece, run through a medium that can produce it in hours instead of weeks.

The single rule

The model is the brush. The artist is the painter. The author is the painting.