The prompt is not the painting
A prompt is not a button. Lugh AI Art on words as first marks, how image generators actually work, and why this medium still needs a human who directs, iterates, and chooses.
In Named, not hidden we said the weather around this work is split. This is the other half: what we think we are doing when we make a picture this way.
People keep asking whether it is cheating. The honest answer is that it is a medium. We start with an idea and make it visible. The first marks are sentences: light, fabric, weather, who is standing where. That is not dragging a pencil, and it is not a button that paints for you. If descriptive text stands in for the old marks, the artist still directs, iterates, and chooses.
In 1859, Charles Baudelaire called photography art’s “most mortal enemy,” then allowed it a job as humble as printing. He was not blessing the camera. Kevin Roose, in the New York Times (2 September 2022), pointed back to that Salon as the template for every later “new tech versus art” fight. Glenn Lowry at MoMA, on CBS (22 February 2026), heard the same echo: skepticism around AI mirroring photography, the human hand taken out of making an image, and the fact that you cannot stop the technology.
Alfred Stieglitz took the other side. On 14 March 1922, in The Sun: “I hold that photography is a positive and additional medium of expression.” Edward Steichen said the camera could say things no other medium could, and later burned his paintings as confirmation of that faith (IPHF / Aperture, 1969 — secondary). Mario Klingemann, in Verse (2 November 2023), put a current instrument in that lineage: AI as a “parascope,” like a microscope or a telescope, and also like a piano. Camera access did not devalue photography; a tool does not make anyone an artist automatically. Look at what the artist was trying to say, not at the stamp “made with AI.”
In Burrow-Giles Lithographic Co. v. Sarony (1884), the camera did not kill authorship if a human determined posing, costume, arrangement, lighting. The US Copyright Office’s Part 2 report (29 January 2025) uses that case as the AI-as-tool template. Prompts alone do not confer authorship: they do not provide sufficient control, and revising a prompt is still re-rolling the dice. What it treats as protectable is selection, arrangement, and modification of outputs. A sentence is a mark. A pile of unchosen outputs is not a work.
Lev Manovich, in 2026, draws a line from Walter Benjamin’s mechanical reproduction to “machine variability,” and treats generative media as a medium that thinks with you: many ideas, not full control over the visualisation. Benjamin did not say the camera replaced the brush.
Named practitioners sound closer to Stieglitz than to a magic button. Refik Anadol, to Laura Cocciolillo in Domus (14 January 2026): “There is no magic button. Behind every artwork there is an enormous amount of human work.” A thinking brush; data painting; models trained on permissioned data. Sougwen Chung (The Art Newspaper, 26 March 2026; MIT Technology Review, 17 February 2025) frames it as collaboration, not a transactional tap — not a replacement for traditional methods; art remains a space to affirm human agency. Anna Ridler (The Observer, 9 January 2025; Right Click Save, 9 July 2026) treats the dataset as creative work, and the artwork as the thinking and the choices, not only the output image. When the field collapses to “AI slop,” she argues, show the parts behind the picture. Christie’s framed AI as enhancing human agency rather than substituting for it, including work generated through refined text prompts. David Holz at Midjourney talks about an imagination engine and a prompt-plus-iteration loop (four variations at a time), not about museum authorship.
Other lines in the room: Holly Herndon (AnOther, 12 September 2024) says it is not about the single image, but the interaction with a model. Mat Dryhurst (SLEEK, 24 June 2026) goes further, toward “protocols or models as artistic media,” away from promptism. That is a serious practice. It is not ours. We work in words, then choices. Hito Steyerl’s “mean images” (New Left Review, 2023) sits on the other side: the statistical average, the picture that belongs to no one. We name it because it is part of why “slop” stuck. We are not endorsing it.
So here is the thesis, without the thin version. Words are the new mark-making. The artist still directs, iterates, and chooses, and often composites or inpaints until the picture holds. One prompt is not a painting. A directed run of them, with a human at the gate, can be.
At Lugh AI Art we name the method. We do not call the files handmade. We do not pretend a generated image is a photograph of a room that happened. The medium is new. The job is old: see something, make it visible, stand behind the choice.