The model is not a student
An image model is not in art school. Lugh AI Art on the human using the model, the copyright line on authorship, and why that file is simply not a student.
In Named, not hidden we said the weather around this work is split. In The prompt is not the painting we said a sentence is a mark. In The stroke and the sentence we said kin to the tablet, not the same stroke. Here is the fourth: how a person learns, and what a generative loop can and cannot rhyme with.
The useful comparison is not “the model learns like an artist.” It is this: the human using the model still does the old jobs. Looking. Selecting. Taste. The closer rhyme is the person in the chair, not the model as an apprentice. Mario Klingemann, in CLOT (7 August 2024): “AI is like a piano.” “We don’t credit a piano for a musical performance. The credit goes to the person who performs on it.” “The machine doesn’t create the art—I do.”
A model is not a student. It has no body, no mentor, no intention, and it does not forget. It does not practice drawing. We will not call a trained file a pupil. Copying-as-learning is also not the same thing as copying-as-theft. Academies have always copied. Forgers have always copied. The difference was never the existence of a source. It was what the copier was trying to become.
Start with the human side, because that is the older record. Joshua Reynolds, in Discourse I at the Royal Academy (2 January 1769), wanted the academy as a repository of authentic models, and the living figure copied so the hand gained precision. In Discourse II (11 December 1769) he sharpened it: “Instead of copying the touches of those great masters, copy only their conceptions.” Drawing facility, he said, was like a musical instrument, built by infinite acts. “Nothing is denied to well-directed labour.” At the Florence Academy, Charles H. Graves puts the same limit on the copy: you copy masters to understand technique, “not about making a forgery.” Kimon Nicolaïdes, in 1941, went further into the body the machine does not have: learning to draw is learning to see, and you feel the gesture in your own body. Cézanne, in 1905, called the Louvre the book we learn to read, not a formula to keep. The Bargue plates ran the same ladder: casts, then masters, then the living model. Look, copy, leave.
What holds, if we are careful. Looking rhymes with gathering. Critique rhymes with selection. Taste rhymes with curation. Iteration is still a human act: try, look, refuse, try again. That is the loop we actually work in: dataset, train or fine-tune, prompt, generate a grid, select, upscale, inpaint or vary, remix. Not a button.
What breaks is the student story. The US Copyright Office, in Part 2 (29 January 2025), will not treat a prompt as authorship. Prompts alone do not confer it. Selection, arrangement, and modification of outputs can. That is a human remaining in the chair, not a model graduating. Refik Anadol, to Laura Cocciolillo in Domus (14 January 2026): there is no magic button, and behind every artwork an enormous amount of human work. He talks about data as pigment, and about starting with permission. Anna Ridler treats the dataset as creative work — repetitive, time-consuming craft — not as “using AI to make an image.” Sougwen Chung, from 2020, has had robots interpret her own twenty-year drawing archive. Drawing remains, she says, “one of the foundations.” Feedback from her hand, not a student AI learning to be her.
Holly Herndon, talking to Hans Ulrich Obrist for Art Basel (2 October 2024), said “The model is the artwork… that can generate infinite artworks.” That is protocolism: the system as the piece. It is a serious line. It is not ours. We make a print for a wall.
So the Lugh AI Art version, without the dunk. A person learns by looking hard at other people’s pictures, then by making their own until the hand and the eye agree. A generative process can rhyme with that if a person is still doing the looking, the gathering, the choosing. The file on disk is not in art school. It is not practising. It is not stealing in the way a forger steals, and it is not studying in the way Reynolds meant. Those are three different verbs. We will keep them apart.
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, or a drawing a student sweated over. The human looks. The human chooses. The model does not become someone.