Should AI images be accepted as art?

Viewpoint 1 — the illustrator

"I spent eleven years learning to draw and eight building a client list. Every one of those clients can now type a sentence and get forty images back in a style that looks like mine, because the model was trained on pictures like mine that were scraped off the web without anybody asking me. I never licensed that. Nobody paid me a penny for it.

"People tell me the tool is just a brush. A brush does not need ten thousand of my drawings in order to work. If a human had copied my style by hand for a decade, at least they would have learned something. What I am objecting to is not the technology. It is the transfer — my labour has become somebody else's free raw material, and the first thing it was used for was to stop commissioning me."

Viewpoint 2 — the graphic designer

"I use generative tools every week and I am not embarrassed about it. Eighty of my mood-board images now take twenty minutes instead of two days, and the client gets to reject the bad ideas cheaply, which is the whole point of a mood board. The finished work is still mine: my crop, my type, my palette, my judgement about which of the forty images was not rubbish.

"Every generation has had this fight. Photography was going to kill painting. Synthesisers were going to end the orchestra. Digital brushes were ‘cheating’ when I was at college, and now nobody paints a poster by hand. The skill moves; it does not disappear. Judgement, art direction and knowing when to stop — those are the things the tool cannot do, and they are the things I get paid for."

Viewpoint 3 — the competition judge

"My problem is narrower than either of theirs. I run a competition with rules, and the rules exist so that entries can be compared fairly. A prize for ‘painting’ is a prize for a skill; if one entry took four hundred hours of oil work and another took eleven prompts, I am no longer comparing like with like, whatever the pictures look like on the wall.

"So I do not think the answer is a ban or a free-for-all. It is disclosure and categories. Tell me what you used, put generated work in its own class, and judge it on the things that class is actually about. The competitions that got into trouble are the ones that had no rule at all and found out in public."

Facts box — things you may quote

• In 2022 an image made with a generative tool won the digital art category at the Colorado State Fair in the United States, which had no rule covering generated entries at the time. The result was widely disputed by other entrants.
• In 2023 the photographer Boris Eldagsen refused a Sony World Photography Award after revealing that his winning image was AI-generated. He said he entered to test whether the competition was ready for such entries; it was not.
• UK copyright is automatic and belongs to a human author. The Copyright, Designs and Patents Act 1988 also has an unusual provision for "computer-generated" works with no human author: copyright goes to the person who made the arrangements necessary for the work to be created, and lasts 50 years from creation instead of 70 years after a death.
• The UK Government ran a consultation on Copyright and Artificial Intelligence in 2024–25, looking at whether AI developers should be allowed to train on copyright material and how creators could reserve their rights. At the time of writing, the UK's text and data mining exception covers non-commercial research only.
• Creative Commons has published guidance stating that much of how AI systems use content sits outside what its licences can control, because it sits outside what copyright was built to address. A CC licence on an image is therefore not the same thing as permission to train on it.
• Prompts themselves are generally too short to be protected as literary works, so "I own the prompt" is a weaker claim than it sounds.

Definitions

• Training data — the images and text a model was shown while it was being built.
• Scraping — automatically downloading material from websites in bulk.
• Text and data mining (TDM) — analysing large amounts of material by machine to find patterns. In UK law there is a narrow exception for non-commercial research.
• Rights reservation — a creator marking their work "do not train on this", and a duty on developers to respect that marking.
• Provenance — a record of how an image was made and what went into it.