Why are artists upset about AI image generators?

Two stories from the past few weeks show why feelings run so high. At the end of August, a picture made with Midjourney took first place in the digital art category of the Colorado State Fair's fine art competition. And this month, MIT Technology Review reported that the name of Polish fantasy illustrator Greg Rutkowski has become one of the most popular words people type into Stable Diffusion, used in roughly 93,000 prompts tracked by the search site Lexica.

Both stories involve tools that turn a written description, a prompt, into an image. And both raise the same questions: who made the picture, whose work made the tool possible, and what happens to the people who draw for a living?

The state fair prize

The winning entry, Théâtre D'opéra Spatial, shows classical figures in an ornate hall gazing through a round window at a glowing landscape. It was submitted by Jason Allen, who runs a tabletop game company in Colorado. He did not paint it; Midjourney generated it from his prompts. He then edited it in Photoshop, enlarged it with an upscaling program and printed it on canvas.

When the result spread online, many artists were angry. Some saw it as a sign that creative jobs are under threat; others felt it was unfair to compete against software in a category meant for human skill.

Allen's side of the story matters too. He says he spent weeks refining a prompt, generated hundreds of images and chose his three best, so human judgment shaped the work. He argues that critics are judging the method rather than the image, and he expects competitions to create a separate category for AI art. He also says he labeled his entry "Jason Allen via Midjourney."

The artist whose name became a prompt

Rutkowski's case shows a different worry. His painterly fantasy scenes for games such as Dungeons & Dragons and Magic: The Gathering are well suited to the look people want, and many of his images are online in high quality, with descriptive captions. That makes them easy for a dataset to collect and easy for a model to connect with his name.

He told the magazine he was first surprised, then concerned when a search for his own name returned pictures he had never made. He does not blame people who use his name; to them it is an experiment. But, he said, "it's starting to look like a threat to our careers."

What artists object to, and what supporters say

Put the arguments side by side and the disagreement is clearer:

Artists' concern	Supporters' answer
Their work was collected for training without permission or credit	The tools learn general patterns, the way human artists learn by studying others
A model can imitate a living artist's style on demand	Style itself is not owned; many artists work in similar styles
Search results and art sites could fill with imitations	The tools are a new medium, and people will learn to label and sort them
Cheap images could replace paid commissions	Prompting, choosing and editing still take skill and taste, as Allen argues

Both sides contain something true. The tools do rely on vast amounts of human-made work, and they do give non-artists a new way to make images. The tension sits between those facts.

A worked example: crediting AI-assisted images

Clear credit is the easiest way to respect both artists and audiences. Here are credit lines for the same illustration, from weakest to strongest:

Credit line	What it tells the viewer
"Illustration by Sam Lee"	Implies Sam drew it by hand; misleading
"Sam Lee via Midjourney"	Names the tool, as Allen did
"Image: Sam Lee, generated with Midjourney, edited in Photoshop"	Shows the workflow
"Image: Sam Lee, generated with Midjourney and edited in Photoshop; no artist names used in the prompt"	Shows the workflow and addresses the style question

The names here are made up. Two habits go with the strongest line: describe the look you want in your own words ("moody oil painting, warm candlelight") instead of naming a living artist, and say plainly when an image was AI-made in contests, client work and social posts.

Try it: Take a prompt you have used that names an artist. Rewrite it so it describes the colors, brushwork, lighting and mood instead of the name. Compare the results, and write the credit line you would use if you published it.

What is the debate over training on artists' work?

Image generators learn by studying enormous collections of pictures paired with text. Stable Diffusion, for example, was trained on images drawn from LAION-5B, a huge collection of captioned images from across the web, assembled by a German nonprofit of the same name. Artists did not opt in, and so far there is no simple way to opt out. Holly Herndon and Mat Dryhurst, two artists working in Berlin, have built a search tool, Have I Been Trained, that lets artists check whether their own work is in that data. Stability AI, the company behind Stable Diffusion, says artists who want their work removed should contact LAION, and that the open questions will be settled over time.

The legal picture is unsettled. A UK lawyer quoted by MIT Technology Review says scraping images without consent to train a commercial tool could infringe copyright there, while the UK government is planning to loosen the rules for AI developers. Some art communities have already banned AI-generated images.

Even researchers building these tools acknowledge the issue. The authors of a recent paper on textual inversion, a method that teaches a model a new concept or style from a few images, warn that it could let someone copy an artist's style without consent instead of paying them. They also suggest it could open new options, such as artists licensing their style. Ideas like that, along with consent-based datasets and clear labeling, are where much of the discussion is now heading.