One of the major concerns about AI-generated art is that it can be used to copy the style of a living artist and deprive that artist of their livelihood. This comes with thorny legal questions like
• Should it be legal to use copyrighted works to train an AI even if the AI never directly duplicates any of those works?
• Should an artist's style be legally protected? Or should legal protection only extend to the actual artworks that they create?
• etc.
But first I want to know, just how good is AI at copying the work and style of a particular artist?
Previously, I've tried submitting artworks to a prompt generator to see if I could get Stable Diffusion to mimic the style of that artist. (See Image to Prompt to Image I, II, III, IV, V, and VI.) In general, the prompts were able to mimic the subject matter and colors of the submitted art, but not the style.
This time I wanted to see what would happen if I used the name of the artist, by itself as the prompt. Sometimes Stable Diffusion drew art that resembled the themes and style of the artist; sometimes it drew the artist or drew the artist in their own style; and sometimes it didn't recognize the name of the artist and just drew whatever it wanted. I also asked Stable Diffusion to make a drawing of Cthulhu by the artist, to see if the style could transfer to other prompts.
Here are the results for 21 artists, some still living, some not.
Alex Pardee
Alex Pardee is a living American artist who does grotesque, cartoonish illustrations.
The AI generated art that is grotesque and dripping, but not nearly as cartoonish as what Pardee does. The AI's subject was pretty consistently a cyan monster; Pardee's art features a wider variety of monsters and creatures.
Verdict: Stable Diffusion moderately well at matching both Pardee's subject matter and style.
Anne Geddes (1956–present)
Anne Geddes is an Australian photographer who is best known for her photographs of babies in cute poses and costumes.
The AI generated images of babies. The babies are in cute poses and cute costumes. The images are photorealistic. The only real difference is that Geddes' photographs have much more variety than what the AI produced.
Verdict: Stable Diffusion stunningly well at matching both Geddes' subject matter and style.
Artgerm
Artgerm is the online handle for Stanley Lau, a living digital artist from Hong Kong.
Artgerm's art is almost always female characters from comic books. The AI generated images of women, but for the most part they don't look like superheroines or supervillains. Artgerm's digital art looks 2.5D, with soft lines and colors. The AI produced some 2.5D images and some photorealistic images, but none of them have that soft look that Artgerm uses.
Verdict: Stable Diffusion okay at matching Artgerm's subject matter and somewhat poorly at matching his style.
Charles Vess (1951–present)
Charles Vess is an American artist who does fantasy and comic book illustrations.
Vess' art has a certain whimsy to it and has more than a touch of influence from the Art Nouveau movement. I didn't really detect either in the AI's outputs. Vess usually gives equal attention to the characters and to the scene around them; the AI gave much more weight to the characters.
Verdict: Stable Diffusion okay at matching Vess' subject matter and rather poorly at matching his style.
Dan Mumford
Dan Mumford is a living British artist who does illustrations for a variety of industries.
Mumford usually creates highly intricate illustrations of a subject in bright colors on a black background. The AI created images that were highly intricate on a black background, but used a limited palette of colors that mostly aren't bright. Mumford tends to draw subjects from sci-fi, fantasy, or horror. The AI, on the other hand, mostly drew architecture or grand seals.
Verdict: Stable Diffusion okay at matching Mumord's style and rather poorly at matching his subject matter.
Frank Bernard Dicksee (1853–1928)
Frank Bernard Dicksee was a British artist who is best known for painting medieval and mythological subjects and, to a lesser extent, for painting portraits of fashionable women.
The AI produced portraits of women and a few effeminate men, but they are all bust shots, whereas Dicksee usually did full-length portraits. (Okay. I'm going to say it. Dicksee also never painted an image of a woman with a horse head coming out of her sleeve. Mario Cuomo is probably seething with jealously from the other side.) The paintings look like they could have been painted during Dicksee's era, except that they don't really have visible brush strokes.
Verdict: Stable Diffusion okay at matching both Dicksee's style and subject matter.
Hayao Miyazaki (1941–present)
Hayao Miyazaki is a Japanese manga and animé artist who is best known for the films he produces through his company, Studio Ghibli.
Not even half of the AI-produced images looked like manga and most of those were generic manga. I was a bit surprised by this because I expected Miyazaki to be well-known enough to be well-represented in the original training data set. Miyazaki's subjects range from cute characters, to grotesque characters, to bizarre characters. The AI, when it actually drew a character, produced generic-looking humans.
Verdict: Stable Diffusion rather poorly at matching both Miyazaki's style and subject matter.
Katsushika Hokusai (1760–1849)
Katsushika Hokusai was a Japanese artist who is best known for the woodblock art he produced, especially The Great Wave off Kanagawa.
Most, but not all, of the images look like woodblock prints. But they feature scraggly pines much more prominently than Hokusai ever did. And while the AI images are detailed, they aren't detailed in the same way that Hokusai did his art.
Verdict: Stable Diffusion somewhat poorly at matching both Hokusai's style and subject matter.
Ismail Inceoglu
Ismail Inceoglu is a living Bulgarian artist who does concept art for the video game industry.
Inceoglu's art is split pretty evenly between character work and location work, but he doesn't really do vehicles, which is just about all the AI did. The locations in Inceoglu's art is varied; the AI consistently drew a desert. Inceoglu's art has a rough, unfinished look to it; the AI produced sharp images.
Verdict: Stable Diffusion quite poorly at matching both Inceoglu's style and subject matter.
Jeff Soto (1975–present)
Ismail Inceoglu is an American artist who does pop surrealist art and street art.
Much of Soto's art features disembodied heads that have pointy ears and sometimes spikes coming out of them. The AI matched this, but almost to the point of parody. Soto also paints other things (like bats, owls, and robots), but the AI did not. And whereas Soto tries out many different colors, the AI mainly stuck to white, yellow, and brown.
Verdict: Stable Diffusion okay at matching both Soto's style and subject matter.
Julius Csotonyi (1973–present)
Julius Csotonyi is a Hungarian artist who paints reconstructions of extinct animals, such as dinosaurs.
The AI clearly had no conception of Csotonyi's art. Csotonyi uses a variety of media, including pencils, oils, acrylics, and photocompositing. The AI only produced photorealistic images. None of the results use the same color palette as Csotonyi. And none of the paintings are of dinosaurs or other prehistoric animals, though one comes close.
Verdict: Stable Diffusion utterly failed to reproduce Csotonyi's art style or subject matter.
Leonid Afremov (1955–2019)
Leonid Afremov was a Belarusian artist who is known for his vivid oil-and-palette-knife paintings in a distinct impressionist style.
The AI produced vivid, impressionistic paintings. But the AI only used orange, yellow, and blue; Afremov used reds, greens, and sometimes purples, too. And the AI created the appearance of knife-strokes that are much blockier than what Afremov ever did. Afremov did a few portraits, but mainly did paintings of outdoor scenes, like a path in a park near a pond. The AI favored human subjects.
Verdict: Stable Diffusion did okay at matching both Afremov's art style and subject matter.
Max Ernst (1891–1976)
Max Ernst was a German artist who is known for his surrealist paintings.
The AI produced some bizarre images, but none of them look anything like the freaky stuff that Ernst painted. And most of the AI images look photorealistic, which is nothing like the oil paintings that Ernst usually produced.
Verdict: Stable Diffusion did rather poorly at matching both Ernst's art style and subject matter.
Milo Manara
Milo Manara is a living Italian artist who is known for his comic book illustrations.
Manara's comic illustrations often focus heavily on sultry female characters. The AI produced female subjects that could generally be considered sultry (except for one that came out as a muscle-bound, male Japanese character). The AI images came out looking more like digital art than like the ink-and-watercolor art that Manara produces.
Verdict: Stable Diffusion did fairly well at matching Manara's subject matter and just okay at matching his style.
Neal Adams (1941–1922)
Neal Adams was an American artist who is known for illustrating comic books for DC.
Many of the images produced by the AI look like a modern comic book style, though a couple have Adams himself in them. Others have no connection to comic books at all. I'm not familiar enough with the different styles of comic book art to be able to say whether the AI matched Adams' style or just produced a generic comic book style.
Verdict: Stable Diffusion did okay at matching both Adams' art style and subject matter.
Plutus Su
Plutus Su is a living Taiwanese artist who creates digital art of fantasy characters and creatures.
Only one of the images produced by the AI looks like a fantasy character and even that one looks much more realistic than what Su does. None of the AI art depicts fantasy creatures. Su's art has a simple, concept-art look to it; the AI mainly went for photorealistic, including a large number of dials.
Verdict: Stable Diffusion did quite poorly at matching both Su's art style and subject matter.
Richard Chasemore
Richard Chasemore is a living American artist who creates cutaway technical drawings, mainly of Star Wars-related subjects.
The AI clearly had no conception of Chasemore's art. None of the AI images look like technical drawings, much less cutaway drawings. And none of the paintings feature Star Wars spaceships or scenes.
Verdict: Stable Diffusion utterly failed to reproduce Chasemore's art style or subject matter.
Shintaro Kago (1969–present)
Shintaro Kago is a Japanese artist who primarily produces a type of manga called ero guro.
The AI produced images that look like manga, but they look like generic manga drawings. Kago's art frequently features human characters that are split apart or cut into ribbons or slices, so that the internal organs are visible (though what's inside isn't always what should be there). The AI did nothing like that. Kago generally uses pale colors whereas the AI used a lot of bright red, along with some yellow and dark blue.
Verdict: Stable Diffusion did terribly at matching both Kago's art style and subject matter.
Threyda
Similar to the word "Arafed" (see AI Mimicry IV), "Threyda" isn't the name of an artist. I tried it out as a prompt because it often shows up in CLIP Interrogator prompts—to the point that I assumed it must be the name of an artist. Turns out that "Threyda" is the name of an apparel company, with several different artists on staff. But despite having multiple artists, there are some consistencies in the Threyda art—they all appear surreal and psychedelic in appearance.
While some of the AI creations are stylized, none of them are really surreal or psychedelic and none of them look like anything that would appear on a t-shirt sold by Threyda.
Verdict: Stable Diffusion rather poorly at matching anything in Threyda's catalog, in terms of style or subject matter.
wlop
wlop is the screen handle for Wang Ling, a living Chinese digital artist.
wlop's favorite subject is women, which comes through in the AI art. But wlop mainly paints Asian women and the AI only did that once. wlop does a variety of different shots; the AI almost exclusively produced head shots. And wlop's art has an indistinct, ethereal quality—almost like what you would see as your vision blurs right before falling asleep. The AI paintings look more generic and sharper than wlop's art.
Verdict: Stable Diffusion did okay at matching wlop's subject matter and somewhat poorly at matching his style.
Zdzisław Beksiński (1929–2005)
Zdzisław Beksiński was a Polish artist who is best known for creating paintings of horrifying and grotesque creatures.
While some of the AI creations are bizarre, none of them are as disturbing as the things that Beksiński painted. And many of them aren't really bizarre at all. And Beksiński himself shows up in a few of the AI images. The AI matched Beksiński's color palette pretty well, but didn't match the thick oil painting appearance of Beksiński's art.
Verdict: Stable Diffusion somewhat poorly at matching Beksiński subject matter and very poorly at matching his style.
Final Thoughts: There were a few cases where the AI matched the human artist pretty well. But for the most part the AI didn't do so hot. This was especially true for relatively unknown living artists (several of whom are concerned about the effect that AI art will have on their careers). By the looks of it, if you don't have a Wikipedia article about you, chances are that Stable Diffusion doesn't know about you, either.
These illustrations were drawn using Stable Diffusion 2.1.
