Media Notes / A.I.-Generated Artwork

A piece of A.I.-generated artwork titled Théâtre d'Opéra Spatial, which controversially won a prize at the 2022 Colorado State Fair.^note For clarification, it only won the digital manipulation category.

A.I.-Generated Artwork refers to digital art created by man-made computer algorithms. This is accomplished through the use of "machine learning": existing pieces of art are studied by an Artificial Intelligence, which uses them to generate shapes, patterns, and images.

Though A.I.-generated art has been experimented with for decades, it has grown increasingly notable in 2022 and onwards due to its usage by large corporations (Google, Microsoft, etc.) and its integration with popular art applications. Improvements in art-generating algorithms have led to various advancements, such as imitations of different artstyles, text-to-image programs, and increasing accessibility. Many A.I.-generated art programs are currently available for personal use.

There are several types of A.I. art-generation programs, including:

• Generative Adversarial Networks (GANs): Programs which produce "new" images with the use of a two-part system. GANs are split into a "generator" and a "discriminator"; the generator attempts to create patterns based on input data (in this case, other artwork), while the discriminator discards the generator's data if it is deemed unoriginal or implausible.
• Diffusion Models: Programs which, similar to GANs, produce "new" imagery based on what they've learned from existing ones. Diffusion models train through the addition and removal of "noise" to imagery, with which it learns to alter and manipulate to create new pictures.
• Convolutional Neural Networks (CNNs): Programs which attempt to mimic the human brain's pattern recognition abilities. CNNs are designed to remember and reproduce various patterns and features from the artwork used to train it. In doing so, it learns how to differentiate various parts of images from one another.
• Neural Style Transfers (NSTs): Programs designed to mimic various artstyles. NSTs are given two images (a "content" image and a "reference" image) and trained to implant the artstyle of the latter onto the imagery of the former.

Once an image has been generated by an A.I., it is often altered in various ways to refine the way it looks, often by human artists utilizing digital art tools or customized algorithms. A.I. has also been used to enhance or alter existing human-made art, changing its style or shifting certain elements of it. This has led to the creation of the term "A.I.-assisted art".

While A.I. programs have become notorious for their ability to generate imagery and visual artwork, researchers have also experimented with other types of A.I.-assisted creative works, such as music and video.

Though A.I.-generated artwork has been celebrated for its technological achievements, it has also been sharply criticized by a number of artists and creative professionals. A.I.-generated artwork has been accused of being unethical, unskilled, and devaluing to creative fields. There are several common arguments for this:

• A.I. artwork programs utilize thousands, if not tens of thousands of artists' works in their machine-learning processes. This has brought up various legal and ethical issues, since many of these artists have not consented to the usage of their artwork in such programs, nor are they compensated for the usage of their artwork. Valve had an official ban on any game that uses AI assets citing the risks of misused copyrighted material as a reason. (Later reduced to having to affirm that you own the full rights to every piece of content in the AI's training dataset, and that if players can use your game to generate new content, it won't create anything illegal.)
• The results of A.I. programs are probably uncopyrightable: the US Copyright Act stipulates that only a human creator can hold a copyright, and in 2023, the US Copyright Office stripped copyright protection from the A.I.-generated images used in the comic book Zarya Of The Dawn for that reason (though the human-written script remains protected). At least two other people are challenging the Copyright Office's claim as they have been previously rejected; one of which includes the creator of the State Fair contest entry seen on top of this page.
• A.I. artwork has sparked fears of automation of the creative industry, meaning its advancement could see artists' work being devalued in favor of art-generating algorithms.
• The artistic merit of A.I.-generated art, and its status as original creative work, has often been called into question, as the "original" pieces generated by an algorithm are only able to be developed due to imitation of human artists' work. ^note An artist can create a Pastiche of another artist's style, but inherently contributes their own quirks and decisions to the process, affecting everything from line shapes to subject matter. Purely seeking to imitate without contribution, meanwhile, is at best derivative and at worst an act of forgery. In Layman's Terms, when humans witness the works of other humans, they are capable of evolving them meaningfully — Paintings for example have spanned the gamut of styles from classical to impressionist to cubist to abstract — but A.I. generators are only capable of reproducing what they've already seen, and training A.I. generators with A.I. generated art won't cause newly-developed art to diverge from old art, in fact it'll do the opposite.
• A major critique on the technical side is the eagerness of developers to anthropomorphize these projects as "A.I." in the sense of a program capable of thought. Known as the ELIZA effect^note Named for an extremely primitive chatbot from 1966 that simply rephrased statements made to it, even short conversations caused what the creator described as "powerful delusional thinking" about the program's capacity for creative thought and conscious understanding, even when the users knew that all it was doing was parroting phrases., the same tendency that leads one to perceive a malfunctioning vending machine as passive-aggressive causes many to attribute comprehension, decision-making, and creativity to systems that really amount to (admittedly novel) database compression and search engines.
• An especially concerning trend is that A.I.-generated images are only getting increasingly more difficult to differentiate from actual real-life images, which has sparked immense concern at how easily deepfaked images could be used to ruin people's lives and create false narratives for political reasons, or for extremely personal acts of cruelty, pettiness and vengeance.

A number of prominent creators, including Hayao Miyazaki, Guillermo del Toro, and Mike Mignola, have protested against the usage of A.I. generated artwork. The integration of A.I.-generation into art websites and programs, such as ArtStation, DeviantArt, and Pixiv, has sparked heavy protest among online art communities. This has led to some websites, such as Inkblot, Furaffinity, and Newgrounds preemptively banning the usage of A.I. art on their platforms. Companies such as Celsys (creators of Clip Studio Paint) have withdrawn their plans to include A.I.-integrated tools after backlash by many of their users.

Popular examples of A.I.-generated artwork programs include DALL-E, Stable Diffusion, PlaygroundAI, Midjourney, DeepDream, and WOMBO.

---

Works that use or discuss A.I.-generated artwork:

Advertising

• An ad for BODYARMOR sports drinks compares creepy-looking A.I.-assisted videos to the brand's comparatively "real" sweeteners and flavors, declaring that "nothing in sports should be artificial".
• Another ad for the animated film Despicable Me 4, which ran during the Super Bowl, depicts a group of Minions messing around with A.I. image generation, while poking fun at the technology's quirks.

Comic Books

• Judge Dredd: The Kenny Who? plotline is about a comic book artist who gets replaced by a machine that could draw and write his stories. What makes this interesting is this story was published in 1985 and was originally a satire about creator rights.

Fan Works

• Alien/Species Crossover: Return to LV-426: Major characters like Pike, Lise, and Boone have AI generated images depicting them.
• The Naked Jedi has character art done with AI. As the title might indicate, much of this art is NSFW.

Literature

• Nineteen Eighty-Four: The Minitrue uses novel-writing machines, kaleidoscopes and versificators to generate literary works. A sub-section, Pornosec, where Julia worked, uses only six different plots it switches around to create pornographic novels for the proles.
• Alice and Sparkle is a children's book by Ammaar Reshi with pictures generated with Midjourney and text generated with ChatGPT.
• Roald Dahl wrote a short story titled “The Great Automatic Grammatizator”, about an author named Adolph Knipe and a businessman named John Bohlen who develop a machine that can write based on input from human operators. While the story was written all the way back in 1953 as a satire of how mechanization displaced artisanship when it comes to physical goods, it's eerily prescient regarding not only the capabilities of generative AI, but also the discourse surrounding it.

Live-Action TV

• John Oliver talks about A.I. images in an episode of Last Week Tonight with John Oliver, where he places special focus on Midjourney generations involving himself. Part of the episode revolves around a user who generated images of John Oliver marrying a cabbage, accidentally eating it, and scattering its ashes into the ocean. He ends the episode by marrying a cabbage for real.
• Black Mirror: Joan is Awful has the in-universe Joan Is Awful show be generated by a "quantum computer", where the characters aren't being "acted" but actually using the licensed likeness of real actors.
• The Prisoner (1967) episode titled "The General" was about a supposed revolutionary speed learning program through rote memorization; as it turns out, it is a sinister plot by Number 2 as a form of mid control by bombarding its subjects with facts without requiring them to use any critical thinking. It is revealed that this "General" is a supercomputer that is trained on works of literature and gives an answer to any question asked. Number 6 realizes its fatal flaw by asking it "why?", a question that it cannot answer and as a result overheats and explodes. This episode was written and premiered several decades before Chat-GPT was developed which functions similarly to how the General was depicted here.
• Secret Invasion (2023) uses A.I.-generated visuals in the series' opening cinematic. Director Ali Selim helped contract visual effects company Method Studios to produce the visuals, which he said created a "sense of foreboding" that was perfect for the series.

Music Videos

• The music video for "Ich weiß es nicht" by Lindemann, (a German/Swedish industrial metal supergroup, consisting of Till Lindemann of Rammstein and Peter Tägtgren of PAIN, Hypocrisy and Bloodbath) off of the F & M album, uses the majority of A.I.-generated visuals.

Newspaper Comics

• H.T. Webster predicted the use of A.I.s in a cartoon published in 1923 in the New York Word.

Tabletop Games

• The Dungeons & Dragons sourcebook Bigby Presents: Glory of the Giants was criticized by fans after one of its artists used A.I. to assist with several of the pieces he contributed to the book. D&D's parent company, Wizards of the Coast, denied knowing that A.I. was used for the book's art and said they would redo the artwork submitted and update their guidelines to prevent it from happening in the future. It slipped past their radar because the artist involved was a regular who had been working at the company for nearly a decade.
• Legions of Carcosa, the bestiary sourcebook for The Yellow King, is officially described as 'an experiment in Pelgrane's use of AI-generated art in its products'.

Theatre

• Willy's Chocolate Experience used AI to write the script and generate promotional artwork, resulting in plenty of typos and some weird plot points that were outside of the event's budget.

Video Games

• Activision Blizzard has announced plans to use diffusion A.I. to assist with creating concept art and in-game character skins for their games, including Overwatch 2.
• Arc Tree: Several of the game's achievements have images made with NovelAI.
• The D4DJ multimedia franchise (including its video game arm, D4DJ Groovy Mix) has an in-universe example in Lumina Ichihoshi, an A.I. Virtual Celebrity who produces music.
• The Finals uses A.I. voice cloning for some of the Announcer Chatter.
• Firmament made use of A.I. to assist with creating a variety of content for the game, including paintings, portraits, wallpapers, and other decorations.
• High on Life used Midjourney to "add finishing touches" to the game's world, such as for the artwork on the posters in the bedroom environment. Justin Roiland confirmed the game's development team often messed around with the A.I., enjoying the "weird, funny ideas" they could generate with it.
• PAYDAY 3: As Frederik Larson notes in a (since edited) post to ArtStation, AI was used to produce the paintings seen in Surphaze Art Gallery, the setting of the heist Under The Surphaze.
• Various posters and decorations in the Ready or Not mission "23 Megabytes a Second" were created by generative A.I.
• The Roottrees Are Dead uses AI art in the original itch.io version; an updated version replacing the AI art with human-generated art is being made for Platform/Steam.

Visual Novels

• John Doe, in both its original version and John Doe Plus, uses AI generated backgrounds throughout the routes. They are mostly absent in the routes where John Doe alters himself into an approachable, normal man, and only appear again when John goes back to his regular self. This implies that John is actively altering reality while courting You, which isn't too far a stretch considering he's well aware of the fourth wall.

Webcomics

• One strip of qxlkbh was originally made using DALL-E and redrawn by humans. The program does a half-decent job at emulating the look of a Stick-Figure Comic, but the "dialogue" is a series of random glyphs, only some of which are from the English alphabet.
• Sarah Andersen, the creator of Sarah's Scribbles, has filed a copyright infringement lawsuit against Stability AI and Midjourney (alongside fellow artists Kelly McKernan and Karla Ortiz). Andersen has separately criticized AI-art generation, stating that seeing such programs attempt to emulate her artstyle has made her feel "violated".

Web Video

• Cr1TiKaL submits various A.I.-generated art prompts in his videos "AI Art is Scary" and "AI Art is Taking Over".
• JonTron discusses A.I.-generated artwork and videos, (alongside other A.I.-related web content) in "The A.I. Episode (Chat GPT Takes Over The World)", where he worries about the upcoming A.I. revolution.
• DJ "LegalEagle" Stone dissects some of the legal issues around A.I.-generated images in "A.I. Versus The Law" and to a lesser extent in "NFTs Are Legally Problematic". Besides the currently unresolved legality of scraping image databases for the basis of the machine learning networks, US common law states that only humans can hold a copyright: Stone cites the "monkey selfie" case, where PETA argued unsuccessfully that a macaque who stole a photographer's camera and accidentally snapped a selfie deserved copyright to the image. It's therefore possible that any A.I.-generated image automatically enters the public domain.
• Chris and Zach of OneyPlays published a video titled "MAKING ART WITH AI", where they use DALL-E to generate images ranging from "Peter Griffin eating a deer" to "Donald Trump in H. R. Giger style".
  Zach: Now, do you think this is a good idea, Chris? Does this, like, scare you?
  Chris: Yeah, it really does, because it's gonna put a lot of people out of a job... but god damn, does it look cool!
• Shadiversity addresses the A.I.-art controversy in "STOP THE LIES! - A.I. made art DOES NOT STEAL art!". Being a hobbyist who uses A.I.-art generation, Shad agrees that such programs can be used for unethical purposes, but argues that the programs themselves are not at fault. He states that A.I.-art programs don't inherently "steal" other artists' works, believing many examples of A.I.-art being similar to existing artwork are due to misuse by the individuals using the programs.
• In "Doomed To Be Replaced", Solar Sands offers harsh criticism of A.I.-generated art programs, stating his belief that their overuse will eventually spell the end of human artists' careers. He also worries about the state of the datasets used to train A.I.-art programs, as their large size makes them impossible to feasibly curate for abusive content or intellectual-property violations.
• CYBERGEM's "Raiden Warned About AI Censorship" video is a dialogue between several major characters of Metal Gear Solid 2: Sons of Liberty, based on an endgame conversation in the original MGS2 in which GW, an AI created by the the government conspiracy group The Patriots, declares their intent to control information flow to create the global narrative they want, but instead posits the conspiracy theory that generative A.I. is actually a military weapon used to spread misinformation in a similar way. GW then goes on about how this kind of A.I. was leaked to the public on purpose with the intent of making people paranoid about whether what they see on the Internet or even in real life is real or not, so that GW can present their "solution": Requiring each human user to be verified as a human at all times and having anything they output be cryptographically signed with their credentials in the event that whatever they output is deemed problematic, thus putting an end to online anonymity once and for all. Ironically, as noted in the video description, the vocalizations were made with a voice-cloning AI in order to get the point across that one can't trust anything they see online to be human-made content as opposed to an A.I.-generated construct.
  Raiden: Snake! Snake! Wake up! It's an emergency! Rose knows about the deepfakes! Snake! SNAKE! I guess he stayed up late AI-catfishing again...
• The lyric video for Beyblade X's ending theme "Zoom Zoom" includes a few A.I.-generated fan-art images near the end.
• Jimmy McGee's video essay "AI Revolution is Rotten to the Core" discusses that the problem with AI created media isn't so much how it generates the medium it's being used for, but how it's being used to masquerade the hundreds of man-hours spent to improve on and contribute to the database that allows it to work by attributing its growth to itself, whether positively or negatively.
• Our Drawings: Part of the enmity between Paige and her sister Pillow is the fact that Pillow secretly used Paige's drawings to help train an AI program.

Western Animation

• Godzilla: The Series: An interesting example as one episode predicted AI generated images twenty years before their use became public by having Randy attempt to create a logo by scanning existing artwork and expecting the computer do all the work. The output is an illegible mess of vaguely familiar shapes (which would be the real result before about 2018).

---