AI art refers to any digital art generated by artificial intelligence — where a machine leverages algorithms to create unique images, videos, audio and written material. Trained on billions of annotated examples of existing artwork, styles and techniques, AI art generators produce creations that mimic the characteristics of human-made works.
What Is AI art?
AI art is content (text, images, video, audio) that has been generated using artificial intelligence. AI art generators work by analyzing patterns in existing art and then using statistical models to create new works based on these patterns.
Some artists embrace the technology and its potential, but others are fighting against it, wary of its broader ethical and legal implications. As the lines between artistic expression and technology blurs, our understanding of ownership, fairness and the very nature of creativity itself is up for debate.
What Is AI Art?
AI art is digital content that has been created with artificial intelligence. While it is more commonly associated with visual mediums like images and videos, AI art also includes written works and audio compositions.
Image generators like Midjourney, text generators like ChatGPT, music generators like Suno and video generators like Sora enable anyone to instantly conjure up realistic content in a matter of seconds — without having to wield a brush, pencil, musical instrument or camera. All you have to do is type a few words into a text box.
The inner workings of AI art generators are fundamentally mysterious — even to their creators — which means their outputs are often unexpected. Controlling them is a collaborative, repetitive process, where users tweak their creations with prompt engineering rather than choosing between pencils and oil paint, or changing the aperture of their camera lens.
“It’s not like water color painting,” digital artist Lev Manovich told Built In. “You don’t have to decide every line or every brush stroke. You give it a direction and you get something — it surprises you.” From there, the user “curates” what the generator makes, he continued, refining their prompt over and over again until the tool produces something they like.
The final product can be anything from a hyper-realistic landscape to a country song. In any case, these creations are often the result of numerous prompt iterations, revisions, detours and hours of work — and all on the back of advancements in machine learning.
How Does AI Art Work?
AI art generators work by analyzing patterns in data and using complex statistical models to understand what the prompter is asking for, and then generating content based on those patterns. These tools use a variety of specific AI models and techniques depending on the medium, but the fundamental process of getting from input to output is usually the same:
- Training: First, generators are fed a massive corpus of existing artworks, along with text descriptions. This can be photographs, paintings, books, news articles, songs, screenplays, movies and lyrics.
- Processing: Then, generators use neural network models to find patterns in their training data and replicate them in new content. For instance, image and video generators often use convolutional neural networks to recognize specific things in their training data, and then generative adversarial networks (GANs) to produce an output. Text generators use large language models (LLMs) built on Transformer architectures, and music generators use recurrent neural networks.
- Generating: Once trained, AI art generators can produce artwork based on users’ prompts, drawing on what it has learned to create a unique piece of content. For example, an image generator trained on pictures of apples has formed patterns based on abstract qualities like “roundness” and “redness,” which then overlap to form an understanding of “appleness.” So, when it is asked to generate an image of an apple, the AI combines these relevant qualities to make a new, imagined picture, rather than just assembling bits of existing pictures.
At the end of the day, creation and comprehension are “really intricately linked” when it comes to AI art generation, Ajay Jain, chief technology officer at AI video generation company Genmo, told Built In. The data must be broken down in order to understand its individual parts, and then brought back together to form new patterns and understanding.
“If a language model can generate texts similar to the distribution of texts that people have written, it’s understood something about how people communicate,” Jain said. “If a [visual] model can generate images or videos similar to the things we’ve seen in its training data, it has to understand that training data.”
AI Art Controversies
Artificial intelligence has shaken up the art world, creating significant controversy along the way. From copyright concerns to fears of human artists being replaced, AI-generated art is at the center of several heated legal and ethical debates.
Art Generators Use Copyrighted Work as Training Data
Several human creators have filed lawsuits against AI companies like OpenAI and Stable Diffusion over the years, claiming their copyrighted work was scraped from the web and used as training data without their permission. For now, though, these companies are allowed to do this work under the protection of fair use — a doctrine in U.S. law that permits limited use of copyrighted material without explicit permission from the copyright holder.
To establish copyright infringement, “somebody has to copy somebody else” with “substantial similarity” between the works, according to Rob Heverly, an associate professor at Albany Law School.
AI companies argue that they transform this material enough for it to qualify as fair use. They claim they’re not selling this work as their own, or copying it line for line; rather, they’re using it to teach their AI models — an entirely different use from the original.
The plaintiffs in these various lawsuits are seeking to strip these companies of the protective shield of fair use, in addition to monetary damages. If the courts ultimately determine that these companies are infringing on creators’ copyrighted work, it could mean a lot more lawsuits.
AI Art Cannot Be Copyrighted
The U.S. Copyright Office has long held that works created by non-humans — including machines — are not eligible for copyright. So, at the moment, art created using generative AI cannot be fully copyrighted in the United States. Without copyright protection, creators may lose control over how their work gets distributed and used, and they may miss out on potential income, since others can freely copy, modify and sell the work without permission or compensation.
The root of this issue lies in the way AI art generators work, which involves recognizing patterns in existing data and replicating those patterns in new ways to form original works of art. In other words: Their outputs are simply a culmination of other people’s work, much of which is copyright protected in its own right.
“The guidelines essentially say that you need to have created the work,” Heverly told Built In. “If you just put in, ‘Give me a picture of a starry night,’ you haven’t done the work that’s necessary to create art. All the ‘choices’ are made by the AI. You haven’t made them.”
The lines get even blurrier when the creative work is the result of a collaboration between a human and a machine, which is often the case with AI-generated art. Under these circumstances, only the human’s contributions to the piece are eligible for copyright. This does not include the prompts the person used to instruct the generator, however.
The Technology Can Be Used Maliciously
The technology that powers AI art generators has been used to create deepfakes, which are fabricated videos and audio of real people that can look and sound convincingly authentic.
While deepfakes do have some useful applications, such as synthetic data creation and customer service personalization, they are commonly used for malicious purposes: Spreading misinformation, carrying out phishing attacks and phone scams, even enacting political sabotage. The technology has also gained notoriety for its role in the nonconsensual creation of fake pornographic images and videos — a disturbing trend that has affected celebrities and everyday people alike.
While Congress has prohibited the unauthorized dissemination of sexually explicit photographs and videos of both adults and children, law enforcement agencies are struggling to prosecute the makers of unauthorized deepfake images — because existing legislation on this issue doesn’t take into account the way this technology works.
“It’s different from other forms of nonconsensual pornography distribution, because it’s not that person’s body actually being used in the video,” Star Kashman, privacy attorney and partner at cybercrime-focused Cyber Law Firm, told Built In. “The law has to catch up to that.”
That said, some progress is being made. The U.S. Senate unanimously passed a bill recently that would allow people portrayed in sexually explicit deepfakes to sue the creators. The Senate introduced another bill in 2022 that would make sharing nonconsensual AI-generated pornography illegal, but that is still in its early stages.
Art Generators Are Replacing Human Artists
AI art generators pose a significant threat to human creative professionals across various industries. These tools can produce high-quality content much faster and cheaper than humans. And as their capabilities improve, demand for painters, writers, musicians and actors may decrease.
This is already happening in fields like graphic design, illustration and stock photography, with AI tools being used to create marketing materials, stock images and book covers at increasingly high rates. The music industry is going through something similar, with AI-generated background music and jingles removing the need for session musicians and freelance composers.
This was also a key issue in the 2023 SAG-AFTRA strike, where actors expressed concerns around being supplanted by artificial intelligence. In the end, union leaders agreed that studios cannot use AI tools to create digital replicas of performers without payment or approval. But they punted on the issue of using actors’ prior performances as training data, which means studios are allowed to create “synthetic fakes” — fabricated characters that meld together recognizable features from real actors — without compensating or asking permission from the actor. Screenwriters have also received some protections from AI after the Writers’ Guild of America went on strike in 2023.
While it’s unlikely that AI art generators will replace human creatives altogether, it will probably reduce the number of jobs available in these fields — especially for entry-level positions, which could make it hard for newer artists to enter these fields and establish their careers.
Examples of AI Art
Some works of AI-generated art have captured global attention over the years, sparking both fascination and frustration. Here’s a handful of them.
“Théâtre D’Opéra Spatial” by Jason M. Allen
“Théâtre D’Opéra Spatial” (French for “Space Opera Theater”) is an image created by game designer Jason Michael Allen, with the help of Midjourney. Combining elements of the Renaissance and steampunk styles, the image portrays people in Victorian garb standing in a Baroque-style room, gazing through a large, circular window at a sun-drenched sky.
Both Allen and his creation made headlines in 2022 after taking home the blue ribbon in the digital arts category of the Colorado State Fair Fine Arts Competition — making it one of the first pieces of AI-generated art to win such a prize, and setting off fierce backlash from some artists, who accused Allen of cheating. But Allen insisted he won fair and square, telling CNN that he had to tweak his prompts hundreds of times to get an image he was happy with, and then manually clean it up with additional software like Photoshop and Gigapixel.
That same year, Allen attempted to get “Théâtre D’Opéra Spatial” copyrighted, but the U.S. Copyright Office refused, claiming it included “inextricably merged, inseparable contributions” from both Allen and Midjourney, and that Allen did not have “creative control” of the work.
“Zarya of the Dawn” by Kristina Kashtanova
“Zarya of the Dawn” is a short graphic novel written by author and software developer Kris Kashtanova and illustrated by Midjourney. The book is a work of science fiction that chronicles the adventures of a non-binary person named Zarya, who travels across various worlds in a quest to improve their mental health and forge deeper connections with other people and creatures.
In 2022, the piece made history as the first known work of AI-generated art to be granted registration with the U.S. Copyright Office. But just a few months later, the office partially rescinded the work’s copyright registration, claiming in a letter to Kashtanova’s attorney that it contained “non-human authorship” that had not been taken into account before. While the book’s text, as well as the “selection, coordination and arrangement” of its “written and visual elements,” remain protected, the images themselves are not because they are “not the product of human authorship.”
“Edmond de Belamy” by Obvious
“Edmond de Belamy” is an image made by Obvious, a Paris-based arts collective, using a generative adversarial network. The piece is a blurry, 18th-century-style portrait, and is part of a series of AI-generated images called “La Famille de Belamy” (French for “Belamy’s Family”) also made by Obvious. “Belamy,” the surname of the fictional man depicted in the portrait, is a tribute to Ian Goodfellow, the inventor of GANs — with “bel ami,” meaning “good friend” in French, closely resembling Goodfellow’s name.
Since its creation in 2018, “Edmond de Belamy” has been displayed in museums alongside traditional works of art, and was the first piece of AI-generated art to be auctioned at a major auction house. The image sold at Christie’s Images New York auction for $432,500, making it the second most expensive artwork in the auction behind a screen print from Andy Warhol’s iconic “Myths” series.
Since then, Obvious has been accused of stealing code originally created by Robbie Barrat, an AI artist who shared the software online via GitHub, in order to make the lucrative image. Although a member of Obvious admitted that they used Barrat’s code with little modification, Barrat has not been officially credited for his contribution or received a share of the work’s proceeds.
The Wizard of AI by Alan Warburton
The Wizard of AI is a 20-minute documentary about generative AI, made with generative AI tools. Created by Scottish digital artist Alan Warburton in 2022, the movie discusses the impacts of artificial intelligence on society, particularly as it relates to human artists and designers. The project was commissioned by the Open Data Institute, a non-profit co-founded by World Wide Web inventor Tim Berners-Lee to promote transparency and accountability through data.
The vast majority of the videos and images in The Wizard of AI were made with Midjourney, Stable Diffusion, Runway and Pika, according to Warburton. On his website, he explained that these tools were used to “discuss and critique the legal, aesthetic and and ethical problems engendered by AI-automated platforms.” But the movie isn’t entirely AI-generated; it also includes real news clips and interviews with tech and art experts. And Warburton wrote the script himself, reportedly drawing from his PhD research Digital Culture and Communication at the Vasari Research Centre in London.
Sunspring by Benjamin
Sunspring is a short film written entirely by an RNN-powered chatbot named Benjamin, which was trained exclusively on human-made screenplays. It was originally created by British filmmaker Oscar Sharp and NYU AI researcher Ross Goodwin in 2016 for the 48hr Film Challenge contest of Sci-Fi London, a film festival dedicated to science fiction, and was later released on YouTube by tech news website Ars Technica.
Set in a futuristic world, the movie stars actors Thomas Middleditch and Elisabeth Grey, as well as comedian Humphrey Ker, who play three people “probably” involved in a love triangle, according to producer End Cue. The whole project was written and filmed in just two days.
Will AI Art Replace Human-Made Art?
AI art has started to shift society’s view of creativity from something that requires high levels of intelligence and imagination to something that can be synthesized and manipulated with nothing more than some data.
But not all art is simply an amalgamation of pixels and letters to look at. At its best, art can challenge us, comfort us, excite us and inspire us. It has the capacity to touch us in a profound way. While AI is really good at mimicking human emotions and empathy, it is still a long way away from actually being able to think and create like a human. And art created in seconds with a few keystrokes often lacks the same prestige as works that require real talent and effort.
“People value skills. If you [generate] some beautiful, realistic image, people will say, ‘Well, did you make it? Did you spend ten years learning how to draw like this, or did the AI do it?’” Manovich, the digital artist, said. “Craftsmanship, education, mastery is valued.”
For these reasons, it is unlikely that AI art will completely replace traditional art. Instead, art generators are expected to become tools for human artists going forward, enhancing their creative process and expanding their capabilities.
“Everybody is going to be a creator,” Genmo CTO Jain said. “There’s going to be a lot more people who are able to tell their stories that aren’t currently able to — they don’t have the means or the time necessarily.”
Frequently Asked Questions
Why is AI art controversial?
AI art is controversial due to concerns over authenticity and creativity, as some argue that artificial intelligence lacks the ability to captivate human expression in that way and is thus devaluing traditional artistic skills. It also raises complex issues around copyright — particularly around ownership and originality — and the utilization of existing artworks as training data. Plus, the technology used to make AI art is increasingly used in malicious ways, including phishing attacks, phone scams and the nonconsensual creation of pornographic images using deepfakes.
Is AI art legal?
While it is not explicitly illegal to produce content using generative AI, the subject of AI art continues to exist in a legal gray area. The primary issues involve fair use and intellectual property, as the legality of a specific piece often hinges on how the AI model was trained and whether the generated art infringes on existing copyrighted works. Plus, AI art isn’t protected by U.S. copyright laws, so questions around authorship and ownership remain unanswered.
Does AI art use real art?
Yes, AI art generators are trained on works of real art that have been made by humans.
