Question

How Does AI Art Work?

Answer

How Does AI Art Work?

AI art is where an individual uses an artificial intelligence solution, such as a large language model (LLM) or text-to-image generator, to generate artwork.

In the case of an LLM, essentially, the user enters a prompt into an LLM like DALL-E 3 (accessible via GPT-4), which uses technical approaches like deep learning, neural networks, and natural language processing (NLP) to understand the text and then creates a relevant image to match.

They can do this because it is trained on a large dataset of training data, including natural language, as well as existing images, paintings, and artwork, which collectively give the model a reference point it can use to create new images that match the user’s search intent.

Is AI-Generated Art Original?

Art created via an LLM can’t truly be considered original because these solutions use scraped images as the basis to produce a new image, often without crediting or financially compensating the human artists whose work was used to train the model.

The debate over originality has led many artists to accuse AI vendors of plagiarizing existing artworks.

Most notably, artists Kelly Mckernan, Sarah Andersen, and Karla Ortiz filed a class-action lawsuit at the start of 2023 against Midjourney, Stable Diffusion, and DreamUp for the use of Laion-5B, a database including over 5 billion images taken from across the internet, a dataset which included the work of many human artists.

Can AI Create True Art?

Whether or not an AI solution can create art is subjective and depends on how you define art. For example, Merriam-Webster defines art as “the conscious use of skill and creative imagination, especially in the production of aesthetic objects.”

If we interpret this definition to mean that human creative imagination is essential to creating art, then we could conclude that AI-generated artwork couldn’t be considered art under the traditional definition. This is also true more broadly if we consider art as a medium for human expression.

At the same time, the fact that language models can’t create completely original works and must instead pull from existing artwork in reference images supports the argument that AI-generated works can’t be considered art in the traditional sense.

However, on the other end of the spectrum, language models are actively being used to create and sell pieces of artwork. So, these pieces have been deemed high-quality enough to have won prizes or fetched high sale prices.

Some of these AI-generated pieces have been deemed high quality enough to win prizes and have sold for extremely high prices.

For instance, Jason Allen won a blue ribbon at the Colorado State Fair’s annual art competition, while the AI-generated portrait “Edmond De Belamy” sold for $432,500 at auction in Christie’s New York auction house.

In addition, AI-generated art and human expression aren’t mutually exclusive because language models can generate images and artwork with human guidance.

If a human user crafts prompts to produce an image with an LLM that expresses the emotion or story that they want to tell, then this could be considered a form of artistic expression.

Is AI-Generated Art Subject to Copyright Law?

Whether or not AI artwork is subject to copyright depends on the jurisdiction. Notably, In the U.S., a number of legal cases have confirmed that AI-generated artwork can’t be subject to copyright law, while in the UK, the Copyright Designs and Patents Act 1988 provides copyright protection for computer-generated works without a human creator.

Likewise, the Beijing Internet Court in China recently ruled that AI-generated images in the case met the requirements of “originality” and should be protected under copyright law.

In the U.S., in August 2023, a federal court ruled that copyright law hasn’t extended to “protect works generated by new forms of technology operating absent any guiding human hand” and argued that “human authorship is a bedrock requirement.”

This position was again reaffirmed in September when the U.S. Copyright Office denied Jason Allen’s request for copyright protection of an image created using Midjourney.

In any case, given the uncertainty surrounding the use of copyrighted images and intellectual property in AI vendors’ training data, we recommend seeking legal advice and conducting a risk assessment before attempting to use language models to generate artwork or images in a commercial or enterprise setting.

Can You Sell AI-Generated Artwork?

Yes. It is legal to sell AI-generated artwork the same way you would a human-created piece of art. As mentioned above, artwork created by AI artists has generated a premium, most notably with the “Edmond De Belamy” portrait selling for $432,500 USD at an auction in New York.

While not all AI-generated artwork can be expected to sell for this kind of price, this example illustrates that works created by AI can generate a significant financial return.

What Tools Can I Use to Create AI Artwork?

There are a number of tools you can use to create AI-generated artwork. These include LLMs, such as DALL-E 3, Midjourney, Stable Diffusion, and Adobe Firefly. Each solution offers a slightly different approach to image generation.

For example, Midjourney allows you to see images being developed in real-time and gives users the option to upscale images and choose between multiple variations.

Likewise, Stable Diffusion differentiates itself by allowing users to enter negative prompts, which detail content that they don’t want to see featured in the image.

Related Terms

Tim Keary

Since January 2017, Tim Keary has been a freelance technology writer and reporter, covering enterprise technology and information security.