Want to keep up with the latest in tech?
Artificial Intelligence 101: What Is AI Art and How Is It Made?
AI art has taken the world by storm, but how does it work exactly?
Artificial intelligence (AI) is more present than ever as the piece of tech that helps you access your devices through facial recognition, filter spam emails, and navigate roads (just to name a few). Beyond the applications that single individuals can benefit from, there’s the fact that entire organizations can shape the world with AI. Businesses today get to make product recommendations or automatically enable chatbots to respond to customer inquiries thanks to AI. It figures also in healthcare, government, and so many other sectors that affect society.
As AI continuously evolves, of course, it has managed to make its way to more niche areas: like the arts. Today, it’s common to see artificial intelligence art all over social media and other digital channels. If you’re new to the concept, you may be wondering: what is AI-generated art? How is AI art made? How does it differ from human art?
The guide ahead discusses all things AI art. Learn how to distinguish it from authentic, manmade artwork below.
What is AI-Generated Art?
AI art is a new form of digital art where AI and machine learning models receive instructions and then generate artwork. To do this, an AI artist can type in a prompt (like an astronaut riding a horse), then the computer interprets it in the most accurate way possible. Popular pieces of artwork in this category are created through Open AI’s neural network, which mirrors the behaviors and capacities of the human brain in order to create its output.
Photo: @OpenAI on Twitter
How Does AI Art Work?
AI artists typically feed data—such as an art style, existing artwork, or photos—into an AI art generator, producing artwork based on the prompts it receives.
So far, AI art generators use one of three types of AI models: the Generative Adversarial Network (GAN), the Convolutional Neural Network (CNN), and the Neural Style Transfer (NST). GAN consists of two components: generator and discriminator. The former attempts to create original art, while the latter sifts through existing images to determine whether the generator’s creation is new. CNN, meanwhile, uses patterns from artworks to produce new, random images, while NST involves following the art style of a specific artwork to create a new image. (For instance, you can instruct an AI application to recreate your favorite celebrity’s photo as Leonardo da Vinci’s Mona Lisa or Johannes Vermeer’s Girl With a Pearl Earring.)
Humans vs. AI: 7 Ways to Know If Art is AI-Generated
Over the years, artists have found ways to create abstract or realistic art using different styles and mediums. You might find art that looks digitally rendered but was actually hand-drawn or, vice-versa, digital artwork that looks like a painting created in person.
Understandably, it’s become harder to distinguish AI vs. real art, but it pays to be aware of what’s manmade and what isn’t. Here, the signs to look out for when trying to spot AI-generated art:
1. Asymmetrical discrepancies
AI may be smart, but it’s imperfect. Asymmetrical features often give away AI paintings or images. (While human faces aren’t always symmetrical, you can still use asymmetry as a basis for checking AI art.) Take the photo below as an example. It may look realistic based on the texture of the man’s face, but closer inspection shows that it’s a GAN-generated image.
Photo: This Person Does Not Exist
What are the other giveaways? The background can look like a mesh of textures and colors. Also watch out for human subjects that have different colored eyes, inconsistent facial hair growth, and other human attributes.
2. Blurry sections
Blurry features also indicate that a piece of art is AI-generated. In the photo below, you’ll see that the background and ship are blurred or smudged. An actual photograph would look sharp and clean.
Photo: @openaidalle on Instagram
3. Extremely saturated colors
Some AI artwork tends to be more saturated in color than real paintings or images. This indicator often manifests when using AI art generators like Craiyon (formerly DALL-E Mini) and Dream by Wombo.
4. Level of realism
In the AI-generated image of Paris below, it’s obvious that the AI art generator resorted to “filling in” the details of its view of the Eiffel Tower. Hence, the diverse scenes of the streets aren’t captured and they instead look uniform.
5. Smudged or clumpy hair
AI models can’t completely mirror real human features. Aside from the face, check for smudged or clumpy hair. Spot if the ends of the hair are blurring into the background, then double-check them against other hair ends.
6. Image quality
Human artists take their time when adding details, depth, and definition to their artwork. As such, you can expect AI art to have low quality and resolution. Some AI images end up pixelated as if they were cropped from an original photo.
7. Identify AI art watermarks
Some AI art generators put watermarks on their artwork to avoid AI copyright issues. For instance, DALL·E 2 always puts a row of yellow, turquoise, green, red, and blue squares at the bottom right corner of its images.
Photo: @OpenAI on Twitter
8. Use an AI art detector
If you are unsure about whether or not a piece of artwork is AI-generated, you can use an AI art detector like Mayachitra’s GAN detector or umm-maybe’s Hugging Face to confirm. While these programs aren’t always accurate, they can at least supplement your initial intuition.
Art in the Age of AI
While it still garners mixed opinions, there’s no denying that AI art is changing the relationship between the arts and technology. You can expect AI art to only evolve further and find a deeper place in the world.
Want more of this kind of content? Read our technology articles to stay updated with the latest tech trends and developments in innovations like AI art.