Can AI Create Original Art Better than an Artist?
Testing the Technology to Find Out
Artificial intelligence has proven itself over the past decade for its computing prowess, the ability to learn from experience, adapt to new changes and perform tasks just like people. However, creativity and self-expression have been solely in the domain of humans. Until now.
The art world is all a-twitter because a person used AI tools and other software to generate a piece of art that won first place in the Colorado State Fair Fine Arts Competition. This begs the question, is it really art?
First, a Little Background
Jason Allen, a game designer from Pueblo West, Colo.,, created the winning entry. He entered three pieces in the emerging artist division’s “digital arts/digitally-manipulated photography” category at the Colorado State Fair Fine Arts Competition. The definition for the category states that digital art refers to works that use “digital technology as part of the creative or presentation process.”
A slight pause here to insert a dictionary definition of art:
“The expression or application of human creative skill and imagination, typically in a visual form such as painting or sculpture, producing works to be appreciated primarily for their beauty or emotional power.”
Allen’s winning image, titled “Théâtre D’opéra Spatial” (French for “Space Opera Theater”), was made with Midjourney — an artificial intelligence system that works with user generated prompts, either written, image or both, to produce an output of digital image representations.
To form an opinion, I signed up to try AI art myself.
Test Driving AI to Create Art
Full disclosure, I have a background as a graphic designer, fine artist and website designer. However, pretty much anyone with average computer skills could use the program I accessed called DreamStudio.ai, the light version of Stability.ai. I chose it because it offers 100 free credits without providing a credit card, which is a few hundred images depending on the settings you choose. It’s also a lot of fun.
I watched a 7-minute beginner’s YouTube video covering DreamStudio’s tools and navigation before opening the program and poking around the menus a bit. This was super helpful to learn how to format and manipulate the prompts–words, phrases and spacers used to generate images. I easily started creating right away.
Even though AI crunched the input I fed it–words and a photo–and returned several versions of my dog on stage with a microphone in the hyperrealism style of Pedro Campos, the prompts never generated quite the image I wanted. Several samples had no microphone. I had typed “standing in front of a microphone” which I replaced with “standing behind a microphone. The microphone is on a microphone stand.” That helped, but then a weird cord appeared wrapped around my dog. One image of my dog was pretty good, but she had five legs!
Next, I tried an impressionist version of a princess on a throne holding a puppy, with a castle interior background. I added a few more complex prompts. I could generate a pretty nice princess, or a puppy, but not both together. Also, I did not give any prompts on skin color or nationality, yet my princess was snow white. Gender and racial bias is an issue in AI, and not just in AI generated art. When I added black skin to the prompt, the character’s facial skin and hair changed, but did not look as realistic. There was a fur blanket covering her from the chest down, assuming because the software did not render the arms, whereas the white princess had lovely arms and a beautiful gown. Note that AI inherits bias from the humans that decide which data sets to feed it and the quality of the data sets. It is not the AI deciding to be biased, it just does not have enough information.
I could see where I could generate some interesting images in pieces. The software allows erasing the background around an image (rather clunkily), so I could have it generate another background. However, it would be much more refined to take these pieces into Adobe Photoshop to compile into a piece of artwork.
And that is exactly what Jason Allen did. One of Allen’s AI generated images of a woman rendered without a head, which he added in Photoshop, along with other artistic embellishments. Allen said he spent over 80 hours on his three entries. The fair awarded him $300 for his winning piece.
So, is it Art?
Yes, I believe that images of the quality that Allen created are art. His winning entry is a stunning, highly detailed rendering that brought together an artist’s vision, computer generated elements, human composition and editing. The artist also adhered to the rules of the contest.
The grumpy complaints that AI generated images are the computer version of a banana taped to a gallery wall is not a fair comparison. It takes skill, vision and time to use AI art tools to create a work worthy of a gallery or contest. However, I remember clearly feeling the same way one time. I was working as a graphic artist and 3D modeler for network television when people started generating drawings using Corel Draw, an early consumer graphic design program. They would print computer drawn images at home and sell them. I thought it was ridiculous, and certainly not fine art at all! My how perception can change.
As an Amazon Associate, Arvig earns from qualifying purchases.
The Future of AI Art
AI-generated art has been around since 1973, when American computer scientist Harold Cohen created the first-ever AI painting. However, because of the large datasets and computer power needed to generate detailed pieces of art, systems were almost exclusively owned by large tech companies. Now, with more efficiency in AI, and faster and cheaper computer processors, the technology has worked its way to the consumer market. Tools released last year include DALL-E 2, Midjourney and Stable Diffusion.
Data bias, and regulations in general, is something that we should address for all applications among the developers of AI. It is not a problem exclusive to art. One AI art bias example noted was datasets portraying nurses as women. Call up a Google Images search with the subject “nurse,” and you will see a page predominantly of women nurses. And using AI art to create hate speech or for nefarious purposes? Well, that is already being done with other tools too.
No matter the concerns raised, since AI art generation tools are already publicly accessible, the genie is not going back into the bottle. Besides, we’ve passively accepted that every major tech player is gathering data on us when we go online, use navigation in our car or operate a smart appliance. This seems way less intrusive.
AI can now help with the creative process, opening up the creative arts to those that struggle with traditional methods, or artists that want to add a new layer to their skills. Let’s embrace this new creative frontier and see where it goes. I know I’m going to continue creating images with DreamStudio.ai, at least until my 100 credits run out. So, if you would like to buy a hyperrealism digital painting of my dog’s rock and roll debut, drop me a line!