Can’t spell “artist” without AI
“Hey DALL·E 2, make me a picture of the Last Supper in Jollibee”. (Image: AJ Dimarucot)
As a former freelance graphic artist and Tumblr apologist, I would say I have both an average knowledge of digital design and a novice understanding of coding.
Although I’ve stepped away from these platforms that fed my creativity and “digital prowess,” I always kept what I learned from them to heart. Now, it seems that the tech-novice art-hoe in me is being awakened once again with artificial intelligence (AI) now having the capability to create complex, hyperrealistic imagery faster than I can load up a blank canvas on Photoshop.
Text to art
Generative modeling takes center stage for the AI art movement. This technology is capable of creating realistic images and art from natural language descriptions. In other words, it can create out-of-this-world art just by inputting simple text.
Generative models are artificial neural networks (like a virtual brain) that undergo rigorous training with large datasets of images and learn to encode these statistically prominent features.
So to differentiate a basketball from a volleyball, for example, it would have to scrape through millions of images of basketballs and volleyballs and analyze each pixel to accurately distinguish one from the other. And over time, the system makes fewer and fewer mistakes.
After training, the software goes through a generative process called diffusion which allows it to produce entirely new images that did not originate from the dataset it initially studied. From incomprehensible noise, the system arranges these pixels into a composition that makes sense to humans: perceivable artwork.
To produce these images, text guides are fed into the system which explicitly describe the desired results; this process is called prompt engineering. Think of it as describing someone’s likeness to a sketch artist, the more specific the details and the wording, the more accurate the result.
Prompt engineering is not just limited to generating images of objects; it can also incorporate painting techniques, period-specific designs, photo lenses, architecture, and even a specific artist’s personal style. Once you enter the text prompt, you are given images that match your visual cues in a matter of seconds.
Software that sparks creativity
OpenAI recently unveiled a new model named DALL·E 2, fittingly named after artist Salvador Dali and the fictional Disney robot WALL-E. Their software can generate remarkably consistent and relevant images from pretty much any text prompt. Here are some pieces from the AI art goldmine that is Reddit:
Some independent, open-source developers have gone on to join a tech company with a similar service called Midjourney, which heavily operates within the Discord platform. And now that the software is open on beta, many users have tried their hand at AI art:
Public perception of AI art is mixed. For one, its early inception means that there are not many copyright measures to protect the existing artists’ works included in the databases. Artist James Gurney is a known supporter of this tech, but promotes artists’ consent to opt-out of including their handmade art in these datasets.
AI artist Lia Coleman also noted another issue with the system’s stereotyped imagery, clearly attributed to the Western-oriented sphere of the internet. If you ask for an image of a CEO, it will show you old Caucasian men in suits. If you ask for a nurse, the system will give you pictures of women in scrubs. After much feedback, the developers are now working on ways to make the software more inclusive.
Although some artists have expressed their frustrations towards this new tech, other professional artists have embraced this technology and have even adopted them into their own practice.
I had a brief conversation with Philippine graphic designer and AI artist AJ Dimarucot; he shared with me that he and artists like Dan Matutina use these systems to easily spark inspiration and overcome artist’s block and have done so for their ongoing art projects.
He suggests that this tech virtually erases the time that would have been used to conduct research and brainstorm, which artists can allocate towards creative execution.
Real art?
The “Is it art, is it not?” discussion has been going on for decades now, simply replacing the digital art advancement that preceded it. People made the same complaints when computers and cameras were invented.
“Graphic artists in the 70s didn’t run out of jobs when the computer came along. They adapted to it, and were able to improve their workflow and productivity as artists,” Dimarucot says. For this established artist, design becomes more accessible when artists learn how to make the most out of the tech available to them.
Cameras and software do not make art, people do. AI art is created by human artists who utilize algorithms as another instrument in their creative arsenal. One shouldn’t underestimate the amount of skill, talent, and intentionality involved in creating mind-bending work.
The future of AI art is bright, with today’s AI-generated art going for as much as $435,000 in auctions and exhibits. Local artists like Dimarucot are hopeful for the AI art scene in the Philippines and wish to see more live AI art exhibits in the near future.
As someone who lacks the artistic conviction and capabilities to express my ideas on a canvas, I too am looking forward to a future where everyone can be an artist.
