AI & Art
Taylor Robertson ’23 details his journey creating art through AI image generators like Lensa and Midjourney, discussing their current capabilities, potential benefits, and ethical and legal considerations.
TAYLOR ROBERTSON // FLAT HAT MAGAZINE
For me, it began with seeing someone I follow on Instagram post a flawless digital portrait. I thought, “Oh, they commissioned it.” Then I tapped to see the next photo in the story, only to find another portrait. Then another. A whole album’s worth.
After investigating, I discovered Lensa. Lensa was able to generate 100 artistic digital portraits of me based on 20 photos of my face, a feature dubbed “Magic Avatar.” Sure, only 20 of the portraits looked enough like me for me to recognize myself, and of those, there were only 10 I truly liked.
Little did I know that this experience was my first plunge into what would become a deep dive into AI art.
As a fantasy fiction writer, I’ve pictured dozens of fantastical places and people in my head, things I can describe with words but have never had the patience to draw or learn to paint digitally. Lensa was just the tip of the iceberg once I started researching what else was out there.
I began to experiment with three different image generation AIs: Stable Diffusion, DALL-E, and Midjourney. Stable Diffusion was the first to market and can run on your own computer (if you have the technical prowess, patience, and RAM). DALL-E was made by OpenAI — the creator of ChatGPT — and is generated on their servers. Midjourney runs through Discord, which makes it handier than the other two; I can easily prompt AI art on the fly from my phone.
Over the past three months, I have generated thousands of AI art pieces. I’ve iterated on a single prompt a dozen times to get a piece of art more specific to what I want. Using those initial avatars I created of myself through Lensa, I’ve even been able to place myself into the art through prompting with a starting image in addition to text, or through a combination of two images.
However, I couldn’t write this article without acknowledging that these image generators scraped the internet without seeking permission from artists or photographers. The images Midjourney or Stable Diffusion generate are a prediction based on everything it has seen before (in other words, everything on the internet is up for grabs). It isn’t capable of true creativity, but the sum of humanity’s artistic creation until now is quite a hefty sum to jumble up and regurgitate. For the modern user who wants to picture their Dungeons & Dragons character better or bring the imaginary wonders in their head to life, AI art is a very strong tool.
TAYLOR ROBERTSON // FLAT HAT MAGAZINE
Attempts to copyright art produced by one of these AIs have failed to hold up in court. Many artists feel cheated out of their work and art styles without any form of compensation. But if any human’s art is merely the recombination of everything they’ve ever seen before, how is Midjourney any different? It’s a question to be further answered by the courts, but there’s no re-sealing this Pandora’s box.
Through my creations, I’ve learned a few lessons of little import:
1) Midjourney is awful at generating flying broomsticks and mirroring fantasy traits (like placing pegasi wings on both sides of a horse’s body or putting a set of bunny ears on both sides of a human head).
2) Midjourney doesn’t know Hogwarts wouldn’t have lamps.
3) Text will be unintelligible.
4) Weird things appear in weird places. Look too closely, and nothing is what it seems.
Ultimately, it makes me happy to feel my dreams realized by art in a way that would have cost tens of thousands of dollars to commission only a year ago. AI has challenged artists, and with ChatGPT and GPT4, it’s threatening other fields as well. My recommendation to everyone would be to create — create whatever AI is currently incapable of creating but also use AI’s current capabilities to create art that will further enrich your life. Even if you’re not interested in AI, being able to use it will likely become one of the most marketable skills.
TAYLOR ROBERTSON // FLAT HAT MAGAZINE