By Zach Collier
Artificial intelligence is all the rage right now, especially when it comes to generating images. I’m going to tell you why you shouldn’t generate an image and then immediately use it for your album art or post it to Instagram as promotional material.
There are so many different things that I could say about this subject. In this article at least, I’m not going to be talking about intellectual property or the ethics of AI. Instead, I’ll discuss things from an aesthetic perspective – specifically why AI can be used as a jumping off point, but not a final product. I’m also going to explain in detail why some of this art is so uncanny and off-putting.
Let’s use the album art from my latest release, “A Blessing,” as a case study. The main image for the album artwork was initially generated using artificial intelligence. This is the original image that was generated with AI:
You’ll notice right off the bat that while it bears a striking resemblance to me, it’s… off. The first reason why it feels weird is that it’s too smooth. That’s because the AI is mimicking the aesthetic of a painted portrait, but isn’t smart enough yet to include details like canvas texture or brushstrokes. If this was a real painting, it would exist on a physical medium. If this was a digital work created by a human (at least one with experience), that person would make it appear to exist on a physical medium.
Secondly, there are some strange, strange things about my anatomy. In this image, I’m missing eyelashes, I have some distortion/artifacting in my eyeballs and pupils, and (in the creepiest turn of events) the skin around my eyes was generated using what looks like samples of skin from human hands. Additionally, my lips are actually made out of two different lips – both of which were rendered from different angles. My Adam’s apple is also incredibly pointy. I could stab someone with that thing.
My clothes are also very strange. Apparently, this shirt I’m wearing is red in the shadows and then turns white in the sunlight. And, for reasons unknown to me, I’m some kind of astronaut wearing a space suit with my helmet off.
This image is also missing some key features that would be important to anyone who has seen photos of me and would recognize my face in real life. I’m missing several easily identifiable moles, and this AI interpretation of my face has also decided to fill out my receding hairline. I’m a father of 3 kids. I’ve lost hair. Gotta make sure I represent myself authentically.
Here’s my first round of corrections to the image:
I’ve added a little bit of stubble, I’ve smoothed out my Adam’s apple, I’ve added my trademark moles/freckles, my eyeballs are no longer made of the skin of my victims, I’ve fixed my wonky red/white shirt, I’ve fixed the artifacting in my eyes, my hairline has been pulled back slightly, and I’ve removed the metal components of my spacesuit. Instead of looking like I’m wearing some kind of gear for interstellar travel, I’ve made it look like I’m wearing a hoodie over a canvas jacket.
Most importantly, I’ve added a canvas overlay to the image. This gives the illusion that this is a painting that was actually done on a real, physical medium. It adds a lot of depth to what was a pretty stagnant and eerily smooth original image.
Despite all of these edits, the biggest remaining issue is that this version of my likeness exists… nowhere. It appears like I’m floating in some kind of void with a vignette around it. That makes the whole thing feel unfinished.
To rectify this, I lifted “me” off of the AI’s original background and painted my own background. The purpose of this background was to pay homage to the lyrical content of my single. The song asks the question: “What if you had met the love of your life under better circumstances?”
To create this background, the first thing I did was add a renaissance-inspired halo to my likeness to emphasize the sacredness of the idea that struggle and sacrifice are all part of some cosmic plan:
After adding this, I got to work on a much more expansive background.
To the right, I’ve run a photo of Utah’s mountains through several different filters to resemble a painted texture. To the left, I’ve done the same thing to a public domain image of Southampton in the UK. I’ve taken elements of that filtered photo and manipulated them into something entirely new. In this new backdrop, Utah represents my reality, and Southampton represents an alternate reality where things went better for me and my relationship. My likeness sits at the intersection of these two realities. I’ve retained the vignette from the original AI render.
Once the image was complete, I didn’t stop there. I took the image and added elements of graphic design (including font pairings, margins, a planned informational hierarchy, and complimentary colors) to turn this into an actual album cover instead of just a cool image. I did this to help this cover jive with past releases.
Here’s a side-by-side comparison between the original AI image and my finished product.
AI is a powerful tool. It can spark inspiration and even build out a framework for an artistic vision. But it shouldn’t be your end product. Do you see how much work went into this? I had to fix distortions, portions of the art that didn’t accurately reflect what I really look like, and weird textures. I then added elements that imbued the piece with meaning so that it actually meant something to me.
There will come a day when AI doesn’t leave behind weird artifacts. This will happen sooner rather than later. No matter where technology goes and what tools are developed, never forget to put forth effort. Doing so will infuse meaning and purpose into your art, and audiences will feel that.
Resonant art is so much more than auto-generated content.
