As a photographer and artist, I’m eyeing AI-generated art. It’s an exploding phenomenon, and a controversial one. A few artists are embracing it as the inevitable new frontier in the world of visual arts. Others are alarmed or hesitant, concerned over copyright issues or being put out of business.
I’m in middle of the pack. I create and use AI art but recognize and respect its shortcomings. For the most part, I request imagery that’s unlikely to overlap with any existing artwork. Today I literally asked my favorite AI art program to create a self-portrait, for example, and used it for the featured image at the top of this page. Did it overlap somebody else’s art? Very possibly. I don’t know where it got the girl’s features.
For a different example: my daughter is an event planner and recently did a Cinderella-themed Sweet 16 party. The quintessential element of a Cinderella theme is the glass slipper – but nobody said it had to be a clear glass slipper. What if it were stained glass?
AI instinctively knew shoes come in pairs, however, so my request for one glass shoe got the set:
My daughter adores them. So do I. However, I can’t imagine the AI went straight to catalogues and pulled pictures of stained-glass shoes, so it’s highly unlikely I’d be in hot water over any copyright claims here.
About the copyright issues – and there are a few.
First, no one is 100% certain how AI creates its artwork. We know it draws from known sources – hence there’s technically an infringement potential for swiping elements from existing art and modifying it. However, two people could put in the exact same set of instructions and wouldn’t get the same result, no matter how many attempts they made. (That’s key, too: it may take dozens of attempts to get a good outcome.) In the sense that a human being is somewhat defined by having two eyes, two ears, a nose, and a mouth – is it infringement if AI intuits the same based on millions of photos, videos, and paintings, then chooses stylized interpretations of what it’s “seen”? And could it be argued human artist are doing the same thing when using a model for their painting or photograph?
The other end of the spectrum is, AI-generated graphics can’t be copyrighted. You may help yourself to the picture of the glass shoes. I don’t own them. Nobody does. Of course once I modify the original, my modifications are copyrighted.
It makes sense. After all, the real artist here is a machine, vs. a living painter or photographer.
There are other considerations. While some of the outcomes can be hauntingly beautiful, most aren’t usable as-is. Eyes are often problematic. Look closely at the AI self-portrait at the top of the page, and it’s one of the better outcomes. An elf child is created with a claw coming from the middle of her belly, needing Photoshop surgery. A boat can be half-in a house and pier, without any splintered wood to account for it. While a few pieces are good to go straight from the machine, most need human intervention to tweak the outcomes.
There are a handful of AI sites out there. NightCafe.studio gives you 5 credits per day, with a few extra after you’ve subscribed to other feeds, etc. Starryai.com uses a similar credits system to NightCafe. However, most of the truly spectacular outcomes, including those I’ve used here, have come from MidJourney.com. MidJourney gives you around 25 free images – actually 25 minutes of server time and based on how their algorithm works, you’ll most likely get more than 25.
With NightCafe, you can upload one of your own images as a “seed”. I’ve had mixed results. I got three incredible outcomes and the rest… not incredible.
MidJourney is different than the other two options in that when you run your first prompt – the set of text instructions you write, telling the AI what you want your picture to be – the AI gives you back a block of four small images. You may then choose to enlarge and enhance one or more of the smaller pictures; run a second set of creations based on one (or more) of those images to get variations of them; or run the whole initial prompt again for a different block of four images. Or, if you realize you forgot something or think you’ll get a better outcome with slightly different wording, just enter a whole new prompt.
One artist friend decreed AI design is here to stay, so we’d better get used to it. If early response is any indication, she’s absolutely right.