I made this comment on a YouTube community post and thought I may as well phone it in to my SubStack because I spent a lot of time on it.
Machine art (it's not AI) tends to generate the same poses and compositions over and over, and they are tediously boring and simplistic. This is because human artists are overwhelmingly NOT VERY GOOD, so they do the same poses and compositions over and over, and the machine works by finding averages.
What we as human beings like and appreciate about art is the novel, the innovative, the groundbreaking. The machine cannot do this unless the human being operating it is novel and innovative and groundbreaking themselves, with a substantial influence from proper curation of the training data. (By people. Never forget the training data is curated BY PEOPLE.)
It is very, very easy for the machine to generate mediocre art. A non-artist can easily confuse the technical execution of the image for art; if you struggle with painting skin, or clothes, or dramatic light and shadow, or anatomy and proportion, you will pull out the machine and say "do the boring derivative piece I am struggling with" and it will merrily crank out a fantastically rendered execution of your boring derivative piece. It is much better than anything you could do, because you have got a long way to go before you are a mediocre artist. You are making the art you make because it is the art EVERYONE makes when they start.
Machine art is relatively easy to recognise, not because the machine is bad at hands and faces - we can fix that, we're getting GOOD at fixing that, the machine can do the fix automatically if you operate it properly - but because the machine doesn't do anything interesting. The operator is still amazed at the technical proficiency of the machine; this person does not exist, but I have a photograph of them, and they can be doing whatever I want.
But what everybody wants is for them to be standing in the street looking at the camera. Like an Instagram post. Because the overwhelming majority of human artists get their portrait reference from Pinterest, which in turn gets all of its portraits from Instagram, so no matter where your training data came from the portrait data is hopelessly derivative of the average Instagram post.
There's no dramatic composition. There's no deep perspective or dynamic storytelling going on here. And when you start trying to make that happen, you rapidly get frustrated trying to make the machine do these things. The tools to make it happen exist, but they're still new and clunky and kind of a well-kept secret. Hardly anybody has combined these things in a way that they're using to generate well-designed images that might constitute art.
Instead, what we mostly have is people who roll the dice until something halfway decent comes out, then lock their seed and tweak the prompt until they like what they have. They pull out all the tools to clean it up, and then say "I've arted." But there's a missing piece here.
For the creator to be an artist, they had to know what they were making BEFORE they made it. They didn't. They just threw things at the wall until they said "that looks like a thing" and then pushed and prodded it into the thing. This limits them to what they might get throwing things at the wall.
And the machine generates nothing but relentless mediocrity.
What we are losing by leaning on the machine - by thinking it is an AI, that it has thoughts and makes decisions and exercises agency - is THE ACTUAL ART. The machine is being given far too much credit. It should be thought of like a camera. Your checkpoint database is the world into which you will point the camera; the prompt is the part of the world you will point at.
All of your tools and settings are just things like focal length and lens radius and F-stops, whatever the fuck those are, and the machine - IT'S NOT AI - just goes click and gives you whatever you are pointed at. In a twelve million dimensional space, so you can't even begin to conceive of HOW you are "pointed" at anything, but the machine is very very good at doing the maths.
Trouble is, YOU are not very good at pointing where you want to point. You're waving the camera around wildly and snapping a picture and looking to see what you got. That's all you can do, really, it's just how the machine works. Which is a fun little journey of discovery, but playing games doesn't create great art. It creates AGGRESSIVELY MEDIOCRE art.
Don't get me wrong, I think it is a great advance that people can go straight from the "cannot art" position to the "mediocre art" position. I also think it is eye-rollingly predictable that these same people are sticking out their chests and proclaiming themselves to be just as good as professional artists, and predicting the demise of art as a career path because the machines will do it all.
Do you know how much shit we hire people to do even though there are machines that do it? Machines we've had available for most of a century? Machines that are often sitting unused in a warehouse because a human being does the job better, primarily by just... not being a machine, and having abilities we don't know how to give to a machine?
Artists are not in trouble. Even the mediocre artists are not in trouble. Machine art is the MIDDLE of the process. Only an actual artist can do the beginning and the end of it. And yeah, doing the middle is most of the work, but if you ONLY do the middle your work is mid.
Machine art is gonna produce more artists, but it's also gonna produce an endless sea of garbage that people try to call art. Just like the camera. When you pulled out your Polaroid One-Step in the 1960s and said "I are a photographer too" everyone with a Nikon 35mm rolled their eyes and knew better. It's not about clicking the button. It's about pointing the camera, tweaking the settings, and developing the film. The button-clickers aren't artists and we all know it.
But by putting a camera in everyone's hands, literally everyone has the tools to be an artist, and that's good.
Hello Caliban, an odd comment. I've just read "Smart Start" 2008-2013 copyright. I probably bought in 2013. Cursing myself for not reading it sooner! Not sure whether to pledge or not. You seem to have moved into a different area. Where I'm back in a ready for a Smart Start position.