I checked out Playform.io for the first time in a while (since SIGGRAPH 2020). Whoa! All sorts of new features. You can train an AI-based off your own image set. You can even upload your linework! (For context, you basically scribble some white lines onto a black canvas, select a filter (ie: Vincent Van Gogh), and then it renders it for you.
I uploaded a scribble from my iPad, and it kind of worked. It rendered out 16 variations, but all of them sucked. This is where I remembered HOW I've worked with an AI in the past. It's all about iteration. Let me explain.
I usually start just by drawing a few lines, and clicking render. Within 10 seconds, it gives me a set of images. This is showing me how the machine is interpreting a few simple lines. It will start to fill in some details in the set. I'll then decide to add in a few other lines, either based on my own impulses, or based on what the machine is starting to show me. After another 10 seconds of doodling, I render my second set of images.
Eventually, based on what the machine renders, I'll start to see shapes emerge. It's kind of like "pareidolia," when you see dogs, or mountains, or Bernie Sanders in the clouds. When a gestalt (a recognizable form) comes through, I'll double down, and try to articulate that form through the line work.
So working with an AI to make a painting is a back and forth process. It's improv between a deep neural network and your own subconscious. I personally love shapes and visual composition, but I'm not great at rendering (painting & shading), so this kind of app is perfect for me.
Here's a look through a series of iterations:
1) Draw a line and a circle, see what comes up.
2) Add in a few shapes based on noise from the first render pass.
3) Look, a wild rabid fox came out of the machine.
4) I give it a tail, and it's starting to look like an animal.
5) I tried to give it more details to make it look more dog like, but it's not working.
6) Abandon ship, just scribble over the fox, and see what comes up.
After 7 iterations, I generated 144 different images. (In some works, this can be up into the 1,000s). These symbols have a way of etching themselves into your subconscious (you're seeing the same core form, but in infinite variations). I have weird dreams if I get caught in this process for more than a few hours.
Here's a set of images that came from this one round.
Here's what came from playing around tonight.
Onion house in a mostly deserted bush field; oddly spaced skyscrapers on the horizon
A heiro-glyphic caterpillar, standing on his back two legs like a dog, looking at his own reflection
Angel humming bird fights a drone out in Utah?
Grumpy triangle head pushes a wheel barrow in a sun storm
A bipedal fish mutant (3.5 million years from now) wears headphones and eye shades
Hybrid between a cave scribble and a peaceful landscape