I’m actually pretty worried about the implications of image generation by artificial intelligence. What is art, if it can be reproduced by a machine? What is an artist, if her work can be done by simply typing a prompt into an image generator and picking her favorite semi-random collection of pixels?
Well, I don’t think it’s time to panic quite yet. (Maybe next year.)
Adobe has Firefly, its AI image generator integrated into programs like Photoshop. I’ve been playing around with it a little bit. So far, it’s a fun toy, but I wouldn’t trust it for anything professional. Here’s a chronicle of my latest foray into the world of machine learning.
I started with Obi, my late aunt’s Persian cat. As she declined, his incredible fur got matted and my mom had to take him to a groomer to be shaved. Don’t worry, he’s actually super happy to be free of all those mats; his face just looks that way.

I decided to try to put his hair back on. That sounds like something AI should be able to do, right? “Take this cat. This one, right here. Add some hair to it.”

First, I selected the subject. That’s one cool thing about AI integration: it can figure out what a picture is probably about and sort out “subject” pixels from “background” pixels with reasonable accuracy. For that sort of thing, it’s an incredibly useful tool that I’m glad to have in my arsenal.

Next, I inverted the selection and turned it into a layer mask. I added some feathering and turned down the intensity. What we’ve got now is a mask over most of the pixels except a few of the cat pixels. The theory is that when I use “generative fill” on the cat, it will weigh the source material proportionate to the prompt I give it depending on the density of that mask. Let’s see how that works, shall we?
I selected the subject again and gave it the prompt “long hair.”

Cute cat. Not Obi.
I tried a few more variants of the prompt, various intensities of the mask opacity, and so forth. No joy. Even “white Persian cat with long hair” got nowhere useful.

Finally I got frustrated and bored. This was no way to honor my aunt, who was a very talented artist and who always encouraged me and who I miss every day. She’d laugh to see her kitty turned into a Furby, but that was really not the goal here. Well, if I couldn’t give Obi his fur back, maybe I could at least give him a reason to look so disgruntled.
So I used the AI to generate him a nice muddy beach to sit in. I used traditional Photoshop artistry to paint around the edges of the carpet with various brushes to push and pull at the opacity of the mask to make it look like part of the background.

The final result looks…okay. Not great. I’d like it better if I’d dug through my old photos and found an actual shot of Torrey Pines to use. But it’s not bad for quick & dirty.

I wish I could say that artists’ jobs weren’t in danger, except I know they are. AI is cheap, and for a lot of people, “good enough” is good enough. But it’s really not “good enough.” Nothing is going to replace actual human artists. Nothing is ever going to be as good as the real thing. The world is changing and we can use the tools we’re given, but we can’t let them replace us.
We’re too good for that.
Leave a comment