Gruber alerted me to the new beta of Photoshop with a compelling name: generative fill. The pitch: ask the robots to fill in parts of your existing photograph with, well, whatever you want.
Yes, I am not showing these initial trials deliberately. The first case I tried was inserting an object into a photograph. A snowman in a wintery scene. Did fine. I then asked it to change the color of a car in the foreground. Didn’t do well at’ll.
Take an existing photo, extend the canvas, select the empty canvas, and hit the generative fill button. The robots merrily fill the open area with what they believe belongs based on the existing photo. Generative fill shines when you don’t ask it to do anything specific at all.
Here’s a photo of my new bike:
I expanded the canvas, but before I hit generative fill, I wondered what the robots would add. I expected endless tree bark background, but what about the partial bike? Here you go.
It drew the goddamn bike. Click on the image and look at the larger size. It took a good swing at an appropriate color palette, and it drew the goddamn bike. Sure, it hallucinated the back of the bike into the tree, and there are other weird artifacts, but this is a beta. This is the first iteration. And this is 30 seconds of effort on my part.
Change is coming.
(p.s. Please note it also generated shadows for the entire imagined thing.)