June 2026

Photoshop's Content-Aware Fill Ain't Half Bad

Discussing generative AI is fraught^images. It's easy to make binary distinctions. Love it, hate it. No in-between.

I'm mostly against it for images. Fully admitting some look dope, the fact that it's making it harder for artists to make a living is too much^time. Not to mention the resource usage^energy.

There is one place I'm fine with it: background generation.

Behind the Scenes

I've taken to playing Magic: The Gathering. The crew I play with is spread across multiple countries. We use Tabletop Simulator^tts for our virtual play surface. The app lets you customize play mats and card backs. I'm collecting a bunch of images for them^playmats.

Take this awesome bear from The Smithsonian's public domain collection^bear.

A photo of a small, plaster sculputer of a bear made by Edward Kemeys sometime between 1871-1894. It looks a little like bronze with the bear sittting on a little platform looking down like it's looking into a river for salmon. The background is black at the top and behind most of the bear before transitioning to white on the surface. The image has roughly a 4 to 3 ratio.

Wonderful. But, not the right aspect ratio. Play mats are 16:9. I don't know what the bear is. Just that it ain't that. I'd been dropping images into the middle of black backgrounds to cover the size. Something like this:

The same image of the bear with black bars on either side of it to extend the width so the image has a 16 to 9 aspect ratio.

Not bad. Better than stretching the image. But, not great.

Then, I remembered Content-Aware Fill^fill.

Better Backing

Content-Aware Fill is a Photoshop feature. It attempts to use existing parts of an image to generate something that looks reasonable for new areas. It uses machine learning to figure out which pixels go where. "Machine learning" being what AI was called before AI became the buzz word.

A few clicks, and we get this:

A wider image of the same bear with the background extended to the sides to expand the aspect ratio to 16 by 9.

16:9 but with a background that matches the original instead of black bars.

Close Enough for Jazz

It' not perfect. The gradient between dark and light is splotchy on the left side. The hard edge of the shadow on the right maybe cuts off a little quick. I could clean it up. Make it impossible to tell where the original ends and the addition begins. That would take time and effort. Content-Aware Fill took a few clicks and about 20 seconds.

Hard to beat.

-a

end of line

Endnotes

A quick video showing how to use Content-Aware Fill.

More videos should be like this. 24 seconds of mouse movement showing you exactly where to click. Most videos aren't even out of their "Hey, Guy!" intro in that time.

Here's Kermit with another example of extending an image with a more complicated visual to match.

Original:

Kermit The Frog sitting on a round blue floaty on top of paper cutout waves. He's holding yellow sunglasses up in one hand and a red cup in the other.

Content-Aware Filled:

The same image of Kermit, but with the background extended with content-aware fill to make the image 16 by 9.

It doesn't do the greatest of jobs. The extended waves are all blurry and it emphasizes the darkening gradient of the background. I'll adjust the lighting a little but it's way better than black bars with nothing more than this.

Footnotes

^{images ⤴}

I'm referring here to image generation. Code and prose generation are separate topics.

^{time ⤴}

Check back in a few years. I have a feeling I'm in the same position from a couple decades ago when professional photographers refused to use digital/phone cameras. There's something different with AI since it's a different category of tool that makes the actual images, but still.

^{energy ⤴}

Everything I understand about generative AI is that it takes a lot of energy and

^{playmats ⤴}

My virtual play mats

The mats (and card backs) I'm collecting to use during rounds.

^{tts ⤴}

Tabletop Simulator

It make my fans spin up like all hell. Small price to pay for being able to play with a globally diverse group. The feature set built in for Magic is great. (It's also hackable. I haven't gone down that rabbit hole yet.)

^{bear ⤴}

Grizzly Bear Sitting from The Smithsonian

Just a little guy.

^{fill ⤴}

Content-Aware Fill

The marketing page focuses more on removing things. That works well too.