With a new machine, I thought it’d be fun to try out the trial of Photoshop CS5. The content-aware fill feature is the one that really draws everyone’s attention, and was indeed something I was psyched about trying.
For one, let me state that content-aware fill is terrible at some things. It’s probably things it was never meant to do, like filling in huge swaths of an image, or taking a set of photos of people and “content-aware filling” them with recognizable human forms. But it’s borderline-magic at filling in small patches with good-enough content that looks right.
So here’s a quick photo series showing what I did after a couple minutes of tinkering. I show a couple of my favorite tricks, too. I should note that there are several other images I tried that didn’t come out well, so don’t take this one example as representative. (But I thought my mishaps created without reading any documentation or even fully understanding the precise purpose of content-aware fill would be unfair to Adobe, too. “Hey, look, if I misuse a tool I don’t understand, I get terrible results!”)
Here’s an image that started out life as a magnificent vista up in the White Mountains area, shot from a car window, which practically guarantees that it will come out terribly. Indeed, it did:
For one, like every photo ever shot through a car window, it’s horribly washed out, which is generally somewhat irreparable. And two, you’ll note that I got a highway stake in the middle of the shot. Hard to plan those at 65mph when you’re on an old digital camera with tremendous shutter lag. In short, if I took a shot like this today, I’d probably have deleted it as soon as I saw how it came out on-screen. As you can see, I took the first step already and drew a really rough outline around the highway reflector/stake. I made no attempt to be precise.
Next, I selected Edit -> Fill, and took the default settings, which use the Content-Aware Fill. It worked pretty well:
If you really scrutinize it and compare to the original, you might note that it’s not perfect, but to a casual observer, it’s really not apparent that something’s been cloned out. Sure, you could have cloned it by hand; it’s not even that tricky in this case. But why bother if the computer can do just as good of a job for us? You might want to circle back and do a little manual-cloning, because there are some really distinctive elements that are notably duplicated, but they’re really not notable to casual observers.
The next two alterations aren’t specific to CS5, but they’re two of my favorite tools, and, coupled with Content-Aware Fill, really helped to save this image:
I think the image directly above actually looks worse than the one before it, but that’s because this is an intermediate step. I used an old favorite cheat of mine, Shadows & Highlights, to bring out some of the details in the shadows, and, to the tiniest degree, the highlights. This tool needs to be used with caution, because it’s easy to get carried away and push things to outrageously-artificial levels. You can ultimately push the dynamic range out a bit, but if you push it more than a little, the result is a monstrosity. That’s okay here, though, because the next image is going to make everything alright. Or, at least, as alright as it can be given the horrible starting point we had.
The above image is after the Levels tool. This image won’t win any awards; it’s really quite mediocre. But if you scroll back up and compare it with what we started with, it’s really pretty remarkable. If I weren’t playing with a new Photoshop trial at the time, I’d have considered the image hopeless.
Of course, the ultimate question at the end of this pseudo-review: Is it worth $700? To a professional, it’s a no-brainer, I’m sure. But $700 is a lot for a hobbyist photographer.