When Adobe added Generative Fill to Photoshop it probably knew how big a deal it would be, but even so it’s been interesting watching everyone get to grips with the new tool’s capabilities.
If you’re wondering how you can start using Generative Fill we’ve got a handy guide for you, but if you’re instead hoping to get a sense of what you can easily achieve with it, I’ve run through a series of tests to give you a sense for what Generative Fill is capable of.
What can you do with Generative Fill in Photoshop?
To start off with, I wanted to give Generative Fill a chance to do some useful tasks on photos that I didn’t think would stretch it too badly.
Make small changes
I started things off simply, with this photo of my family cat from a few years ago – I’m happy with it, but thought it could be tidier on the left-hand side, and have a more roaring flame for some drama.
I used two fills on this – one around the fire to get it more roaring, and one on the far left of the frame to get rid of that dangling wire. The fire worked better than the wire, I’d say, which has left behind a weird warped space. Still, this doesn’t look too obvious and does enhance things.
Expand photos
Next up I thought I’d see how the feature fares with expanding a portrait-orientated shot into a landscape, with a couple of holiday snaps in moody black and white.
It’s interesting comparing the results – in both cases, they’re mixed. The street view has worked well on the left, but was incapable of filling out that cut-off car on the right, and left a visible seam no matter how many times I regenerated it.
The waterfall, meanwhile, is a smoother result but has altered things a lot to make for a way bigger waterfall that’s now pointing off to the right – and if you look too closely at the left things do fall apart.
Change the mood
Speaking of waterfalls, our next test involved three steps – could it first expand a tall colour shot, then change its character?
Well, the answer is basically yes – although again the seams on either side are pretty visible. The most impressive step by far is the addition of storm clouds, which do add a thoroughly different mood to things.
Next, I amped things up with more complex and hopefully challenging subjects. Take this shot of Kuthodaw Pagoda in Myanmar.
Asking Photoshop to clear the clouds from the sky leaving it blue left us with a weird, stunted reality – and asking for a sun to then be added only offered cartoony results.
Remove annoyances
This parade from Liverpool’s last Champions League win also offered interesting results.
Removing a big traffic light worked pretty well, but doing so again in the distance left a bit of a word soup on the bus. It’s the sort of thing an AI would call acceptable, but a human would spot quickly.
Add new things
Let’s turn to actually adding things into my images, though, another major selling point of the feature. This loch’s reflective surface was a challenge, to my mind, and as predicted many of the generated assets had no reflection when I asked for a plane flying over. However, the one I picked was impressively close, for all its jaggedness.
Similarly, the sunset marina changes impressed me by acknowledging that sunset and accounting for it (although the fills I rejected didn’t do the same).
Get weird with it
Finally, though, I know that people will want to see if they can get weird, spooky or uncanny with this feature – and you absolutely can. Adding a thunderstorm and the Slenderman to this image gave me some options that I frankly couldn’t use since they looked so disturbing. Best of all, asking for the Slenderman and storm to be removed left me with that final image, his slinky legs extending endlessly into the sky.
This should showcase that the tool Adobe has added is indeed pretty remarkable, but also that it’s consistently fallible at the moment. This is something it realistically shares with the likes of ChatGPT and Google Bard, but the clear upside to this is that it can make for some really outlandish and fun results alongside occasionally very useful tweaks.
