Am I impressed by Photoshop’s AI Generative Fill and Generative Expand?

All right, yes. Impressed, for sure, but not yet overwhelmed. There are both technical and moral issues with AI content that we all need to think about.

It’s true that I started Life after Photoshop with the intention of covering everything BUT Photoshop, but over time things have changed.

First, Adobe is no longer the only company pushing subscriptions. Capture One is steadily moving over to a subscription based system, ON1 Photo RAW arguably works best as a subscription package, especially if you want the online sync features, while Skylum is making it harder and harder to justify a perpetual licence over one of its ever-changing kaleidoscope of Luminar subscription deals. I know many folk hate them but I really don’t think subscriptions are all bad.

• Subscriptions vs perpetual licenses: pros and cons explained

The second thing is that we do occasionally need to check out what Photoshop is doing, if only to maintain a reference point for the current state of photo editing and the tools available.

And right now, AI is, of course the big talking point. Not just the AI masking in Lightroom, for example, or the AI noise reduction in DxO PureRAW. It’s also the ‘generative’ AI now appearing in Photoshop.

My generative AI images are no longer enhancements of the original scene, but fabrications built around it. They can still work very well as illustrations or artwork, but that delicate thread between ‘photography’ and ‘reality’ has definitely been broken.

RL

Generative Fill and Generative Expand

So right now, if you download and install the Photoshop 2024 beta, there are two types of generative AI you can try out. The first is ‘generative fill’, where you can select an area of an image and then prompt Photoshop to fill it with an object of your choosing. The second and newer option is ‘generative expand’, where you can extend the image area and have Photoshop fill the extra with what it thinks is background from the current image or another background based on your prompts.

The generative fill works really well – within certain limits. Adobe is very wary of trademark and IPR issues so Photoshop generates ‘generic’ objects which aren’t recognisable brands. And while it can generate people, after a fashion, it doesn’t pay to zoom in too close because their faces can look like something out of a horror flick.

I’m finding the generative expand option much more interesting. One of my many weaknesses as a photographer is that I crop too tight in-camera, so that if a website or an art editor wants an aspect ratio other than the one I shot in, I’m stuck.

But so far, the Photoshop beta’s generative expand feature has done a borderline miraculous job on every image I’ve tried it on. I’ve even used it to remove cluttered backdrops by cropping in tight on my main object and then expanding the image back out again and getting Photoshop to add its own AI backdrop rather than the one that was originally there. It’s a great way of creating simpler, cleaned-up images.

But where do you draw the line between enhancement and fabrication? That’s always been a tricky crossover point in photography, and if the AI revolution has done one thing, it’s to throw this into sharp perspective.

My generative AI images are no longer enhancements of the original scene, but fabrications built around it. They can still work very well as illustrations or artwork, but that delicate thread between ‘photography’ and ‘reality’ has definitely been broken.

The Photoshop Beta should be available to anyone with a paid Adobe All Apps or Photography Plan subscription via the Creative Cloud app.