Photoshop’s beta Generative Fill

Like many people I’ve been giving Adobe’s AI-based Firefly and now Photoshop’s beta Generative Fill a good workover and thought I’d share a few words and examples. While there’s a lot of hype – and fear – about Generative AI, I’m still not sure I’ve been able to use it to produce anything that could fool anyone.

So for example, in the first group of images here, I can do better Photoshopping away the boxes myself, although to its credit I thought it was a nice touch that Generative Fill inserted a reflection of a traffic cone. Yet I did spend a lot of time typing in a variety of phrases for it to use to replace the area, and I got lots of artificial-looking results like the basket of roses (and I asked for four) until I tried puddles and hit on one that worked. So was the big benefit for me that the AI had access to stock examples of puddles? That might be valuable, sometimes, but is it really something to fear?

The second set of images came from Firefly, Adobe’s web-based tool, which I asked to create entirely-new photographs, sorry photograph-like pictures. I deliberately used phrases like “Lake District” to test if it would still produce scenery from New Hampshire rather than north west England, and wondered if using “Derwentwater” or “fells” might steer it to the right region. You can see the phrases under the pictures, and while there’s a lot wrong and some strangely-bred sheep, there’s also a lot that is indeed reminiscent of the Lakes. I can recognise an Ullswater and the rainy colour palette.

What practical use is this, I just don’t know!

But of course, the beta’s available and it’s not hard to try it and inform yourself, which has got to be way better than listening to all the hype.