Playing With Generative A.I. In Photoshop

Adobe has been slowly rolling out newer and more sophisticated Artificial Intelligence (A.I.) features in Photoshop. I’ve been playing with their implementation of Generative A.I. since it came out in the Fall of 2023. I decided to try and use it to create some “photo art” for my wife on Mother’s Day.

I started by taking some updated photos of our two Rottweilers – Thor and Athena. The photos were taken with my Canon R5 and the least used but probably best lens that Canon makes – the RF 85mm F1.2 DS. The photos were taken between F1.8 and F2.0. And I also used my K&F Concept Nano X Pro 82mm Variable ND filter to make sure I could get the correct exposure at the fairly wide open aperture settings I was using. Here’s a couple of samples of backyard photos I was getting:

 

As you might imagine, we have tons of photos of our dogs in the backyard. So I wanted to see if I could use Photoshop’s Generative A.I. features to put the dogs in a different setting.

My normal workflow is to make any necessary corrections/enhancements that are needed in Lightroom Classic before bringing the image over into Photoshop.  At this point, I’ll explain my process for Generative A.I., but there’s probably more than one way to skin this cat.  First I “Select Subject“ and then invert the selection which essentially selects everything in the photo except the two dogs . Now I click on the “Generative Fill” button and get the command prompt. I then describe what I want Photoshop to place in the selected area.  It’s more art than science. “Insert a mountain top at Sunset background.”  You can describe what you want in natural language and for the most part it does a very good job of interpreting what you tell it.  If I don’t get the results I’m looking for then I tweak my request a bit.  After some processing time Photoshop will serve up three different choices based on it’s interpretation of what you’ve requested. You can either select one of the options or you can hit the “Generate “button again and see if you get better results.  The results you get are quite random – so if you request the same thing multiple times, you’ll get different results.  This is both good and bad – good because it’s doubtful you’ll ever run into another photo using the exact same A.I. elements you generated in your photos.  Bad because if you do happen to get exactly what you were hoping to see, you’ll never be able to duplicate the results in the future.  Here’s the photo I settled on for my wife – which she loved:

Note that I didn’t ask for the golden grass that surrounds the dogs in the image – I wasn’t smart enough to do that.  Very impressive. 

WEAKNESSES IN ADOBE PHOTOSHOP’S GENERATIVE A.I.

Oddly, there’s one pretty big area where Adobe’s Generative A.I. doesn’t work well – text.  In one of the photos I took of our two Rottweilers, our male (named Thor) is looking away from the camera with a stoic and almost heroic look on his face.  So I wanted to see if I could create an Avengers movie poster type image, playing on our male dog’s name.  So I repeated the process I described above to get a background I liked.  Now I needed some “Avengers” text (I wanted to call their superhero team The New Avengers.”  But any request for text based results returns garbage results – the words are completely scrambled – it seems to be a common problem with all generative A.I. tools currently available.  The other issue is that Adobe’s A.I is fiercely protective of copyrights – so you can’t really use any brand names.  You could tell the the A.I. prompt, “put a wrist watch on the selected subject” and it would generate a generic watch.  But you can’t tell it to “put a Rolex Submariner” on the wrist of the selected subject.”  You could put a sports car in the background…but you couldn’t put a Porsche 911 in your picture.  Good for a commercial tool…but it significant’ reduces the fun factor.  There are other generative A.I. tools out there that are of course less ethical. 

Knowing the limitations of Photoshop, I had to do a Google image search for the Avengers Logo, and for Thor’s hammer.  I found what I was looking for and incorporated them into my movie poster type image:

Note that this is fair use of these two intellectual properties as I’m using them for educational purposes and not for profit. 

What really stands out to me is just how good you can, literally with just one click, select your subject.  Removing a subject from an image used to be so, so painful doing it manually – I was horrible at it.  It’s also neat how the generative A.I. has attention to small details like it will match the lighting to the subject, match the appropriate foreground and background blur, etc.  Notice in the two A.I. generated backgrounds, the dogs feet are obscured…because they were obscured by grass in the original image. 

I’m having a lot of fun playing with the generative A.I. tools and look forward to future enhancements.