Here are some helpful tips for adding, removing, or extending with Adobe’s new Generative Fill feature in Photoshop (beta). Warning: It’s about flying saucers.
What is Photoshop Generative Fill?
Adobe included Glowworm (AI generated art app) to Photoshop (beta) 12.4.6. It allows you to add, remove and expand your image by typing text prompts.
Or even not typing in text prompts, as we’ll see.
Generative fill attempts to match the perspective, lighting, and style of the original scene. In other words, it tries to create an image that also has the right shadows, reflections, lighting, and perspective.
AI-generated art is under fire. Some characterize it as a plagiarism tool because many systems mine millions of images in an attempt to generate their own images. Adobe gets around this by letting the model learn only from Adobe Stock images and public domain content with no copyright restrictions. In addition, as part of Adobe’s Content data feature, AI images created in Photoshop are encoded with an invisible digital signature that indicates whether they are human-made or the product of AI.
We’ve described what it is. We discussed ethics.
Now on to the fun stuff.
Generative fill: add, remove and expand
I’ve tried three general types of generative fill: add, remove, and extend. Yes, that means adding new elements (Bigfoot!), removing unwanted elements (ex-boyfriends or girlfriends?), or expanding your image to make it wider or taller. I’ll also give my opinion for instances you might find helpful while not creating “fake” images.
But I will also fake a few. You know. Because that can be fun.
Add to your image using Generative Fill
Of course I wanted to add things first. We’re all curious about this, aren’t we? First I tried to add “a very realistic looking metal flying saucer flying in the night sky”.
First I selected the part of the sky where I wanted my flying saucer to go.
Then I pressed “Generative Fill.” The field is empty. You can choose to leave it blank and let it try to fill in the area, just like it could with Content Aware. Or you can type a text prompt. You can see my “flying saucer” text prompt above.
This was the first flying saucer it generated. And also my favourite. It didn’t look as realistic as I had hoped, but it’s a flying saucer. In later experiments, I found that it didn’t seem to matter whether I put “realistic” or “detailed” in the text prompts.
But then again, who am I to judge? I’ve never seen a flying saucer from space. Maybe they look exactly like this.
You may also notice that in addition to the flying saucer, it also mimics the night sky around it.
Just for fun I thought I’d try a “very realistic jet plane”.
I have a jet plane. I don’t know if it was ‘very realistic’. The hind wings look quite crooked.
But nevertheless, this technology is a lot of fun and quite amazing. If someone does graphic design this can be extremely useful and easy, easier than inserting clip art.
For night photography, mainly because I have little interest in putting anything “fake” in my photos. But that doesn’t stop it from being fun.
Thought I’d try another generative filling. This time my text prompt was “A woman in western clothes.” It has achieved the above. This is fine on a smaller image or on social media.
Zooming in closely, however, revealed a woman with a rather molten face.
Remove part of your image with Generative Fill
I thought I might experiment with a night shot of an old abandoned Nye County Search and Rescue vehicle in the Nevada desert. I noticed that I had “blown out” the left part of the windshield and part of the headlight. They were overexposed by my light painting. It happens.
I could clone out these two spots or even use Content Aware. However, I was curious how Generative Fill would work.
To remove these overexposed hot spots, it was better not to leave a text prompt. This would make Generative Fill work more like Content Aware, only with the added benefit of being able to make use of Firefly’s AI-generated art as well.
Above are the results of my first attempt. The filling looks extremely realistic. Best of all, it did this without any effort on my part, and it looked better than Content Aware might have done.
I might consider using Generative Fill in the future to remove unwanted hotspots or minor issues like profanity graffiti or other unwanted elements. I suppose this means my photo can be marked with an invisible digital signature to indicate it was created by AI. This, of course, blurs the lines between AI-generated art and photography. On the other hand, on a purely technical level, many feel that Lightroom Denoise AI or Topaz Labs Denoise AI already do this.
What is your opinion?
Expand your image using Generative Fill
This time I thought I’d try expanding my image with Generative Fill. I have made a selection. I found that since this was a full size photo, expanding my selection by 30 pixels onto the original photo produced the best results.
This is what the image looked like after applying the above. You will see a vertical stripe in the upper left of the sky. It wasn’t perfect. But this can be solved very easily.
And this is what the image looked like when I added about 30% more to expand the image. About half of the image is no longer my photo!
This is quite remarkable. I found this to be much faster and easier than using Content Aware Fill. And with significantly better results.
Why would I want to extend my photo?
Artistically, I don’t.
However, when I print my photos as a wrapped canvas with many printers, I lose up to 1.5 inches on the sides, especially if I choose to frame it. That’s a big part of the picture that gets lost.
If I print it on paper or metal, I don’t lose 4.5 cm from the edge. But for many printers, a canvas wrap is very different.
Therefore, if I really want to print on canvas without cropping my image, extending the edges is best. One way to do this is by using Generative Fill. It doesn’t have to look perfect, but it can work wonders for this kind of print, especially if your aspect ratio doesn’t exactly fit the frame perfectly.
Bonus tip: how to save an image with a generative fill
Above, the initial “Save As” is rather limited. Photoshop (Beta) only allows you to save this as four file types.
However, selecting “Save a copy” offers the usual array of file types.
I have given some of my opinions above. I’ve found I have some “good use cases” for using Generative Fill to add, remove, and expand images. And to me, these scenarios don’t feel like I’m making a “fake picture.”
But not everyone has the same view on this, or on AI-generated art or AI in general. What is your opinion? What are some of the best uses for Generative Fill? Are you going to use it?