Generative AI and the Landscape Photographer
“Not everybody trusts paintings but people believe photographs.”
Everywhere you turn these days, you will find breathless coverage of the advances in artificial intelligence and the great risks it poses to society. Within the landscape photography community, the discussions of AI's impact on this segment run a very wide gamut. There are those who not only believe that the use of generative AI will destroy all commercial opportunities for photographers but that the widespread use of AI will also zap the very soul of the art form. There’s no need to stand in the cold darkness with your tripod preparing to capture the rising sun over a perfectly aligned landscape when you can just ask your computer to create that very image. On the other side of the spectrum are photographers who see AI as a powerful toolset that can help them to create powerful images that not only make for happier clients but also helps produce ever-purer, improved landscape images. Unlike other creative segments that can only wonder what AI’s impact will be, the advent of powerful generative artificial intelligence is already a reality in photography.
So which is it? Do we stand at the precipice of landscape photography’s destruction or the opening acts of a new golden age? Well, after using some of the AI tools that are available to photographers today, I think we can dispense with all the talk of a photographic armageddon. Still, these generative AI tools, because of their ease of use and increasing sophistication, do create ethical and actual risks that can’t be ignored. As Ansel Adams said in the opening quote, people have traditionally believed in photographs. Photography, since its inception, has been a means of actually capturing the real world. What one sees in a photograph must be what existed in front of that camera lens…right? Well, any veteran photo editor knows that for decades people have been able to manipulate and manufacture photographs to a great degree. But now, with generative AI capabilities embedded directly into some of the most popular and widely used photo editing programs, the speed and ease of manipulating images or creating fantastic images with photographic qualities, has never been easier.
Generative AI - The Good (even Great)
Let’s start with a quick review of some of the great positives that generative AI tools bring to the landscape photographer. But first, let me establish the fact that every great landscape image we can think of today is an image that has been edited and manipulated. The notion of the pure photograph, unchanged from its initial conception within your camera is just not a reality. Since most of us shoot in RAW, editing is truly a mandate to create a vibrant image. We’ve all zapped dust spots and pieces of garbage in our photos. Even Ansel Adams, the patron saint of landscape photographers, was an inveterate editor of his images. If you compare his original negative of Moonrise over Hernandez with the heavily edited versions he shared with the world, it’s clear that Adams, as he told us, actively and enthusiastically edited that most famous image (great video). Now did Adams’ edits make Moonrise over Hernandez a less “believable” or pure image or did those changes make improve the photo? I think far more the latter. Adams’ edits that darkened the sky and highlighted the moon did not change the essence - the reality - of the image. No physical objects that weren’t present at that moment were added. No sky or moon was inserted from somewhere else. The changes that Adams worked to produce were changes to elements that existed at the time he pressed the shutter. Those changes heightened the photograph’s artistic expression without injecting unreality into the image. His edited image is believable.
“Dodging and burning are steps to take care of mistakes God made in establishing tonal relationships.” – Ansel Adams
While I was living in Old Town Alexandria, Virginia, I would wander the streets early in the morning with my camera in hand. During many of those photo jaunts, I would come across some of my favorite subjects being lit with beautiful natural light. And invariably those historic buildings or views would also include people’s cars parked bumper to bumper along the old roads. I used to pray for some sort of annual “Car Free” day in Old Town that would allow us to capture images of the historic buildings and cityscapes unmarred by dusty SUVs and BMWs. But alas that city proclamation never came, and my curses of those cars were always under my breath with every shot. But with the advent of new generative AI tools (ex. Adobe Photoshop Beta release 25.0), it’s not hard to remove those modern intrusions. See my before and after images below.
In the example above, Photoshop Beta did a fantastic job removing those cars parked near the Atheneum in Alexandria. Not only did Photoshop’s generative AI tools eliminate the four cars in this image but it also did a truly bang-up job filling in those spaces with imagery that was faithful to the scene. Huzzah. “No Car Day” is now a reality…at least virtually. Believe me, once I figured out that AI could do this, I spent some serious time “de-car’ing” a good number of my favorite Old Town photos. But did I cross some sort of ethical Rubicon by subtracting those vehicles from my image? Would my photo be less “believable” if I removed them? I must admit that this concern crossed my mind. No cars in Old Town? Pshaw!
The first consideration here is whether the photograph you are sharing is being used as news. As we will discuss later, I strongly believe that news photography should be held to a higher standard of authenticity. This photo was not news. Secondly, the eliminations I was making were of objects that were not inherent to the scene - they were temporal intrusions, much like an empty water bottle lying in the street. In a way, I came to see that Audi SUVs or Dodge Challengers as just some form of steel dust spots that should be removed. Their removal improved the focus of the image on the cobblestoned street leading to the historic Athenaeum in the background.
After figuring out how to remove cars, I went after people who had wandered into my shot (will no one stop him!). I’ve always liked this image of the ancient beehive huts atop Skellig Michael in Ireland, but all those tourists and their coasts and bags diminished the image. Removing those people would be a serious challenge, so I had set aside this image and never really attempted to edit it. But playing with Photoshop’s Generative AI tools, it was quite easy to take out those instructions - both people and backpacks - and thus create an image worth editing further.
Ok, removing cars and people from my photographs is one thing, but what about using Photoshop’s Generative AI tools to extend or expand the margins of an image? Here we are moving from the removal of “temporal” things that are not inherent to the scene to the full-cloth creation of something that was not in the image file. The example below is of the Berlin Cathedral. I’ve always liked the moody quality of this image, especially with all the birds flying around it. But this image has always bothered me a touch in that the cathedral was not very well positioned within the image. Using Photoshop Beta’s Generative Expand function, I was able to successfully extend the image to the left, allowing the cathedral to be more centered in the frame. Oh, yes, I also removed that blue scaffolding at the base of the dome (yes, another temporal addition removed).
So is it now ok to add things to your image that were not there in the original file? Didn’t you indicate above that adding things was a threat to the believability of the image? Hmm. In this case, I feel that the integrity of the image was retained since the added area was a continuance of the background, not something different from what existed. But you be the judge.
In summary, I believe that Generative AI’s ability to remove dust spots, trash, parked cars, scaffolds, and even people who are temporal intrusions into the permanent scene is a very positive development and should greatly help us create better, more focused images. I also believe that Photoshop’s Generative Expand capability, if used in moderation, is also a very helpful function that can restore balance to images. These tools are real game changers - fantastic improvements for all of us.
Generative AI - The worrisome
So let’s look at the other side… if there even is another side. Before I am accused of being a photographic Luddite, let me first say that there is absolutely nothing intrinsically wrong with using generative AI to create a piece of art with photographic qualities. Generations of artists have learned how to use new tools and media to push art in brave and exciting directions. I’m sure that this is already happening and we will see artwork created by generative AI that will be dazzling to behold. But if you use your computer to create an image with photographic qualities that you did not capture with a camera, or manufacture a scene that never existed, is that end-product a photograph? Even when a photographer seeks to capture an intentionally abstract or moody image, that photograph captures the light found at a real-world location and at a specific, actual moment in time. Images created whole cloth from generative AI should be called something other than a photograph and clearly identified. A Photographic Illustration? A Synthetic Photograph? A painting with photographic qualities?
Now what about hybrid images that include real-world locations augmented by AI additions? What about a so-so landscape image that you supercharged by using AI Sky Replacement to inject a brilliant sunset where only a bland sky existed before? Or what if you took a good base photo and decided to use generative AI tools to add into it significant people or objects that were not there? Here’s where I think we as photographers need to be thoughtful. I’m far more ok with a photograph that may blend together two or more images taken by you at the same location at different times than I am with a “photograph” that plucks an AI-created sky from who knows where and attempts to pass it off as a natural part of the captured scene. Again, nothing wrong with doing it, so long as it’s understood that what you’ve created is different from a photograph and is something that should be identified as such.
By far the most worrisome concerns about generative AI in the photographic world have to do with manipulated images used in “news” presentations. We are already well aware of how “Deep Fake” technology is allowing people to create videos with someone else’s image and voice - often for dangerous or nefarious purposes. Deep Fake photography is even easier. And in our current world, with passions inflamed by incendiary tweets or a short video clip, the creation of fake images designed to foment deadly mayhem is an all too real phenomenon. And as generative AI becomes ever more sophisticated, will we even be able to use AI to determine which images are real (believable)? Some recent news stories suggest this is already a losing battle. Show a fake image of an assassination of a popular political figure or a fictional nuclear devastation of a city; the snap human reaction could be a truly tragic one.
Fake news is a true concern, but is fake landscape photography something that should worry us? In my example below, I asked Photoshop Beta’s Generative AI tool to place two wading hippopotamus into my photograph of a creek in South Carolina. And to push things further into the absurd, I also asked the AI to add in some 4th of July fireworks - why not!
“No man has the right to dictate what other men should perceive, create or produce, but all should be encouraged to reveal themselves, their perceptions and emotions, and to build confidence in the creative spirit. ”
Now if I tried to pass this “Hippos and Fireworks” image off as a news story, that would be clearly wrong (Dangerous Feral Hippos invade the Lowcountry!). And if I attempted to pass this off as a most remarkable landscape photograph, I would be labeled a fraud. Share this image for what it is - a fun creation that used generative AI to create a fantastical scene - that’s completely ok by me. But between these three options lies a universe of gray that will take time to come to terms with. In the meantime, I plan to continue my crusade to “de-car” all of my photos. Welcome to my AI gray zone.
Rob Shenk, Mount Pleasant, SC
“To photograph truthfully and effectively is to see beneath the surfaces and record the qualities of nature and humanity which live or are latent in all things.”