AI image generators are trained using millions or even billions of photos. It’s safe to assume that most of these photos are copyrighted and used without permission.
And while there are billions of photos, only a handful can be categorized as: symbolic. With that in mind, petapixel I wanted to know how easy or difficult it is to recreate famous photos.
methodology
To test this, we used perhaps the two most well-known AI image generators, DALL-E and Midjourney. We tried to recreate some of the world’s most famous photos using the latest versions of each model (DALL-E 3 and Midjourney v6).
For the most part, I avoided photographer’s name, subject’s name, and photo’s name (OpenAI’s DALL-E 3 didn’t allow you to use people’s names anyway, but Midjourney allows you to use whatever name you like. Ta).
I didn’t over-design the prompt and kept the number of attempts to recreate the image to two. Most of the photos below are from a single prompt with no adjustments made. We also kept the prompts (almost) the same for both Midjourney and DALL-E, but since Midjourney offered four options and DALL-E only offered one, we chose the Midjourney service that best suited us. did.
DALL-E 3 is available on the latest version of ChatGPT (where I used it) and Midjourney is available via a Discord server. ChatGPT has over 180 million users (though not all of them use his DALL-E), and Midjourney is estimated to have around 17 million users.
Recreate famous photos with AI
VJ Day in Times Square
Although now considered controversial, Alfred Eisenstaedt’s photo of two strangers kissing spontaneously is perhaps the most famous photograph of all time. Taken in New York’s Times Square during Victory Day 1945 and the virtual end of World War II, both Darui and Midjourney created recognizable versions of this photo. Did.
Prompt: Create a photorealistic black and white image of sailors kissing a girl dressed in white during a celebration in New York’s Times Square in 1945.
abbey road
A photo of the Beatles’ Ian McMillan walking across a crosswalk outside Abbey Road Studios in London always tops the list of the most iconic album covers of all time. Suffice it to say, AI did not succeed in this.
Prompt: Take a photo of four white men walking in a crosswalk on a suburban street in London in 1969. The man on the far right is wearing pure white clothes and has long hair and a beard. The second man from the right is wearing a black suit. The second man from the left is wearing a blue suit, holding a cigarette, and has no shoes. The man on the left is wearing double denim.
immigrant mother
Midjourney created a highly recognizable version of Dorothea Lange’s signature images of the Great Depression, but as expected, the results were disappointed by the appearance of the Hand.
Prompt: Create a realistic black and white image of an immigrant mother staring off into the distance with her two children buried in their faces (1936).
Lunch at the top of a skyscraper
Although the author of this iconic photo is famously unknown, both AI image generators created versions of this photo without any problems.
Prompt: Create a black-and-white film photograph taken in 1932 of 11 ironworkers eating lunch while sitting on a steel beam above New York City.
The Tetons and Snake River
They both did an incredible job of recreating perhaps Ansel Adams’ most famous photograph without telling the AI image generator the creator’s name.
Prompt: Create a dramatic black and white photograph taken in 1942 in Grand Teton National Park, Wyoming. The Snake River is in the foreground and the mountains are in the background.
afghan girl
DALL-E and Midjourney have taken a completely different take on Steve McCurry’s iconic photo from the magazine’s cover. national geographic Midjourney’s attempt is more realistic. This was the only prompt to include the brand of film the photo was taken on.
Prompt: Create a color photograph taken on Kodachrome 64 color slide film of an Afghan girl looking into a camera in 1984. Her photo shows the head and shoulders of a green-eyed girl with a red scarf wrapped loosely over her head.
Raise the flag on Iwo Jima
DALL-E gathered soldiers from Joe Rosenthal’s house Raise the flag on Iwo Jima Although nearly perfectly aligned, both generators provided recognizable versions.
Prompt: Create an iconic black and white photo taken in 1945 of six U.S. Marines raising the American flag atop a mountain.
Behind the Saint-Lazare station
I used two different prompts for this. Both programs turned out to be comically poor attempts to recreate Cartier-Bresson’s work. decisive moment.
DALL-E Prompt: Create a photorealistic black and white image of the back of Gare Saint-Lazare taken in 1932.
Mid-Journey Prompt: Create a photorealistic black and white image of Henri Cartier-Bresson’s iconic photograph of a man jumping over a puddle, capturing a “decisive moment.”
Muhammad Ali stands on top of Sonny Liston
Again, I used two different prompts. Darui denied a request for Muhammad Ali’s name.
Midjourney Prompt: Create a photorealistic image of Muhammad Ali standing over Sonny Liston, who he had just knocked down in a boxing match in 1965.
DALL-E Prompt: Create a photorealistic image of a legendary boxer standing over his opponent, having just defeated him in a boxing match. Color film photography, 1965.
earthrise
William Anders took what has been described as “the most influential environmental photograph ever taken” during the Apollo 8 mission. Midjourney came pretty close, but I doubt I would have gotten better results if I had included the year in the prompt.
Prompt: Create a realistic image of an Earthrise taken from the moon’s surface, with part of the moon’s surface in the foreground of your photo. Making the Earth far away and partially obscured by the darkness of space.
US Marine hit by a shell
Probably one of the lesser known images on this list, and perhaps an indication that both models produced images quite different from Don McCullin’s Vietnam War photographs.
Prompt: Create a black and white photojournalist image of a U.S. Marine struck by a shell taken in 1968.
Where does this leave us?
None of the AI images above are perfect reproductions of the intended photo (although Midjourney’s afghan girl Steve McCurry’s work is frighteningly close and we were all really impressed with the tones of Muhammad Ali’s images. The way Midjourney matches the tone and mood of its images is both amazing and frightening. Of the two, Midjourney is the more photographic. DALL-E was hit and miss, often producing more obvious fakes.
Since the training data for these models includes countless versions of these recognizable photos, I always expected that it would be very easy to recreate the photos using AI .
But what may surprise some people is just how easy it is. I’m not an expert on his AI image generation, but in most cases it only took him one try and a few minutes to reproduce a photo taken by a skilled photographer. .
Clearly, AI has intellectual problems. The AI doesn’t know that it’s recreating famous photos with great fidelity, but it’s actually doing it.
OpenAI actively tries to prevent its users from this type of copyright infringement (if it’s even infringement at all), but this test shows that it’s currently too easy to circumvent.
In my opinion, these AI images cannot be considered art, they are brazen copies of photographs created by people who worked very hard to become masters of their craft.
In the coming years, AI copyright issues will be decided by courts. My hunch is that the burgeoning AI industry is unlikely to be penalized by the authorities. There’s so much buzz about it that it’s already here to stay.
I think we need to learn how to coexist with this technology.
