AI Photography Is Testing the Limits of Artificial Art

After disrupting the world of illustrators, artificial intelligence is causing controversy with a new type of machine-generated content: AI photography.

While many artists were furious about the impact of AI on their authorship—both commercially and ethically—others are exploring the possibilities offered by AI photographs, hoping to escape the same risk of copyright infringement.

Instead of mimicking human-made drawings and illustrations with a high level of fidelity, like painting the Mona Lisa in the manner of Jackson Pollock, AI photography works by portraying non-existent subjects or events that never happened as if they were shot on a camera in real life.

Since their release this year, visual artists have opposed text-to-image AI art generators, like DALL-E and Midjourney, accusing them of theft and copyright infringement, claiming their work was used to train AIs without their consent.

But AI photographers, who mostly come from artistic fields other than photography, believe that their medium does not exploit other people’s style, as it consists of fake photographs of fictional subjects.

“As an illustrator first and foremost, I try not to use Midjourney to create works that are too similar to the pieces I make by hand. It feels wrong, and disingenuous,” artist Gossip Goblin, who specializes in fantasy illustrations, said to the Daily Dot.

“However, I appreciate the AI’s ability to mimic photography, and to generate realistic compositions of fictional people and places that would be extremely difficult to create otherwise,” he added.

AI photography exploded on Instagram in November, after the release of the latest art generators, with Midjourney v4 the preferred choice of many artists contacted by the Daily Dot. 

The hashtag #AIphotography on Instagram now has over 47,000 posts.

The medium attracts unique and disparate visions, ranging from retro-futuristic scenarios to nostalgic tributes to 1980s television and monsters, but they all have one thing in common. They couldn’t be photographed in real life.

More of a tribute or re-imagination of a specific period’s aesthetics and mediums, rather than an actual copy of an author’s style, AI photography might escape the same copyright concerns that marred AI-generated illustrations.

Gossip Goblin created a National Geographic-style series exploring an imaginary Asian country called Urumquan, where he fuses Japanese cybergoth with late 1980s Soviet aesthetics, without using specific artists as references.

Sam Finn (@Ai.s.a.m), a 3D artist who uses Midjourney to recreate a slightly off version of 1970s America, said to the Daily Dot that he does not command AI to work in the style of specific artists, individuals, or movies. 

He asks the software to replicate the output of a specific medium as if it was using “70s stuff like cameras, isos, filmstocks,” he added.

Artomaton works on his retro-futuristic series by patchworking many 1960s references, “primarily the 1964 World’s Fair,” he explained to the Daily Dot, “but also movies like Logan’s Run and early James Bond films, TV shows like The Avengers.”

As with other forms of AI art, a user needs to feed the art generator a text, or prompt, to create a picture.

In his prompts, Artomaton admits asking Midjourney to recreate specific styles, though not that of photographers.

“I sometimes use fashion stylist Pierre Cardin for the uniforms, and director Mario Brava for his lighting, architect Eero Saarinen for the backgrounds,” he specified.

Does that mean infringing someone else’s copyright? 

“Not at all,” said Artomaton, “because these names are mixed with each other and many other words I use in my prompts. The names influence some of the shapes, but the images don’t copy existing buildings, they simply evoke them.”

To obtain their preferred style in AI photographs, Gossip Goblin feeds art generators a string of specific terms like, “hyper-realistic, 80s mood, blockbuster movie,” and the medium, as “photographed on Arri Alexa, Super Panavision 70.”

These AI-powered photographers are indeed conscious of the ethical issues surrounding art generators but believe that by giving the output an original meaning as part of conceptual experimentation, they can reclaim its artistic singularity.

For Gossip Goblin, that means re-prompting the images several times until any resemblance with other artists’ styles is canceled.

“It’s possible that, for example, adding ‘blockbuster’ [to the prompt] will bias the composition towards Hollywood hits, thereby creating works that are more derivative, but this is typically offset by further modification,” he explained.

From this perspective, mimicking the output of a specific camera or the style of a specific period isn’t different from using a filter on Instagram.

Additionally, every picture on Gossip Goblin’s profile is accompanied by a text description that constructs a narrative, something that many of these AI-generated series have in common.

“The power of AI image production lends itself to, in my opinion, much more fascinating experimentations in storytelling and cultural collage than simple mimicry,” Artomaton pointed out. “I want to transport people to a past that never existed, evoking a future that was never meant to be.”

But that diving into the past can delve deep into the intimate, creating an uncanny world people may have not consented to being a part of. 

Part of the data that AI is trained on consists of people’s pictures uploaded on the internet, as made clear by Manufactured Memory, an Instagram profile run by veteran art director Ryan Wendell Bauer, who posts fake family albums created on Midjourney.

“I was fascinated by the idea of manufacturing new memories,” he told the Daily Dot. “I reckoned that if these neural networks were trained on millions of images from the entirety of visually recorded human history, our regular everyday memories were in there somewhere too.”

The concerns and the risk of impersonation make AI art and photography authorship controversial among artists.

Illustrator LRNZ is starting a campaign to police the companies behind AI art generators. He told the Daily Dot, “It is OK if you take an AI-generated image and give it a different meaning, but you cannot say that the image is yours.”

But Bauer believes differently, that these images and their usage can help bring back a time when the internet was less concerned about ramifications and more about people’s unique ability to use technology to bring about community. 

“I want to help people remember the way things used to be,” he said. “There used to be all these weird little corners of the internet, and the only way you could find them was to be turned onto it by some cool, weird friend.”