AI-Generated Imagery Poses Challenges for Steampunk Creators
Top: Ada Lovelace, Frederick Douglass, Mark Twain, Benjamin Franklin, Abraham Lincoln. Bottom: Nellie Bly, Edgar Allan Poe, Geronimo, Mary Shelley, Harriet Tubman.
Wednesday, January 18, 2023
Do the images on this page look familiar? They’re all original and were created specifically for this story. But no doubt you’ve seen other images like them, because they were produced in Midjourney, an example of generative AI software.
In recent months, hundreds – if not thousands – of AI-generated steampunk illustrations have been shared on Facebook and other social media platforms. For reasons I’ll discuss later in this story, steampunk is particularly well-suited to generative AI, even if the concept seems antithetical to our subculture’s do-it-yourself ethos.
Before I go any further, let me state that outside of this article, you will rarely see AI-generated images on this website. At least that’s my hope. For now, my policy is to avoid use of such images unless they’re intended to help explain the technology. You might see AI-generated artwork in a story about AI-generated artwork. You won’t see an AI-generated illustration of Nikola Tesla in a story about the inventor.
A Quandary for Creators
If you’re a creator in the steampunk world, say a self-published author or independent musician, you may find yourself facing a similar quandary. The next time you need a book or album cover, you may be tempted to jump into Midjourney, Stable Diffusion, or one of many similar tools, and produce it yourself instead of paying an artist. Having toyed with the technology over several hours, I can attest that it is highly seductive.
But understand that many (though certainly not all) visual artists are freaking out about this technology. They fear with some justification that it’s an existential threat to their livelihoods. And the reach of generative AI is not limited to visual art: Similar technologies, in varying stages of development, are generating music, poetry, video, essays, fiction, and other forms of creative content. Visual artists may be most at risk, but others are not far behind.
If you don’t believe me, check out my interview with ChatGPT, a chatbot capable of remarkably human responses to questions, despite being in an experimental preview stage.
I write as someone whose chosen profession – journalism – has been decimated by earlier disruptive technologies, specifically the web and social media. These technologies didn’t eliminate journalism, but they upended the business models that publishers relied on for decades. Thousands have lost their jobs over the past 10 to 15 years, and though many of us have adapted as best we can, it’s been a tough slog.
So my policy regarding generative AI is largely an act of solidarity with all the other steampunk creators likely to be affected by this latest disruptive tech. If it seems like an overreaction, bear in mind that many artisans you meet at steampunk events have day jobs as professional artists and designers, where they’re likely to feel the biggest impact.
How It Works
The technologies described here are examples of what’s known as machine learning. To put it as simply as possible, they begin with a neural network – a computer system that loosely replicates certain aspects of biological brain function. Developers train the neural network to perform a task, at first by exposing it to a dataset. Then they use a variety of techniques to improve the AI’s capabilities, sometimes by using a second neural network to provide feedback for the first.
One example is a neural network that automatically colorizes black-and-white images. The initial dataset might contain millions of image pairs, one in color and the other in black-and-white. The neural network learns to associate visual forms with specific colors, so it can colorize images it’s never seen before.
Similar approaches have been used to create AI software that can perform a wide range of audio- and image-processing functions, such as sharpening images or removing background noise from audio files. Tools like this are not controversial – they’re designed for use with existing content, and vastly improve on conventional software.
Generative AI is different because it can produce entirely synthetic works that resemble human creations. Midjourney, DALL E, Stable Diffusion, and their brethren were built by exposing the neural networks to billions of online images that were accompanied by captions or other descriptive text. They learned to associate words with imagery, allowing them to generate new images from text prompts entered by a user.
The prompts can be simple, such as “Benjamin Franklin as a steampunk,” or more elaborate, like “photo of an extremely cute alien fish swimming an alien habitable underwater planet, coral reefs, dream-like atmosphere, water, plants, peaceful, serenity, calm ocean, transparent water, reefs, fish, coral, inner peace, awareness, silence, nature, evolution.”
Because the AIs were trained with online images, developers now find themselves contending with issues involving copyright and image ownership.
Some artists have claimed to see evidence of their copyrighted work in images produced by the AIs, and now some of the developers are facing lawsuits.
Getty Images, which has banned AI-generated artwork from its stock image services, announced this week that it is suing Stability AI, the company behind Stable Diffusion. The suit was filed in the UK, where Stability AI is based.
Three artists have filed a U.S. class action lawsuit against Stability AI, Midjourney, and DeviantArt. The latter, a popular online art community, has launched its own text-to-image service based on the open-source code in Stable Diffusion.
The lawsuits contend that use of online training images constitutes copyright violations because they were scraped from the web without consent of the creators. The developers claim that use of the images is permissible under fair use provisions in copyright law. Ultimately, courts will decide if the lawsuits have merit.
Generative AI is ideally suited to mashups, which is probably one reason why you’re seeing so many steampunk images produced by these tools. Steampunk itself is a mashup: Past and present; historical drama and science fiction; futuristic gadgets and old-time craftsmanship; and so on. Beyond that, steampunk art often consists of taking an existing work and adapting it to a steampunk aesthetic.
In the gallery below, you’ll see what can happen when someone with practically zero artistic skill (yours truly) gets his hands on these tools. I created these images over the course of four or five hours, using simple prompts like “Edgar Allan Poe as a steampunk.”
What the Boosters Say
To be fair, these tools are still in their early development stages, and I did not take full advantage of them. I entered a few words and left the results to chance.
Proponents of generative AI will tell you that it becomes art when you spend many hours with the software crafting prompts and refining the results. “The sobering secret of this new power is that the best applications of it are the result not of typing in a single prompt but of very long conversations between humans and machines,” wrote one of those proponents, Kevin Kelly, in an article for Wired.
Kelly described a new category of artists adept at creating these prompts. They operate “almost as directors,” he wrote, “guiding the work of their alien collaborators toward a unified vision. The convoluted process required to tease a first-rate picture out of an AI is quickly emerging as a fine-art skill.”
He made some valid points, but he also demonstrated what strikes me as naïve optimism, claiming that “not a single human artist will lose their job because of this new technology.” Instead, he argued that users of AI-generated art will be those who would never hire a professional artist in the first place.
I am skeptical. Professional illustrators may still find work in the upper echelons of publishing and corporate communications. But when you reduce the cost of producing digital assets, those assets tend to proliferate to the point where they lose value. That will have an impact on a large swath of commercial artists, especially those who serve smaller cash-strapped clients.
Out of Stock?
One area where I expect to see an early impact is stock image services. As a publisher, I’m as small as they get, and many images you see on this site are licensed from Adobe Stock or Shutterstock. Lots of small publishers rely on these services, and I could see many turning to generative art as an alternative.
Getty Images, owner of Unsplash and iStock, has taken a firm stand against AI-generated images. However, Adobe Stock and Shutterstock have both embraced the technology. In December, Adobe announced that it would allow AI-generated art under guidelines that include a prohibition on text prompts that mention artists’ names. In addition, all AI-generated images must be clearly labeled as such.
My no-AI policy extends to images on these services. But I can’t honestly say that the policy is set in stone, because no one knows what the future holds. What if AI-generated stock images become so dominant that conventional ones are no longer available? What if a top steampunk artist decides to adopt the technology? What if I want to feature a new book or album whose cover was generated by an AI?
I also want to see how the larger steampunk world reacts. Maybe people will become comfortable with the new technology. Maybe the fears of widespread job losses will prove unfounded.
The nature of disruptive technology is that it becomes inevitable. We can scream and moan about the negative social impact of companies like Amazon and Facebook, but we still rely on them.
Generative AI is far from that point even if the trajectory seems clear. So at least for now, with the exceptions noted above, AI-generated art is not welcome in this corner of the steampunk world.
AI Art Generators: Where We Stand (Graphic Artists Guild)
A Statement from The Society of Illustrators (The Society of Illustrators)
AI Generated Art: Copyright Critique (Arts Blog - Americans for the Arts)
The following links were added after the original publication of the article.
An interview with Midjourney founder David Holz (The Verge)
This article was updated with additional links and more information about legal issues involving text-to-image AI systems.