AI Art Tool
FOR THE LAST FEW MONTHS, Elle Simpson-Edin, a chemist by day, has been writing with her wife on a novel she describes as a “grimdark queer science fantasy,” which will be released later this year.
Simpson-Edin decided to experiment with illustrating the book’s content using one of the powerful new artificial intelligence-powered art-making tools, which can produce eye-catching and even photo-real visuals to fit a text prompt, while she was preparing a website to market the book.
However, the majority of these image generators are meant to limit what users may display, prohibiting obscenity, violence, and images of actual people’s faces.
Every idea she considered was too conservative. “The book is quite heavy on violence and sex, so art made in an environment where blood and sex are banned isn’t really an option,” Simpson-Edin explains.
Simpson-Edin was fortunate to come across Unstable Diffusion, a Discord group for users who use unrestricted versions of Stable Diffusion, a freshly published open source AI image tool.
Users upload sexual or horror-themed artwork and simulated pictures, as well as numerous images of naked humans rendered ugly by the software’s lack of grasp of how bodies should truly look.
Simpson-Edin was able to produce some appropriately sexy and violent visuals for her work using the unfiltered tool.
Other picture generators would not have been able to create these, despite their relative tameness and lack of nudity.
“The big selling point of the uncensored Stable Diffusion variants is that they allow for so much more freedom,” Simpson-Edin explains.
The world’s most sophisticated AI programs remain locked within huge tech businesses that are hesitant to give them unrestricted access—either because they are so valuable or because they may be exploited.
However, in the last year or two, several AI researchers have begun developing and publishing sophisticated tools that everyone can use. The development has raised concerns about the possible misuses of AI technology, which may be used for a variety of purposes.
Some 4chan members have considered utilizing Stable Diffusion to make celebrity porn or deepfakes of politicians in order to promote disinformation. However, it is unclear whether any attempt has been taken to do so.
Some AI art enthusiasts are concerned about the impact of eliminating guardrails from picture generators. Bakz T. Future, the host of a YouTube channel dedicated to AI art, believes that the Unstable Diffusion community is also making stuff that might be deemed child pornography.
“These are not AI ethicists,” he contends. “These are people from the darkest corners of the internet who have been essentially given the keys to their dreams.”
Emad Mostaque, an ex-hedge fund manager from the United Kingdom, launched Stable Diffusion in partnership with the Stability.Ai collective, which is working on a number of open source AI projects.
According to Mostaque, the goal was to make AI image production more powerful and accessible. He has also established a business to market the technique.
“We support the entire open source art space and wanted to create something anyone could develop on and use on consumer hardware,” he adds, adding that he has been blown away by the variety of applications people have discovered with Stable Diffusion.
Developers have produced plugins that bring AI picture production to current tools such as Photoshop and Figma, allowing users to rapidly apply a certain creative style to an existing image.
The official version of Stable Diffusion has guardrails to restrict the production of nudity or gore, but because the AI model’s whole code has been disclosed, others have been able to remove such constraints.
Mostaque says that although some images made with his creation may be unsavory, the tool has not done anything different from more established image making technologies.
“Using technology has always been about people’s personal responsibility,” he says. “If they use Photoshop for illegal or unethical use it is the fault of the person.
The model can create bad things only if the user deliberately makes it do so.”
Image generators such as Stable Diffusion may produce what appear to be authentic images or hand-crafted artworks of almost anything a person can envision.
This is made feasible by algorithms that learn to connect the attributes of a large collection of photographs from the web and image databases with the text labels that go with them.
| “The official version of Stable Diffusion does include guardrails to prevent the generation of nudity or gore, but because the full code of the AI model has been released, it has been possible for others to remove those limits.” |
Algorithms learn to create fresh graphics to fit a text prompt by adding and deleting random noise from an image.
Because techniques like Stable Diffusion employ web-scraped photos, their training data frequently contains pornographic images, allowing the software to generate new sexually explicit images.
Another issue is that such technologies may be used to make photographs that appear to depict a genuine person doing something compromising, thus spreading disinformation.
The quality of AI-generated photography has skyrocketed in the last year and a half, beginning with OpenAI’s January 2021 release of a system named DALL-E.
It popularized the technique of producing pictures from text prompts, and in April 2022, it was succeeded by a more powerful successor, DALL-E 2, which is currently offered as a commercial service.
OpenAI has always limited who may use its picture generators, allowing access only through a prompt that restricts what can be asked.
The same can be said about a rival service called Midjourney, which was introduced in July of this year and helped popularize AI-created art by making it readily available.
Stable Diffusion is not the first open source artificial intelligence art generator.
Soon after the original DALL-E was published, a developer created a clone dubbed DALL-E Mini, which was made freely available to the public and immediately became a meme-making sensation.
DALL-E Mini, subsequently renamed as Craiyon, retains guardrails identical to those found in genuine DALL-E variants.
According to Clement Delangue, CEO of HuggingFace, a firm that hosts multiple open source AI projects such as Stable Diffusion and Craiyon, having only a few huge companies dominate the technology would be problematic.
“When it comes to long-term development of technology, making it more open, collaborative, and inclusive is actually better from a safety standpoint,” he argues.
Closed technology is more difficult to grasp for outside specialists and the general public, he claims, and it is preferable if outsiders can examine models for issues such as racial, gender, or age biases; also, others cannot develop on top of closed technology.
Overall, he believes that the benefits of open sourcing the technology exceed the dangers.
According to Delangue, social media businesses may employ Stable Diffusion to develop their own algorithms for detecting AI-generated photos used to disseminate misinformation.
He adds that developers have also donated a method for attaching invisible watermarks to photographs created using Stable Diffusion to make them simpler to track, as well as a tool for detecting specific images in the model’s training data to delete problematic ones.
Simpson-Edin became a moderator on the Unstable Diffusion Discord after becoming interested in the game.
Certain types of content, such as photographs that might be regarded as child pornography, are prohibited on the server.
“We can’t control what people do on their own machines, but we’re very strict about what gets posted,” she explains. In the short run, humans may be more important than machines in controlling the disruptive consequences of AI art-making.
Making AI picture generators more widely available may provide individuals with an exciting new arena to explore. However, the technology has a major negative side.
