Flooded with AI-generated images, some art communities ban them altogether

Enlarge / An assortment of robot portraits generated by Stable Diffusion as found on the Lexica search engine.

Faced with an overwhelming amount of AI-generated artwork, some online art communities have taken dramatic steps to ban or limit its presence on their sites, including Newgrounds, Inkblot Art and Fur Affinity, according to Andy Baio from Waxy. org.

Baio, who follows the ethics of AI art closely on his blog, first noticed the bans and reported on them on Friday. So far, major art communities DeviantArt and ArtStation haven’t made any AI-related policy changes, but some voice artists on social media have also complained about the amount of AI art they regularly see on these platforms.

The arrival of widely available computer-generated imagery models such as Midjourney and Stable Diffusion has caused an intense online battle between artists who view AI-assisted artwork as a form of theft (more details here). below) and artists who enthusiastically embrace new creative tools.

Established artist communities are at a crossroads as they fear non-AI artwork will be drowned out by an unlimited supply of AI-generated art, yet the tools have also become particularly popular among some of their members.

Banning CGI art from its art portal, Newgrounds wrote: “We want to keep the focus on art created by people and not flood the art portal with generated art. by computer.”

Fur Affinity raised concerns about the ethics of how CGI models learn from existing artwork, writing, “Our goal is to support artists and their content. We do not believe it is in the best interests of our community to allow AI-generated content on the site. “

These are just the latest advances in a rapidly evolving debate about how art communities (and art professionals) can adapt to software capable of producing an unlimited number of beautiful works of art. at a rate that no human working without the tools could match.

Part of a larger debate on the ethics of art

An array of non-AI artwork used to form Stable Diffusion, assembled by Waxy.org.
Enlarge / An array of non-AI artwork used to form Stable Diffusion, assembled by Waxy.org.

The current wave of image synthesis tools allow users to enter a written description (called a “prompt”) and produce a corresponding image, almost like magic. The results often require cherry selection and dedication to be perfect, but with a skillfully crafted prompt, the results can mimic the works of human artists in sometimes breathtaking detail.

The most successful prompts often refer to existing artists and art websites by name, but rarely alone. Mixing artists can create innovative new stylistic blends. For example, here is the prompt that created the robotic woman in the center of the image at the top of this article in Stable Diffusion:

Beautiful tears! female mechanical android!, half portrait, complex detailed environment, photorealistic!, complex, elegant, very detailed, digital painting, art station, concept art, smooth, sharpness, drawing, art by artgerm and greg rutkowski and alphonse mucha (Seed 79409656)

The most popular image synthesis models use the latent scattering technique to create new works of art by analyzing billions of images. In the case of Stable Diffusion, these images come directly from the Internet, thanks to the LAION-5B database. (Images found on the Internet often come with descriptions, which is great for training AI models.)

Recently, Baio and AI researcher Simon Willison retrieved data from over 12 million images in LAION-5B and created a search tool that allows users to peek into a small but representative slice of the much larger whole. (You can also search for artwork in the LAION5B image set, or even your own name, in a demo hosted on Github.)

A search for Jeremy Lipking, a living art painter, returns results in the LAION dataset.
Enlarge / A search for Jeremy Lipking, a living art painter, returns results in the LAION dataset.

Ars-Technica

A few weeks ago, some artists started discovering their works in the Stable Diffusion dataset, and they were not happy. Charlie Warzel wrote a detailed report on those reactions for The Atlantic last week. With battle lines firmly drawn in the sand and new tools for AI creativity coming out at a rapid pace, this debate will likely continue for some time to come.

Comments are closed.