Illustrator discovers her art was used to train an AI art generator

Originally published at: Illustrator discovers her art was used to train an AI art generator | Boing Boing


Being told “it’s not illegal” and “you can’t stop it” are always great signs that a technology is going to work out for the best for everyone. Heaven forbid we find a way to do this without exploiting people.


“I’ve invented a new way of being a dick that is so innovative that there aren’t yet any laws in place to stop me from doing it!” always felt like a weird flex to me.


Such a techbro take


I’m really tempted to train an ai on dilbert strips or somn


The other day I tried to use Stable Diffusion to produce an image that looked like it had been made by a graphic designer – a stylized illustration with lots of big blocks of bold color – so that I could more easily auto-trace it to convert to SVG.

I forget what prompt I used, but it worked well, except for one detail. The rendered image was covered with regularly spaced white text-like glyphs that looked eerily like … the iStockPhoto watermark.

Yep, the Stable Diffusion model was apparently trained on watermarked vector images from stock photo agencies, so when my prompt successfully nailed down that particular category, I got the watermarks too.


I’d like to order up a Batman comic in the style of Rembrandt, please.


It’s a start.

I’m kind of shocked no one seems to have painted Batman into the “The Night Watch”


Prompt: Scott Adams as the Pointy Haired Boss, art by Scott Adams, in the style of Dilbert comics


They were too cheap to pay for stock images so they scraped the previews instead.


I think “Boort Bos. .” should be Scott Adams’ unofficial nickname. On here, anyway.


I am torn on this:

Illustrator discovers her art was used to train (human/AI) generator.

At what point does this become plagiarism/derivative works as opposed to a transformation?
Are we going to have different standards for humans as opposed to AI? Should we?

Do we impose these same standards onto the (human/AI) that scours the Internet looking for infringements? Who is liable when the (human/AI) makes the wrong call?


as a graphic designer i know that the end is near for people like me, just as i saw a lot of old school illustrators and paste up artists get phased out when computers replaced hand-created art for production. i get it, i accept it – technology will one day make many vocations obsolete. i just hope that AI generated design is still not perfected for a few more years until i get to retirement age. not optimistic though! but your story did make me laugh, so thanks.


Graphic design can’t be obsolete because AI only interpolates, never explores anything outside its training set. It can be completely unacknowledged for its contribution though. :frowning:


His take was very practical: he thinks it’s legal to train and use, likely to be determined fair use in court, and you can’t copyright a style.

The question then is did he breach her copyright by using the images without permission? Unless the original artist waived her copyright, then she continues to own them and can reasonably restrict their use.


The question then is did he breach her copyright by using the images without permission?

If something is viewable on the Web, you have legal ability to train off it. This would seem to suggest that a lot of artists with publicly available portfolios – artstation, etc – will be going dark to all but vetted customers. We will all lose something when they do.


I am convinced that copyright laws and copyright squatting ironically do more harm than good for humankind. Moreover, humans are finite beings with finite limits, thus, even under the absolute strictest laws regarding copyright protection, it is literally, as well as is physically, impossible to truly enforce copyright laws and for two legal cases that present the validity of my comments, I present Exhibit A, which is Data East USA Inc. v. Epyx International Inc., and Exhibit B, which is Data East Corp. v. Capcom U.S.A., Inc. Those two court cases alone could, as well as technically should, invalidate any and all copyright claims, as well as copyright squatting, and copyright trolling. I could literally cite more legal cases, not just in the USA, but also worldwide, where courts literally tossed out the claims of both sides of the court case, as well as dismissed both arguments from both sides of the trial, as well as had also told both sides to go pound sand.

1 Like

Picassoesque. Inevitable.

Fancy Dali?


That “transformative” part of the analysis of fair use is often misunderstood. It doesn’t really mean that you altered the original work in some way, although that can be part of it. What it really means, though, is that the use is transformative. In other words, it serves a different purpose than the original. For example, if you are doing a critical review of a movie, and you include a clip of the original movie, your use is transformative, because it serves a different purpose than the original. The purpose of the original was to entertain. Your purpose is to critique. Thus, your use is transformative. Another way to look at it is can the the copy be used in substitution of the original. In the case of a movie review, generally no. The key here is what is the infringing use? It isn’t the art produced by the AI. The infringing use is the use of the original art to train the AI. That is likely a transformative use, because it’s clearly for a completely different purpose and character of the original. However, that’s only one factor in a 4 factor analysis used to determine fair use, and the engineer here is ignoring the other 3 factors. One of them is the nature of the copyrighted work. Creative works get more protection than factual works, is basically what that means. Here, the original copyrighted works are creative. That factor goes in Mengert’s favor. The third is how much of the original work you’re using. Movie reviewers use short clips, not the whole movie. Here, the AI engineer is using the entire artwork, and multiples of them, to train the AI. Another factor in Mengert’s favor. Lastly is the effect of the infringing work on the market for the original work. This one also seems to clearly favor Mengert. Three out of four factors in Mengert’s favor. If I were her, I’d consult a copyright attorney. I could be wrong, though. But I think she has a case.


No kidding. I wonder how many Disney, Illumination, Warner Brother, or Pixar images he used to train his AI.

Trick question, of course. He didn’t dare because those “creators”’ would squash him flatter than a bug in a hydraulic press.

I hope she can find good, pro-bono representation and go House of Mouse on the ahole.