Not only have I been learning a lot about AI text-to-image generators and writing a post about the ethical issues involved, but I’ve also been thinking and discussing with collegues what we can do to solve those very issues as soon as possible – on both a large and small scale, and in both the long and short term. Here’s a little roundup!
Table of Contents / TD;LR / To-Do List
(1) Research (e.g. via artists or the press; links provided), exchange information and network across the board
(2) Support jurisdiction, e.g. through donations to pro-artist initiatives and / or complaints to commissions
(3) Educate and exclude, e.g. through comments in social media, letters to associations, competitions and press
(4) Protect your own work, e.g. via opt-out, blocking name for prompts (?), special watermarks and non-use of AIs
(5) Thinking ahead: How can we outsmart the bots?
(1) Research and Network
First of all, I think it is crucial to inform yourself about what is happening here and where exactly the problems lie. The starting point can be, for example, my first post, “No to AI-generated images”, the detailed article “AI Art Explained Easily: Development, Problems & Tips” by Sandra Süsser or the explanatory video “The End of Art: An Argument Against Image AIs” by Stephen Zapata Art. At the bottom of this post, you will also find a list of links!
At the moment, most of us pick up random information here and there, but there is no central point of contact. I would be so grateful if someone could set up such a central contact point, e.g. a Facebook group for the exchange of information! Unfortunately, I don’t have the time at the moment, but I’d be available as a co-admin.
(2) Support Jurisdiction
Legally, AI text-to-image models seem to operate in a grey area. Copyrighted images used for training without consent by and payment for the respective artists were collected for research purposes (Fair Use), but are used commercially now. LAION 5B’s databases is accessible, but for others, we don’t know what is included or where it came from.
I assume that sooner or later, there will be clear laws against training AI models with copyrighted images without consent and payment of their authors, but as we know, such things take time – probably years rather than months.
Donate for Initiatives to Protect Artists
To support and speed up legal action, you can donate to the Gofundmes of Mestieri del Fumetto (EU) and Concept Art Association (US). On their pages, they explain the problem and solutions very clearly.
Write to Commissions or File Complaints
In addition, you can write to commissions and decisionmakers that deal with the topic, and describe your concerns. Nicole Miller has written a thread on Twitter that is well worth reading, which shows the possibilities in the USA in particular. In the EU, as far as I understand it, complaints about practices or websites that are questionable in terms of copyright can be addressed to the EUIPO, and in Germany, according to WIPO, to the Ministry of Justice.
Participate in Class Actions Yourself
“On January 13, 2023, the Joseph Saveri Law Firm, LLP, filed a complaint in the U.S. District Court for the Northern District of California on behalf of Sarah Andersen, Kelly McKernan, Karla Ortiz, and a class of other artists and stakeholders against Stability AI Ltd.; Stability AI, Inc.; DeviantArt, Inc.; and Midjourney, Inc. This suit alleges copyright infringement, DMCA violations, right of publicity violations, breach of the DeviantArt Terms of Service, unfair competition, and unjust enrichment. It likewise seeks damages and injunctive relief to compensate the class for harms already incurred and to prevent future harms.” – Joseph Saveri Law Firm. As far as I understand it, artists may be able to join the class action, and / or register with Stop Consumer Harm.
In the meantime, however, AI models continue to develop at a rapid pace, learning from ever larger databases and more and more prompts every day. Right now, it is still relatively easy to tell (if you look closely, e.g. at hands, eyes, shadows) whether images were created by humans or generated by AIs, but the differences are becoming blurrier.
Soon we will be at a point where, from an entirely technical point of view, the usage of AI images in commercial projects – and even the replacement of illustrators with images generated in their style! – is no longer completely out of the question. From an ethical point of view, however, that would be more than questionable.
That is why we should do more right now than just watch and wait until legal action happens!
(3) Educate and Marginalise
Right now, it’s crucial to educate others (esp. artists, clients, associations, press, AI users) and raise awareness about the massive ethical problems of AI images trained with copyrighted images. And everyone can do this!
Comments and Discussions in Social Media
I’ve recently started commenting every AI image I’ve seen on social media, and actually had some surprisingly interesting discussions. As far as I can tell, there are a lot of misconceptions, but they can be cleared up.
The points that I frequently mention there are:
(a) Use of copyrighted images without consent by or payment of their authors to train AI text-to-image models
(b) Complete devaluation of art as a craft and profession, plus the role of art for future generations
(c) Possibility of legal problems due to the intentional or unintentional reproduction of copyrighted works
(d) Double standards compared to the music industry, which explicitly does not train copyrighted tracks
Contact Associations, Competitions and Portals
Many artists have already reached out to their professional associations and competitions, most of which have already spoken out strongly against AI images and explicitly exclude them. I hope portfolio portals will follow.
Letters to the Editor
In the press, on the other hand, there are still far too many articles that take a completely uncritical approach to AI-generated images, some even glorifiying them, and I think it makes sense not only to comment on these or write letters to the editor, but also to make suggestions for more critical articles that highlight the unethical practices.
When the ethical issues of AI-generated images become widely (or at least by colleagues, clients and press) understood, this should have the effect of making AI-generated images unusable – at least in commercial projects.
A social awareness of the ethical problem might not only influence jurisprudence, but also be an important step towards AI operators deciding to work with rather than against human artists, even before appropriate jurisprudence.
Ideally, we will reach this point before AI images become visually almost indistinguishable from human ones.
(4) Protect your Work
There are more and more possibilities to remove one’s own work from the databases used for AI training, or to prevent it from ending up there in the future – and I recommend everyone to make use of these possibilities!
Opt-out via “Have I been Trained”
I recommend everyone to use the option of searching for their own images via the image search (text search does not work so well) at haveibeentrained.com in the LAION 5B database, and to opt your images out of the database.
Block Names for Prompts?
In addition, I read that Midjourney is supposed to make it possible for artists to “have their name blocked for prompts” (Der Standard, “Zarya of the Dawn: Streit um Copyright für Comic, der mit Bilder-KI gestaltet wurde”, 22.12.2022), but couldn’t find out more. If someone knows how and where this can be done, please let me know!
NO AI (AI Watermark Generator)
Edit Ballai has developed a small app called “NO AI (AI Watermark Generator)”, with the help of which you can at least protect new images from scraping when uploading them, by adding an AI watermark that outwits the bots in an astonishingly simple way. The app is available for PC on GitHub, a version for Mac is to follow in the next few days. At this point, I’ll probably wait and see how it works for others first, but I think the approach is pretty amazing!
Don’t use AI Image Generators
Tempting as it may be, I would definitely advise against using AI tex-to-image generators yourself at this time, as they are trained further with each prompt. Instead, it might be helpful to tell the operators that due to ethical reasons, the use of their service is currently out of the question for you, and how unfortunate you find that :)
(5) Think Ahead
I sometimes wonder with colleagues whether there might be other ways to efficiently outsmart the bots…
The idea is that the bots themselves are rather stupid, and just learn from everything you put in front of them. In this case, that contains all our images, the associated descriptions (alt texts), and of course the users’ prompts.
How can we possibly outsmart those Bots?
For example, if text-to-image AIs were presented with enough images calling tables chairs, they might eventually generate chairs that look like tables… In a database of 5.85 billion pictures though, I assume it’d take a whole lot of false labels to create any visible effect! On a smaller scale, you could link a style that is not your own to your name.
The problem is that databases are not up to date; as far as I can tell, LAION 5B was scraped around mid/late 2021. So the only way to cheat ight away would be to exchange images under the adress of links that already exist there.
The effect of prompts on the learning of AIs is not yet entirely clear to me. If many users were to request a certain detail (e.g. tomatoes on the eyes :)), would the AI expect that other users would also want to see that, and would it start generating such tomatoes for them as well? And if so, how many such prompts would it take to see them?
It could be interesting (and maybe even a tiny little bit funny) to pick up on things AIs often get wrong on their own (f.e., number and placement of limbs, shadows, stereotypes), and to reinforce that with picture, text or prompt…
But after all, all I really want is for AI operators to keep copyrighted works and the names of living artists out of their databases, and for consent as well as payment to become mandatory.
Further suggestions are very welcome!
As soon as I hear anything else that I consider relevant in this context, I’ll include it here and in my Post on Facebook, and I’ll post again separately if necessary. This post is very welcome to be shared! If you like, you can use my respective posts on Facebook, Twitter, Instagram or LinkedIn and follow me there as well for further information.
I am writing this post to educate, help and share my own views. I’m happy to include fact corrections or helpful resources, but I have neither the time nor the nerve to discuss with anyone whether theft is okay. Talking about what AI is allowed to do or how useful it might be for creators is sth we can do once companies stop using copyrighted material for training. Under the current conditions, from my side, it’s a very clear No to AI-generated images.
About my Logo
I created the picture for this post based on Alexander Nanitchkov’s logo, which you might have seen on ArtStation and social media profile images. I would like to thank Eliseo for the wonderful idea of including the copyright © instead of the prohibition sign, in order to point out – beyond the platforms where AI-generated images should not compete with man-made art – this particular issue, as well as Bert and Kimberly for pointing out possible misunderstandings of former versions and Jens, Stephen, Ryan, Henning and Matthias for constructive criticism. Feel free to use my logo, edit it further and tag me when posting, because I’d really like to see how it evolves :)
Collection of Related Links
Here I collect not only sources mentioned in the post, but also other links that I find informative and important. The list is regularly updated (last update on January 5th, 2023), the latest additions are marked in colour. If you think of anything else that should definitely be listed here, I would be happy to receive a comment or a mail!
★ My first post about AI generated Images, “Statement | No to AI-generated Images” by Iris Luckhaus ➔
★ German Translation of this Post, “Statement | Nein zu AI-generierten Bildern” by Iris Luckhaus ➔
★ Detailed Video “The End of Art: An Argument Against Image AIs” by Stephen Zapata Art, 18.10.22 ➔
★ Blog Post “Why I’m Done Using And Boosting AI Art” by Chuck Wendig on Terrible Minds Blog, 8.12.22 ➔
★ Great video about the issues with AI art, “Why Artists are Fed Up with AI Art” by Sam Yang, 24.12.22 ➔
★ “AI Art Explained Easily: Development, Problems & Tips” by illustrator and coach Sandra Süsser, 27.12.22 ➔
★ “AI Is Coming For Commercial Art Jobs. Can It Be Stopped?” by Rob Slakowitz in Forbes Magazine, 16.09.22 ➔
★ “AI selfies – and their critics – are taking the internet by storm” by Tatum Hunter in Washington Post, 8.12.22 ➔
★ “Stability AI plans to let artists opt out of Stable Diffusion 3 training” by Benj Edwards, Ars Technica, 15.12.22 ➔
★ “The Alt-Right Manipulated My Comic. Then A.I. Claimed It.” by Sarah Anderson in NY Times, 31.12.22 ➔
★ “I asked Chat GPT to write a song in the style of Nick Cave and this is what it produced. What do you think?” ➔
★ Toolset “Have I been Trained” for finding and opting-out of Images from the LAION Dataset ➔
★ App that protects from futre scraping, “NO AI (AI Watermark Generator)” on GitHub (PC, Mac folgt) ➔
★ Gofundme “Help protect our art and data from AI companies” by MeFu / Mestieri del Fumetto (EU) ➔
★ Gofundme “Protecting Artists from AI Technologies” by Karla Ortiz, Concept Art Association (USA) ➔
★ Equity “Stop AI stealing the show: Campaign to strengthen performers’ rightsin response to the rise of AI” (UK) ➔
★ “Getty Images is suing the creators of AI art tool Stable Diffusion for scraping its content”, J. Vincent, The Verge ➔
★ Stable Diffusion Litigation by Matthew Butterick and Joseph Saveri Law Firm ➔, Case Website ➔
★ “AI Image Models Illegally Using Art without Permission in their Datasets”, by Stop Consumer Harm Reports ➔