Deciphering AI images -- Clues to look for. Where's the transparency?? (updated 2.25.24)

 As more people "like and share" AI generated images on line, the more important transparency and attribution becomes. Why is it okay to make and share these -- and then make the general public decipher the clues about what they are looking at?? AI generated images are here to stay. The madcap proliferation of these images comes at a price.

Social media thrives on image sharing. Companies profit by keeping eyes engaged on "content." This word is reductive. Creative work has value. It's produced with effort and experience. It's posted with pride and as a way to connect with fans. It's not provided for unauthorized commercial use. Or exploitation.

Social media has built a successful business model by appropriating artWORK, posted by creators, which then gets shared without credit or compensation -- or even a reliable way to trace back to the original source. Being able to trace back to the first post would  help the artists connect with fans who "like" their work as it gets shared around. Instead we have business models that work well for platforms. But not for people. 

Meanwhile, in cases like this "Gustave Klimt -- Garden Cat," the "credits" seem intentionally misleading. This image was posted on a public Facebook page. And the comments thread was quickly filled by fans calling out the use of AI. However, the attribution to the image never changed.

 All of the AI generated images are only possible thanks to copyright protected art that is scraped for datasets and used without credit or compensation. This is wrong. And NOW is the time to put guard rails in place. 

Artists are testifying in Washington, DC. Artist industry organizations are educating members, and lawmakers, to push for protections to vital creative rights.

More on these efforts here: http://stuartngbooks.blogspot.com/2023/07/testimony-of-karla-ortiz-and-others-at.html

The link below features a statement by artist Jean-Baptiste Monge on how AI generate images have robbed him of agency over his own name as well as his signature style of work.

http://stuartngbooks.blogspot.com/2023/09/jean-baptiste-monge-on-impact-of.html

Concept Art Association AI Town Hall -- August 2023. 

Lightbox Expo 2023 Panel on AI and Glaze Protection -- Oct 2023

Normalizing AI generated images is dangerous for everyone! https://stuartngbooks.blogspot.com/2024/02/normalizing-ai-generated-images-is.html

Is AI scraping of images "Fair Use" under current copyright laws??

The artists who own the images that are being used to build the AI data sets.. and who are testifying before lawmakers and striving to raise awareness at the impact of AI generated images on their livelihoods... don't think so. They consider what's being done with AI to fall outside the Copyright office's definition of fair use as "... a legal doctrine that promotes freedom of expression by permitting the unlicensed use of copyright-protected works in certain circumstances." 

While the courts work out the legal language of all this --what's happening to artists isn't fair. And it's up to art fans to be informed. 

Photo below features quote from artist Molly Crabapple. She has written about AI generated images in an Op-Ed for the LA Times: https://www.latimes.com/opinion/story/2022-12-21/artificial-intelligence-artists-stability-ai-digital-images

and here https://artisticinquiry.org/AI-Open-Letter This June 2023 interview w/ her is also very informative:

 https://www.techwontsave.us/episode/174_why_ai_is_a_threat_to_artists_w_molly_crabapple

This next link from the Copyright Office acknowledges "Fair Use: is a thorny issue: "Fair use is a judge-created doctrine dating back to the nineteenth century and codified in the 1976 Copyright Act. Both the fact patterns and the legal application have evolved over time, and you should seek legal assistance as necessary and appropriate."  In general, Fair Use allows using excerpts of copyright protected work for educational use or critical review. This link has an excellent primer on Fair Use.  https://www.copyright.gov/fair-use/index.html

Regarding the mis-attributed AI generated images we all need to be aware of -- Thank You to art history instructor Susan Rochester and others who contributed this link to the public Facebook page that posts many AI fakes:

https://arthive.com/publications/4056~Art_fakes_that_you_probably_came_across_on_the_Internet?fbclid= (below is a sample of an AI generated image that uses elements stolen from works by Henri Matisse, and has been shared often online, wrongly attributed as "Cat with Red Fish - Henri Mattise, 1953") None of this Title/attribution information is factual.

This post on the blog has more resources for basics on copyright and other creative rights: 

https://stuartngbooks.blogspot.com/2023/09/copyright-trademark-and-ai-image.html

If only AI could be trained to post with a disclaimer/attribution any time it is used. Artists have no no "opt out" for the dataset building/scraping process. The AI generated images produced can be captivating. But we must be discerning about what we see online. Especially when there is a suggestion of cultural context... or appropriation...

Another case study is these images identified as "watermelon art." Each had thousands of "likes" -- and were assumed by most in the comments to be carved from watermelons. Meanwhile, real artists who sculpt watermelons into decorative shapes called out these fakes in the comments threads. 

The clues are the paws!

We all have an obligation to view what we see online with a discerning eye. Be proactive. Be protective of creative rights. Have a higher expectation than what's easy. 

For now, AI struggles with depicting hands. Anything with digits (paws etc), often looks wrong. This remains a strong clue that AI image generation was used.

This 2023 Vox video explores "Why AI Struggles with Hands"

Another clue is proportion. Especially as you enlarge the image. 

Looking at captivating online images with a discerning eye is a critical first step we all need practice with. There are clues here that indicate this image has been manipulated. The biggest one is the scale of the flowers to the building is off. The proportions are fantasy level. If you enlarge the image there is a “@verbamystica” credit at the top. Once you enlarge the image the inconsistencies in the depictions of the windows also become more obvious. These images are fun to enjoy. But we all need to be mindful to decipher the clues. An online search for “who is verbamystica?” reveals an Instagram account for a digital artist.

Some images are posted with a camera icon, or other "credits," that attempt to validate the picture as original and unique. But usually there is no credit at all. Just a fantastical image. With viewers left to wonder... is that real??? 

Update 2.25.24 -- the Facebook Page "AI Spotting" posts tutorials on how to decipher AI images. Here are samples from that page:

Deploy the SIFT system. Stop. Investigate. Fact-check. Trace back to the original source.

http://stuartngbooks.blogspot.com/2023/08/sift-stop-investigate-fact-check-trace.html

UPDATE 9.18.23 -- You Tube video "AI art is going to have consequences" from channel TB Skyen circa March 2023. 

Check out this blog post on AI images being appropriated to sell non-existent products:

http://stuartngbooks.blogspot.com/2023/08/beware-ai-image-generated-non.html

Also this post on the Nov 2022 episode  "Why Society Hates Creative People (and what to do about it)" from the You Tube channel "Design Theory" http://stuartngbooks.blogspot.com/2023/08/why-society-hates-creative-people-notes.html