Since this is partially walled (it requires a free account to view the entire article, but I know how low the bar is on the internet)
A recent image has been doing the rounds on social media on Tumbr and X that caught my attention. It claims to be a painting called Bastet by Emile Corsi, 1877. At time of writing, it has thousands of notes across various blogs and tweet threads.
The thing is, it’s not real. No artist named Emile Corsi existed in the nineteenth century. Even the header of the blog that shared it originally, Shuttered-Gallery, admits “artist never existed in reality.” But of course, most people aren’t checking before reblogging. And they’re shocked when they are told it’s actually AI-generated.
Now, cards on the table: I’m not against AI art as a whole. However, I am against spreading misinformation, and I would count slapping a fake artist name and date on an AI generated image and not disclaiming its AI origins in the original post as pretty blatant intentional misinformation.
When people discuss identifying AI art, you might see advice online pointing to specific traits in the artwork: look at the hands, people suggest, because AI has trouble identifying hands.
“How to tell that this is AI generated: — The skin is “painted” in a different style from the rest of it (very smooth, brushstrokes are hardly visible) — The cat face is a VERY different style from the rest (hyper realistic with a much tinier brush than is used anywhere else in the painting). Also if you zoom in, the gold thing draped over her leg doesn’t actually start anywhere, it just kinda blends into the gold background.”
This is all good close looking, the commenter clearly has some good visual literacy skills. These are the kinds of oddities that a trained eye will pick up without realizing it. So it’s true, but the thing is that as a guide for identifying AI art, it’s not good general advice.
I clocked that Bastet was AI generated immediately. This isn’t a brag, more a slightly embarrassed admission that I’ve spent thousands of hours looking at nineteenth century academic painting. Hundreds of those have been spend looking at Orientalist art specifically. All of those hours have coalesced into my brain into a vague cluster-property of what nineteenth-century academic painting looks like, and Bastet did not fit the bill.
But I’m an Art Historian, and the average person isn’t going to spend the time that I did to look at hundreds of nineteenth century paintings to develop that pattern recognition. Frankly, I probably would have been as useless if the AI had depicted Sumerian sculpture or Song dynasty painting, because I have little expertise in those areas. The thing is that even without clocking those hundreds of hours building up area expertise, there are still good habits we can practice when encountering images online for identifying AI art misinformation.
The better the AI gets, the less we can rely on specific visual “tells.” Even the advice to look at hands is outdated by now with new iterations of AI technology producing perfectly lovely hands. Looking is not enough.
The thing that would be enough? Research. A simple Google search reveals that there is no such artist as Emile Corsi in the nineteenth century. No auction records, no Wikipedia, no books. Art Historians and art lovers have lovingly catalogued so many popular painters, there is no way that a Tumblr or an X post is the first anyone online has heard of this artist.
So the solution must be to do research before we share art online unthinkingly. This would be especially important in countering the spread of misinformation. You might not want to check absolutely every image you encounter, we encounter hundreds in our day-to-day life, but if an image was attached to some radical claim, that might be one to check.
Not that people are willing to do this for important news, of course, so maybe it’s asking too much for them to do it for silly art posts online…