In yet another stunning article, this one titled “We Asked AI to Create the Joker. It Generated a Copyrighted Image,” the New York Times provided more examples of “memorized” visual works, which resulted from prompts, including some nonspecific ones, on Midjourney (5 examples), ChatGPT (2 examples), and Microsoft Bing (2 examples). The AI generated works are strikingly similar to images from movies and the cartoon characters Sponge Bob and Mario of Mario Brothers.
The examples were drawn from a study conducted by Gary Marcus and Reid Southern, who published their results in an article for IEEE Spectrum.
Interestingly, the mere prompt “popular movie screencap” produced on Midjourney a character straight out of a Marvel movie.
Midjourney had no comment, although probably the most egregious outputs were from that AI generator.
I haven’t had time to test whether these prompts still generate similar images. But, hopefully, the AI platforms have addressed the apparent problem of “memorized” works.