Prompt: Male student, standing, asks female professor "Can I get extra credit?" Professor responds calmly, "No." Generated image is a seated, smiling stick figure in a blue dress, facing forward, with no speech balloon at all. So, pretty much zero relationship between prompt and output.

Pretty good at rendering good hairstyles for "a professional looking Black female librarian consults with a 20 year old Korean American university student at a library help desk," except that half of the student images look more South Asian than Korean American. The student is always female for some reason. When the student was rendered with a face mask, and I appended the "Neither person is wearing a face mask" to the prompt, both people were shown with face masks. Books in the image were frequently shown cut in half at the spine. In several images, one or both people was shown wearing a name tag (I didn't zoom in to see the text, but I suspect given usual AI that text would be mis-rendered in some way) - and that continued even after I specified "neither person is wearing a name tag." I could not have used any of the generated images in any public presentation or web work.