The Persistent Flaw in Image Models: Time

It’s a bad time for AI image generators – literally. Most notable AI models, including the likes of DALL.E 3, Midjourney, Stable Diffusion, and Ideogram 2.0, are all struggling to generate analog clocks, often defaulting to 10:10.  When AIM tested DALL.E 3 with a simple prompt asking it to generate the image of an analog clock showing the time as 6:30, both of the images it produced were timed at 10:10. We even tried the recently released FLUX.1 AI Pro model with the same prompt and the results were similar to what DALL·E 3 produced. What Explains This Fixation?  At the root of this issue lie the datasets used to train these AI image models.  Most of the clock and watch images used to train these models come from product photos and advertisements.