Why it’s so hard to get image generation models to create pictures of people with eyes aligned vertically, as explained by ChatGPT
(Last updated 3:15 pm CST February 19th, 2026)Me, after trying three times to get ChatGPT to create a picture of a man with eyes aligned vertically: That’s correct! Why was that so hard for you to generate in the way I described it? ChatGPT 5.2: Short answer: because image generation models are strongly biased toward human facial anatomy, and “removing” expected features is harder than “adding” unusual ones. Here’s the straight talk. 1. Human-face priors are extremely strong These models are trained on vast numbers of normal human faces. The learned prior is effectively: When you asked for vertically aligned … Read more