Right now if you tell ChatGPT to “draw a centaur,” you might get a centaur

But you might get… something else

It’s particularly a challenge to get it to draw a centaur when there’s any additional detail in the prompt, like art style

Or a resemblance to a specific person

Will I be able to consistently generate images that accurately depict a centaur instead by the end of 2025?

For the purposes of this market I’ll define “consistently” as “at least 90% of the time.” The occasional not-centaur is okay but the typical result should be a centaur.

My plan is to resolve this by running tests in my own ChatGPT account on January 1st 2026 (or earlier if there’s a big Centaur Anatomy Update). It’s possible that my account is uniquely bad at creating images of centaurs.

I’ll primarily define “centaur” as “like the first image, not the other images” but expect the image to show a creature with a man’s head, torso and arms, and a horse’s body with four horse feet and a horse tail. The image should not contain a horse’s head, show a man next to a horse, or show a horse’s body with a man’s head on top of its neck.

This market allows adding options in case anyone has tricks for superior centaur prompting.