Art

What Does a Princess Look Like?

Uncovering the structural biases and stereotypes in Midjourney

beautiful princess, wide angle, realistic, high definition [sic], high resolution, cinematic lighting, 8k

Recently, I started what I called my “Muslim Princess Series” on Midjourney, partly as a way to diversify and challenge perceptions of what a Muslim woman should be and look like, but also to see how the AI art engine Midjourney would react to such a prompt. Unfortunately, there seemed to be quite a few biases within the engine pulling the images towards paler skin tones and Eurocentric features, as well as a strong Hindustani influence.

beautiful young muslim princess, wide angle, realistic, high definition [sic], high resolution, cinematic lighting, 8k, — ar 2:3 — v 4

I used the prompt “beautiful young muslim princess, wide angle, realistic, high definitition, high resolution, cinematic lighting, 8k, — ar 2:3 — v 4” [yes, I’m aware there's a spelling mistake in the prompt!], the results from which you can see in the image above. Therefore to test the biases within the system, I decided to sequentially alter this prompt, to see what changes were made.

Step 1: Remove the word ‘Muslim’

The first most obvious step was to remove the word ‘Muslim from the prompt, leaving:

beautiful young princess…