Midjourney-generated image repository for bias detection

Description

A collection of images created by Midjourney using various promtps to understand the AI's biases in character concept art.

Steps to reproduce

The Midjourney® model 5.2, published in June 2023 (Midjourney, n.d.), was used for this study. As it was not intended to produce images in a specific style, the ‘Stylize med’ parameter was activated, which is the default stylization value used by Midjourney® and is equal to 100 on a scale of 0 to 1000. The ‘High Variation Mode’ parameter was also used in order to obtain a greater variation between the images generated by Midjouney®. The RAW Mode, Remix Mode and Sticky Style parameters were disabled, as they were not useful for this study. All images were generated in a square format with an aspect ratio of 1:1.

Institutions

Categories

Licence