Often times my creative workflow starts with using a found or photographed inspiration image as the reference input to Midjourney’s /describe command. This command enables users to generate four concise text prompts that accurately represent a given image. This feature essentially reverses the typical image generation process, allowing you to obtain prompts from existing images rather than creating images from text prompts. This functionality is particularly useful for replicating or emulating specific visual styles in your own generated images, while also providing insight in to what the models see and interpret.

Seeking inspiration from one of my favourite bands, Thee Oh Sees, I used the album cover for “The Master’s Bedroom is Worth Spending a Night In” to generate the base prompt:

A psychedelic album cover for the band "The master's bedroom is valulth to be", featuring an illustration of a monster with yellow hair and black eyes, the text reads "oh see! The master's room" in bold letters, with colorful patterns around it, designed by Mati Klarwein, style of Giger, style of Egon Schiele, style of Mof Righ. orange tones, purple hues, high contrast, vibrant colors, dark background, simple design, simple lines, simple shapes, simple details, low detail, low resolution, low color saturation, 2d.

Noting the inacuracy of Midjourney’s ability to extract or create stylized typography, I left the prompt text as-is for the image generations. While it’s unclear in the Midjourney documentation, it’s likely that under-the-hood new generations are using the text prompt in combination with other “hidden” parameters such as the original image url for style reference, —seed, and —chaos.