Creating With Generative AI: It’s More Than a Prompt
I recently read an interesting article (can’t recall where) that attempted to debunk the notion that the generative AI art making process was like sampling in music or collage in visual art. What can’t easily be debunked is the remixing aspect of the process, especially in Midjourney. As I’ve explored in several Medium articles, remixing as a modality is what drives sampling and collage. In music, sampling utilizes more originality to make a new song, whereas a remix uses more of the previous song to create a new version instead of a new song. The key characteristic of a remix is that it appropriates and changes other materials to create something new.
Although his name and work was not referenced in my prompt, my recent series titled “The Gathering” reminds some people of Aaron Douglas who was an African-American artist/muralist who played a leading role in the Harlem Renaissance. Douglas had a unique artistic style that fused his interests in modernism (ex. Art Deco) and African art. I studied his art, specifically the “Aspects of Negro Life” series as a young art student. I was interested in many Harlem Renaissance and post HR artists such as Romare Bearden, Augusta Savage, Jacob Lawrence, Elizabeth Catlett, Charles White, and John Biggers. Through them, I was introduced to the collage/quilt aesthetic, which features prominently in my art.
The polyrhythmic, pieced together aspects of many of these 20th century artworks bring to mind African American quilt patterns. Identifying and making use of these patterns and elements requires domain knowledge, which requires knowledge about the environment in which something was created. After posting my GenAI images on social media, I’m often asked “What prompt did you use?” Depending on the image, the answer is either simple (text) or complex (text + other content). For example, I recently revisited an image I created using Midjourney in 2022, that was inspired by the ring shout.
Before prompt-based AI generators were made available for public use I was using neural or image style transfer to create images (see above). Image style transfer is a subset of deep learning AI, which is similar to the way artists combine art styles (ex. Art Deco + African Art). This is where I began my generative AI art journey… and then came Midjourney.
The Remix Mode in Midjourney let’s you change prompts, parameters, model versions, or aspect ratios between variations. This form of remixing will take the general composition of your starting image and use it as part of a new Job. It can help change the setting or lighting of an image, evolve a subject, or achieve more complex compositions.
In the two image above, I used the same prompt but tweaked it a little to change the results. The colors and subject matter is similar but I added/changed the composition and effects. For the latter I added “full body” and “hyperbolic paraboloid” to the prompt for variations.
The collage/quilt aesthetic, inspired by artists such as Aaron Douglass and others is still there, as it my interest in portraying Black women in a different light (not objectifying the body). In the variations are the experiments with mathematics and science. The polyrhythmic aspects of the images tap into the domain (emic) knowledge shared by many Black artists who recognize and use these complex patterns.
Citing “Aesthetic Principles of Afro-American Quilts”, Elsa Barkley Brown refers to polyrhythms as a guidepost for viewing quilt work. Wahlman/Scully write:
When the colors of the strips are different from the colors in the row of blocks or designs, two distinct movements can be seen: one along the strips and the other within the designs … This represents a textile aesthetic which has been passed down for generations among Afro-American women who were descendants of Africans … Afro-American quilters do not seem interested in a uniform color scheme. They use several methods of playing with colors to create unpredictability and movement.
We (AI artists) can use mathematics to generate these patterns or we can reference our knowledge of remixing via Harlem Rennaissance murals, traditional quilts, ‘Dilla Time’, or Midjourney to embed our knowledge and experiences into the prompts we use to generate images. The results are like nothing that existed before but contains just enough of our histories, knowledge(s), and experiences to pay honor to the culture.