Generative Adornment: AI & the Representation of Identity
An Overview
According to researchers, Black/African American girls and their experiences around hair have been largely neglected in psychology and body image research, and more research on this topic is required to gain a better understanding of the role it plays in developing young girls. We, and by we I mean the Black girl I once was, have often experienced hair discrimination and bullying. I remember teaching art classes in a Cabrini Green (Chicago) elementary school and being asked by a Black girl student why I wore my hair in a natural style, as opposed to straightening it. I told her it was what God gave me. Then and now, I see a positive correlation between Black hair and identity development. This practice, I believe, can be enhanced and celebrated using generative AI.
For centuries, Black hair was a symbol of identity and social hierarchy. In pre-colonial Senegal (Africa) hairstyles often made it easy to identify a person’s ethnicity, age, marital status or fertility. Enslaved African girls and women are thought to hav used their cornrows as a communication tool to deliver messages. In 2021, Vogue published a piece featuring Ethiopian–American artist, Marian Mereba, who shared about her hair journey and how she presently uses her hair as a storytelling tool in her music videos.
Black women’s hair styles in Robert Pruitt’s artworks are reminiscent of Nigerian photographer Johnson Donatus Aihumekeokhai Ojeikere who spent decades capturing hundreds of magnificent, complex hairstyles, masterpieces of plaits and braids, astounding architectural compositions towering above women’s heads. The titles of the photographs usually refer to the place where Ojeikere took them or the style itself, such as Pineapple or Onile Gogoro — Yoruba for “tall houses,” which was a style that developed during the 1960s in post-independence Nigeria. According to a museum curator, the “tall house” hovering over an African woman’s head is an Afrofuturist symbol for an Africa that shatters clichés. In 2017, I began this development using generative AI, specifically Deep Style.
A Neural Algorithm of Artistic Style
At the time, what I called Deep Style (via Deep Dream Generator), was the output of a convolutional neural network, an AI-driven method that teaches computers to process data in a way that is inspired by the human brain. Deeper layers of neural networks extract more general features. In the case of the convolutional neural networks, the lower layers of the network extract very local features, such as edges, corners, and colors. In deeper layers, these edges and corners are combined together to represent increasingly global features, such as shapes, faces and styles. Researchers such as Leon Gatys developed an algorithm for this method, so when I say I use ‘deep style’ to make images this is what I’m referring to.
They argue that a mixture of lower layers of the network can extract textures and color schemes, while one of the deeper layers can extract the global structure of an image, and, that we can combine those features somehow in one image to perform a style transfer.
The ‘style’ I used for “Angel” (see top) was a vintage 19th century tapestry collaged with other images. Tapestry’s of that time were often woven with metallic fibers or threads and other adornments. I was fascinated with the word adornment that describes a method or way to embellish, enhance, or distinguish someone or something. Neural style transfer allowed me to apply this method to photographs of Black women and girls. Next, came text-to-image tools and user-provided textual prompts that serve as input for an AI model to generate an image. Recently, I used Adobe Photoshop’s Generative Fill tool to extend the sides of “Angel,” which was created years before using neural (image) style transfer. Generative Fill works by using an AI model (Adobe Firefly) to generate new content within a selected area of an image, either based on a text prompt or the surrounding pixels.
An Evolution of Style
The Whiting sisters had long, brown hair that their mother must have spent hours working with. Every three or four inches of hair, Ms. Whiting secured another barrette. When all finished, the Whiting sisters easily had 50 to 60 barrettes in their hair, creating a mosaic of dainty colors and shapes on a backdrop of lush, deep brown plaited hair. — Dana Nickson
Many Black women can recall a time when, as girls, our hair was embellished with ornaments or accessories (ex. beads, baubles and barrettes). A quick Google search results in a ‘tapestry’ of photos of Black girls with dozens of these colorful objects in their hair. These images come from websites salons offering to do hair, tutorials on how to create the styles, and from stores selling the accessories. This is how we have managed to hold on to ritual and practices from centuries before. It is a part of our heritage and has been passed down from the past to the present, to be cherished and preserved for the future.
To create “Barrettes & Universes” I started with a rear-view photograph of a Black girl with several colorful barrettes in her hair. Next, I uploaded the photo into ChatGPT and asked the model to create an image with the details I wanted. This is the result.
Next, I used Adobe Photoshop’s Generative Fill to extend the image. I added a prompt to inform the process. I imagined myself as a Black girl and I recalled my experiences doing Black girls’ hair and using the typical adornments. I also remembered what we wore (ex. blue denim jumpers with fringe). Generative Fill is not possible in the free version of ChatGPT.
I was not satisfied with Photoshop’s Generative Fill, so I used Midjourney to get the ‘blue denim jumpers with fringe’. Unlike ChatGPT Midjourney’s AI model is less ‘on-the-nose’, meaning the results are more varied. I used the Describe tool to analyze the image and generate words and phrases to describe the image. I chose one of the suggestions and made changes to the prompt based on my idea. This process gave me more images to choose from but my focus was the girl’s clothing. I did not like the hairstyles and opted to merge two images in Photoshop.
As a final step, I added a background that is inspired by Black American artist Alma Thomas who was known for her “color-filled, impastoed geometric abstraction(s) of tessellated brushstroke patterns.” The paintings have been compared to Byzantine mosaics and the pointillist paintings of Georges-Pierre Seurat. Some of the titles Thomas came up with referenced themes that can be found in Afrofuturism (ex. ‘eclipse’ and ‘Mars dust’). I wanted the Black girl with colorful barrettes to look into a far off, future place, so I added a colorful Thomas-inspired background.
In Conclusion
Adornment is heritage. My generative AI artworks draw inspiration from heritage, historical elements, craftsmanship, and traditions to create a ‘tapestry’ of images, many of which tap into a deep cultural lineage and my personal experiences. AI-generated images like “Angel” and “Barrettes and Universes” are created using multiple methods and tools. This includes the Gatys algorithm (neural style transfer, deep style), Generative Fill and text-to-image AI technology that uses large language models (LLMs). I also use digital collage or image compositing to create my final images. This process goes beyond prompt engineering and venture into new artistic practices and modalities, using generative AI.