Exploring the Boundaries of AI: A Case Study in Fashion and Identity

As a creative professional, I've ventured into the world of AI, eager to test its capabilities for a client project. However, I thought, why not try it out for something more personal, like my Easter Parade look, rather than relying solely on traditional sources of inspiration such as Instagram and Pinterest?

After conducting some research and viewing beginner tutorials on YouTube, I discovered the MidJourney platform, which proved to be both user-friendly and accessible, even for someone with ADHD like myself. I started by generating a few images with basic keywords such as "Easter Bonnet, top hat, white flowers, high fashion." To my delight, the AI produced four stunning images, including a woman wearing an incredibly contemporary hat that looked like it came straight out of a future runway show.

Encouraged by this initial success, I refined my prompt with "Asian man, bright background, realistic", in order to generate a more specific look. I was impressed with the AI's ability to produce a finely featured, dark-skinned man with a beautiful floral hat -- but he was clearly African. Undeterred, I tried again, this time including a prompt of "light pinkish skin, Asian hair, Asian skin," and various other Asian identifiers. The AI once again provided me with the same handsome black man, with some modified features that included more Asian-looking eyes. Obviously it still wasn't quite the look I was after, but I was able to draw inspiration from the AI's creation, and use its images to communicate with a friend who was developing a complementary outfit. 

as seen in some of the publications

Thanks in part to the AI-generated look, the Easter parade outfit that I made was featured in several high-profile publications, including the New York Times, TimeOut, and the New York Post. The AI was a helpful stepping stone, but clearly with remaining problems. A week after I settled on my final look, I attempted to use the same keywords again, only to receive results that were borderline racist, featuring extremely stereotypical Asian features such as slitty and crossed eyes. Perhaps the AI is being influenced by implicit biases in its dataset. Despite these limitations, I remain grateful for the AI's ability to aid my work, and hope that it continues to evolve and improve in its representation of diverse identities.