The process of generating a digital self-portrait by DALL·E was an insightful exploration of how AI interprets identity through textual prompts and visual elements. Though this final image represents a gothic, rebellious me that reflects my aesthetic and cultural identity, it is merely a relatively close match in terms of facial and fashion elements after generating more than twenty images. DALL·E actually failed to produce one that fully met my expectations.
For the background, I incorporated gothic and Harajuku fashion elements, metal chains, and an eerie graveyard setting to express my personality. These details were essential in aligning the image with my self-perception, particularly my understanding of crows (as symbols of both fortune in ancient China and misfortune in the European culture), death (graveyards), and the natural cycle of life (mushrooms). However, the background's aesthetic seemed to influence the facial features, giving me red eyes that I had not initially intended.
For the facial features, I provided detailed prompts emphasizing a "round face," "peach-blossom eyes," "voluminous curly hair," "no bangs," and "a Chinese." Despite these specifications, many iterations required manual adjustments. DALL·E exhibited a categorical aesthetic preference, frequently generating angular jawlines and Westernized facial structures instead of the round face I requested. I repeatedly specified thin lips, peach-blossom eyes, and double eyelids, yet the DALL·E often defaulted to thick lips, upturned eyes, and monolids. Even when using keywords like "freckles" and "blemishes," DALL·E produced flawless skin. Interestingly, DALL·E struggled to interpret "no bangs," often generating images with bangs. This illustrates the negotiation required between AI's interpretation and my intended representation, as well as the model’s reliance on entrenched data biases. DALL·E revealed aesthetic biases favoring Eurocentric ideals, promoting an idealized version of perfection that prioritized flawless, symmetrical, and conventionally attractive features.
This experience exposed both DALL·E's capabilities and limitations in self-representation. While AI successfully captured my alternative aesthetic and gothic style, initial outputs often misrepresented structural facial features, reinforcing the challenges of identity depiction through machine learning. This experiment also raised questions about AI’s role in shaping identity narratives: Does it merely reproduce existing biases, or can it serve as a tool for self-exploration?
Ultimately, the iterative revision process highlights the worrying complexities of digital self-presentation today. While AI is a powerful creative tool, it still demands continuous and prolonged human guidance to accurately capture the intricacies of individual identity. Even sustained intervention does not always guarantee success.
Comments
Post a Comment