I experimented with a couple of generative AI platforms to generate a selfie, and the final product of DALL-E
felt the most familiar. The generation by openart.ai.com felt stylistic whereas DALL-E aimed for realism.
In other words, it was the selfie that most resembled what I would share on social media.
This led me to question: what made the most “shareable” selfie the more correct representation?
Because realistically, none of them looked like me.
I prompted DALL-E by asking for “a selfie of a young woman, Middle Eastern features, round eyes, curly hair, one cheek dimple.” The last few descriptors were to add specificity, but it seems the model was hung up on the first two: young woman and Middle Eastern. After a few renderings, the AI model stuck to an ultra-feminine, ‘cutsie’ facial structure. In an attempt to guide it to a more accurate selfie, I prompted “more androgynous features.” I wanted to steer away from the animated princess-style features, but it just gave me a sharper jaw and made the picture quality slightly worse.
As I attempted to describe myself, I realized I often draw on observations made by others unto me; I am often mistaken for Middle Eastern and told I look like my dad. I deemed the DALL-E selfie as shareable because it happened to fit how others have perceived me; it was difficult to find descriptors that weren’t based upon others’ perception. This is not to say that they are wrong, but it left me with the idea that I wouldn’t know where to begin if it were left all up to me.
Comments
Post a Comment