Me, Myself and A.I.

I saw a prompt online that promised a cinematic tech portrait without the labor of building it manually in Photoshop. That was the appeal. I wanted the look, not the hours. So I tested it.

It started well. It looked stylized, synthetic, obviously AI, but still good enough to make me curious about what would happen if I kept going.

The first result frames the body through platform aesthetics. Floating music cards and soft interface glow place the subject inside a polished creator economy fantasy.

The visual language shifts from streaming platforms toward computation. The portrait is now embedded in circuitry, dashboards, and luminous data structures.

The model intensifies the cultural trope of the coder. Code fragments and technical overlays become the scene, and the subject starts being rendered through that stereotype.

That was the point where I noticed the face changing too. Not only the background. Me. The rendering got rougher. More masculine. The category was stabilizing.

The fourth prompt is the beard prompt. Even after explicitly asking for no beard, the model resolves the ambiguity by adding one anyway. That is the reveal.

When the model stabilizes gender

It took four prompts, not five. That matters because the beard was not some late accident after endless iterations. It arrived precisely when I tried to correct the drift.

I am a masculine-presenting lesbian. I am comfortable in that aesthetic. The model, apparently, was not. It translated masculinity into male facial hair because that is one of the strongest visual shortcuts available in training data.

Queer theory has long argued that gender is not an essence but a performance shaped by norms. Judith Butler describes gender as something repeatedly produced through expectations about what bodies should look like and how they should appear in public (Butler 1990).

Generative systems inherit those expectations statistically. Jill Walker Rettberg argues that generative AI can flatten cultural difference into dominant training patterns, streamlining expression into the forms most available in its data (Rettberg 2024).

The easiest statistical solution was simple.

Add a beard.