Hallucinated Pornography: AI, Synthetic Erotics, and the Reverse-Engineered Image of Desire

This paper explores the emerging aesthetics of AI-generated erotic images, emphasizing how hallucination, estrangement, and synthetic opacity replace the immersive realism and emotional transparency of traditional pornography. Pornography is viewed here as a visual regime centered around legibility, immersion, and bodily availability, while erotics is proposed as a mode in which desire develops through opacity, latency, and estrangement. Through a comparative analysis of works by artists Arvida Byström and Jake Elwes, I argue that synthetic erotics do not aim for direct representation or immediate sensuality but operate through ambiguity, atmospheric affect, and resistance to being captured. Arvida Byström’s In the Clouds mobilises AI-generated nude self-portraiture as a feminist strategy of opacity, pushing machinic codes of sexual legibility toward the point where they collapse into aesthetic estrangement. Jake Elwes’s Machine Learning Porn exposes the spectral traces of a neural network trained to classify and censor explicit content, rendering desire as glitch, residue, and machinic symptom. Both practices utilize hallucination as a technique, reconfiguring the erotic as a space of tension and delay. To analyze these images, I use reverse engineering not only as a technical metaphor but also as a semiotic method, revealing how erotic hallucinations interfere with systems of recognition, legibility, and algorithmic control.