The Latent Space of Noise. The Prompt as Semantic Friction: Transduction Strategies in Text-to-Image Models

This essay investigates visual transduction in text-to-image generative models, proposing a methodological reversal: using the prompt not as a descriptive-denotative instruction but as “semantic friction” — an engineered condition for observing how generative systems select, suppress, reorganize, or literalize semantically dense material. A reflective first-person prose on contemporary noise and on Umberto Eco’s age of vision, information, and noise is administered to thirteen local and cloud-based models in a 1:1 format. Four recurring response patterns emerge: defensive transduction (aestheticizing removal), entropic transduction (collapse or historicization), conceptual transduction (metatextual decomposition), and didactic transduction (hyper-literalism and bodily somatization). Framed as a single-prompt, single-seed case study offering an exploratory grid rather than a validated taxonomy, the essay cautions against attributing algorithmic consciousness or hermeneutic intentionality to such outputs and reconsiders authorship in Generative AI Art, situating the artist’s activity between rigorous control of the computational environment and critical hermeneutics.