Visual Pleasure & Neural Cinema: How can cinematic myths in AI video systems be identified, measured, and navigated?

This research investigates methods to identify, quantify, and creatively navigate the cinematic myths embedded in generative AI video systems. It is premised on the hypothesis that cinematic iconography, mythologies, and visual tropes—particularly those associated with heteronormative romance, desire, and pleasure—are encoded within AI video models. This hypothesis arises from the critical analysis of cinema as a site of ideological reproduction and the study of bias in AI systems. However, a fundamental challenge emerges: unlike traditional media, AI models lack a stable "text" for analysis, rendering established methods of close reading insufficient. This paper proposes a hybrid methodological framework to address this challenge. By adapting techniques from AI bias detection, such as prompt probing and latent space analysis, for the unique temporal and dynamic nature of video, we aim to make the model itself the object of study. The goal is to develop a framework for diagnosing cultural normativity in AI video systems, while simultaneously empowering artists to transcend their homogenizing limitations and forge new aesthetic possibilities.