Negative Prompts and the Affective Economies of Failure in Visual Generative AI Systems

Abstract Every negative prompt is an algorithmic ‘thou shalt not,’ an act of negating and negotiating what “we do not wish to see” in a generative media output. This study identifies negative prompts in text-to-image generative systems as a crucial site where the digital error discourse becomes materially realised. Employing a multimodal digital ethnographic approach, this paper seeks to understand how these prompts extend beyond their instrumental use to see how they are entangled with the notions of failure and repair in AI systems. Through close consideration of this tool, the paper contends that generative systems sustain user engagement through an affective economy of failure: where errors are rendered ordinary and becomes the burden of users while a relentless update culture naturalises the logic of ‘versions-as-evolutions,’ ultimately cultivating a sort of cruel optimism among its users.