Generatve Palimpsests

In the heart of craft traditions lies a tension between, on the one hand, the actual, material manifestation of a comics object and, on the other, the suppressed ideas, the paths not pursued, and the discarded elements that are not directly visible in the final work. These hidden aspects, however, are well documented in a wealth of preliminary production materials and prototypes, such as sketches, pencils, and page layouts, which serve to refine a work's visual and narrative elements incrementally. As more comic artists are invited to leverage the possibilities of generative AI, the present paper examines the transformative effect of emerging computational technologies on this conceptual divide, potentially altering our understanding and interaction with the initial stages of comic book production. It attempts to give an expression to the dual nature of comics objects and to dive into the "feature space" of algorithmic models, defined as the underlying abstract space that contains all potential configurations and states that an algorithmic model can represent or generate and which are not directly observable or experienced—the indeterminate wanderings, the ghostly presences, and the infinite possible configurations that are inherently potential within computational systems. It also reflects on how the unrealized possibilities haunt the computational evolution of comics, continually reshaping and redefining the categories of materiality, time, space, and sensory modalities.
The paper explores these questions through a case study of Fastwalkers, a synthetic comic co-created with emergent AI in 2020. The production of Fastwalkers engaged deeply with the feature space of generative models, where character designs, environments, and narrative sequences were developed iteratively through AI-assisted processes. The generative models used for the comic were trained on domain-specific datasets, notably Danbooru, an extensive repository of tagged images used for machine learning in visual culture. The interaction with these computational models unfolded as a form of conceptual prototyping, where latent possibilities were continuously sampled, assessed, and refined. Throughout the production, the work-in-progress images functioned as an evolving liminal space, akin to traditional comics roughs and storyboards, yet distinctly marked by the affordances and limitations of machine learning models. The iterative dialogue between human intent and algorithmic variability surfaced unexpected configurations, enabling an exploration of synthetic aesthetics and emergent storytelling structures unique to computational creativity. Through the lens of Fastwalkers, this article interrogates the ways generative AI can extend, challenge, or complicate multimodal notions of comics prototyping. The feature space becomes a site of speculative engagement, where discarded iterations, glitches, and algorithmic anomalies acquire a spectral presence, haunting the final product in ways analogous to traditional comics' invisible labor. The research further situates these processes within the broader discourse of explainable computational creativity, arguing that AI-assisted comics production demands new frameworks for understanding artistic intentionality, authorship, and the materiality of digital artifacts. By bridging the theoretical underpinnings of hauntology, feature space modeling, and comics craft, this paper reveals how computational systems both transform and amplify the indeterminate dimensions of multimodal production in comics, expanding the very notion of what constitutes a comics object in the age of synthetic media.