無題(潛域邊疆) Untitled (Latent Frontier) investigates the circulation and reproduction of the cowboy image through machine-learning reconstruction. The work reconstructs images from Richard Prince’s Untitled (Cowboy) using latent diffusion models. Low-resolution JPEGs of Prince’s photographs were downloaded from auction websites and computationally divided into thousands of sub-images. Each image was processed with the prompt “Untitled (Cowboy)”. Over 20,000 individual images were reconstructed into large-scale prints. The work retains the legibility of Prince’s originals while resolving into a statistical field of cowboy figures derived from LAION-5B, none of which is attributable to a single author. 

The work situates itself within a lineage of image appropriation. Prince’s Untitled (Cowboy) rephotographs advertising imagery from the “Marlboro Man” campaign. That campaign emerged from a 1949 issue of Life magazine featuring cowboy photographs by Leonard McCombe and was later conceived by advertising executive Leo Burnett to construct Marlboro’s masculine brand identity. The cowboy myth thus circulates through successive commercial and artistic functions. This iteration extends the circulation into the internet as compressed auction JPEGs.

As detail is degraded across multiple reproductions, the work enlarges the poor image and delegates reconstruction to a diffusion model. The system statistically reconstructs image detail in latent space, substituting pixel information with the dataset’s probabilistic understanding of “Untitled (Cowboy)”. The resulting surface aggregates thousands of Untitled (Cowboy) images, from heroic to explicit, revealing how the myth fragments at a dataset scale. 

Engaging with questions of authorship, reproduction, and technological mediation central to Walter Benjamin’s analysis of mechanical reproduction, Untitled (Latent Frontier) proposes a shift in the work of art from mechanical reproduction to statistical reconstruction. If Prince once deconstructed the American archetype, machine learning extends this logic on an automated scale. In this context, the cowboy functions not only as a subject but as a structural metaphor. The American frontier myth that once naturalised territorial expansion reappears in machine learning through large-scale data extraction and aggregation. The diffusion model’s reliance on mass image datasets reflects the logic of late capitalism, in which visual material circulates as an endlessly extractable resource.