This study investigates the intersection of surrealism and architecture through AI-generated image reconstructions of interior illustrations by Chilean artist Roberto Matta (1935–1940). Using automated inference and heuristic methods, Matta’s collages are translated into structured prompts that generate realistic interior images. These images are analyzed using atmospheric frameworks by Zumthor and Mestre, focusing on spatial configuration, interior form, and materiality. The results reveal hybrid spaces that merge architectural realism with surrealist ambiguity—biomorphic furniture, totemic stairs, and umbilical lighting—creating heightened tension between interior and exterior. By treating machine learning as a creative partner, the study proposes an experiential methodology that positions AI as a catalyst for rethinking architectural atmosphere and expanding the limits of design representation.
Lara et al. (Fri,) studied this question.