Abstract
Recent advancements in latent diffusion models have enabled the synthesis of highly nuanced surrealist architecture that seamlessly blends classical lighting techniques—specifically chiaroscuro—with impossible spatial geometries. However, the perceptual and emotional impact of these AI-generated environments on human observers remains insufficiently quantified. In this study, we present "Algorithmic Chiaroscuro," an empirical framework investigating how luminance contrast and structural paradoxes in diffusion-generated surrealist spaces influence human visual attention and emotional valence. We recorded high-resolution eye-tracking metrics (pupil dilation, fixation durations, visual exploration paths) and self-reported affective states using the Self-Assessment Manikin (SAM) from 48 participants exposed to 120 AI-rendered architectural scenes. The stimulus set systematically varied across low-to-high chiaroscuro contrast and non-Euclidean spatial complexity. Our statistical analyses reveal that high chiaroscuro contrast in impossible spaces significantly increases mean fixation duration (p < 0.001) and pupillary dilation, indicating heightened cognitive load and aesthetic arousal. Furthermore, focal convergence on structural anomalies correlated positively with positive affective valence when mitigated by soft shadows, whereas harsh lighting contrast elicited negative valence and elevated anxiety responses. These findings demonstrate that generative lighting parameters directly modulate aesthetic perception in synthetic visual environments, providing quantitative design principles for neuro-architecture, virtual environment design, and affective human-AI interaction.
