IAEA 2026 · Image composition and aesthetics: Three principles manipulated, validated and tested

Lisa Koßmann, Milan Meulemans, Eline Van Geert, Stefanie De Winter, Michiel Willems, & Johan Wagemans
Biennial Congress of the International Association of Empirical Aesthetics 2026 · Jena, Germany

Compositional goodness is a key concept in aesthetics and art, yet it has rarely been operationalized and tested empirically in rich images. Here, we focus on three principles often mentioned to underlie composition: Balance, Emphasis, and Repetition. We directly manipulated and validated them separately and tested their effects on aesthetic preference. Using generative AI and image-editing software, we created 180 images from 60 different motifs, suggesting semi-realistic or fantasy-like scenes, each with a baseline version and versions with more or less of the targeted principle. We also collected demographic data, art interest and knowledge, and qualitative reports of task-specific decision criteria. Experiment 1 was a manipulation check (N = 37), in which participants ranked the three images within all triplets from least to most of the compositional principle (e.g., least to most repetitive). Plackett–Luce models showed that, in all but three cases, the manipulations produced the expected directional deviations from baseline. In Experiment 2 (N = 48), participants ranked the images with the same motif (within triplets) by either composition or preference. Global Plackett–Luce models revealed that the pairwise comparisons confirmed the expectations (all p < .001), with stronger effects for composition than for preference. Experiment 3 (N = 345) used a 2AFC task, with all pairwise comparisons across the three levels and motifs, in which participants had to choose either the better-composed image or the one they preferred. Item-level worth estimates derived from Plackett–Luce models showed a significant medium-sized effect of manipulation in the composition condition, but no meaningful overall effect in the preference condition. Data collection for Experiments 2 and 3 is still ongoing, but our current results show that the manipulation of compositional principles significantly affects participants’ composition judgments more than their preference judgements.