Mustafa Alperen Ekinci^1,2, Daniel Kaiser^1,2; ¹Neural Computation Group, Justus Liebig University of Giessen, Germany, ²Center for Mind, Brain and Behavior (CMBB), Universities of Marburg, Giessen, and Darmstadt, Germany

In real-world situations, our aesthetic experiences are shaped by exposure to constantly changing visual scenes. Yet, most research on visual aesthetics has relied on static or simplified displays, leaving it uncertain how aesthetic experiences form and evolve over time in naturalistic settings. Here, we examined how the visual characteristics of continuous naturalistic and artistic movies influence aesthetic experiences as they unfold. In two separate experiments, participants watched either the documentary “Home”, presenting a broad range of real-life visual scenes, or the animated drama “Loving Vincent”, based on Van Gogh’s paintings and his life story. The two movies differ vastly in terms of style and context: while “Home” features a range of photorealistic natural scenes, “Loving Vincent” features artistic animations akin to dynamic Van Gogh paintings. During the movies, participants continuously rated aesthetic appeal by adjusting a response slider. To predict these moment-to-moment ratings, we trained linear regression models on image-computable visual predictors extracted from every movie frame. Predictors included low-level and higher-order properties, including the efficiency of coding in a deep neural network (DNN) model, color statistics, symmetry, and motion energy. For both movies, the trained models successfully predicted aesthetic ratings for held-out movie parts. Predictions were robust within individuals and across individual observers, suggesting that perceptual features similarly shape aesthetic experience across people. Reduced model variants revealed that DNN coding efficiency and color statistics contributed most strongly to the successful predictions. Interestingly, color similarity also allowed for successful predictions across the naturalistic (“Home”) and artistic (“Loving Vincent”) movies, highlighting that color preferences prevail across vastly different visual inputs. Our findings demonstrate that visual features reliably predict aesthetic experiences even for complex and dynamic stimuli.

Acknowledgements: Supported by the Deutsche Forschungsgemeinschaft (DFG), KA4683/6-1 (project no.~536053998); and under Germany’s Excellence Strategy (EXC 3066/1, “The Adaptive Mind”, project no.~533717223).