Animation, but not facial femininity, increases attention to female faces in male mandrills

Abstract Previous correlational studies in wild mandrills suggest that males may show greater socio-sexual interest toward less feminine-looking females, challenging the expectation that facial femininity enhances attractiveness. However, whether facial femininity itself causally influences male response remains unknown, as natural faces vary simultaneously in many visual and social attributes. We tested this question in semi-captive male mandrills using paired looking-time assays with synthetic female faces generated by a generative artificial intelligence (AI) model trained on mandrill face photographs. This approach allowed us to manipulate facial femininity along a calibrated axis while holding identity and other visual features constant. We also used the same framework to animate static portraits, testing whether dynamic presentation increases attention to facial stimuli. Contrary to predictions from previous field observations, males showed no detectable preference for either more feminine or less feminine female faces. By contrast, they looked longer at animated rather than static versions of the same faces. These findings suggest that facial femininity alone is insufficient to elicit a consistent visual preference in male mandrills under controlled conditions, whereas motion enhances the salience of conspecific facial stimuli. More broadly, our study shows how calibrated generative-AI stimuli can experimentally isolate socio-sexual signal components that are difficult to disentangle observationally.