Digital Mirrors in Motion: A Tool for Measuring Gendered Movement in Short Videos

Abstract

Visual self-representation on social media is a central concern in communication research, as bodily posture and movement convey gendered social codes that shape audience perceptions and contribute to the reproduction of stereotypes. While previous research has identified gendered body-pose patterns in static Instagram imagery, contemporary platform communication is increasingly dominated by short-form video formats, where self-presentation unfolds dynamically over time. Building on Tsolak & Kühne (2025), who identified 150 body-pose clusters in Instagram imagery using OpenPose, this paper introduces an open-source research tool for analyzing gendered bodily self-presentation in video. Using the original raw keypoint dataset, we implemented an independent unsupervised learning pipeline to derive 150 pose prototypes through K-means clustering. These prototypes form the basis of a browser-based application that performs frame-by-frame pose classification using the MediaPipe framework. To bridge differences between OpenPose and MediaPipe landmarks, the tool employs a custom mapping layer that aligns live video data with the original clusters. The framework enables communication researchers to trace how gendered bodily performances unfold over time in short-form video content that now dominates social media engagement.