Mirror Face: A Large Language Model-Based Workplace System for Emotion Regulation

Mirrors in the workplace provide opportune moments for pausing, inviting users to stop briefly and engage in moments of self-reflection. In this paper, we present Mirror Face, an LLM-based affective mirror system designed to facilitate emotion regulation in the workplace. Through emotion recognition, gestural interaction and generative feedback, Mirror Face transforms short office breaks into active emotional regulation experiences. We conducted two studies: a quantitative comparative experiment in simulated office environments (n = 37), followed by a field study of one week in realistic field settings (n = 12). We collected both qualitative and quantitative data to examine usage patterns, user experience, and emotion benefits. Our results show the effectiveness of Mirror Face in enhancing emotional regulation and boosting users’ pleasure and psychological arousal. Based on our findings, we discuss the role of affective systems in shaping the workplace ambience and derive a set of design implications for emotion regulation, focusing on iterative design and broader applications.