Emotion regulation is a fundamental process that shapes cognitive, affective, and behavioral responses to emotional stimuli. Traditional emotion regulation models conceptualize regulation as a sequential modulation of emotional responses. However, they do not fully explain how emotions are constructed in a way that allows such regulation to occur. Predictive processing (PP) provides a mechanistic framework for understanding emotion generation by proposing that the brain minimizes prediction errors (PEs) to optimize perception and behavior. Yet, standard PP accounts reduce valence to PE minimization, failing to explain how PEs can generate different subjective feelings. To address these limitations, we propose a valence-focused model of emotion regulation that integrates predictive processing with self-referential cognition. We incorporate emotional valence as interpretative processes of the self-model, which assigns emotional significance based on goals, values, and autobiographical context. This model bridges the gap between emotion generation and regulation, highlighting the dynamic interplay between prediction errors, subjective valuation, and self-referential processes. This approach not only advances theoretical understanding but also opens new avenues for computational modeling and empirical research into the adaptive and maladaptive aspects of emotional experience.