The importance of affective schemas for emotional behavior is well-established, but their properties remain poorly understood. Here, we examined affective schema acquisition (over a ~24-hour consolidation period), updating (assimilation and accommodation), and inference across three novel experimental paradigms. We show that specific learned affective associations become semanticized into generalized affective schemas (negative, positive, and neutral). Once acquired, these schemas can assimilate similar information, accommodate to new inconsistent information, and generate new affective inferences. Valenced affective schemas are acquired faster than neutral schemas, are more difficult to reverse, and facilitate rapid learning and memory for related information. Negative-valenced schemas are most prioritized for learning, most resistant to change, and more effective at facilitating gist-based inferences. These findings demonstrate the properties of semanticized affective memories, how they organize into schemas, how those memories are modified, and how they generate affective meaning-making, offering new directions for understanding how complex generalized emotional memories operate.