A narrative is affectively polarized when the distribution of expressed feelings towards such a narrative shows two separated peaks. In this paper, we explore the distributions of feelings expressed by online users on Twitter in March 2020 using a sample of 7.9 million tweets related to the narrative of the COVID-19 pandemic. How people influenced each other affective states? Does affective polarization emerge? To answer these questions, sub-narratives are distilled from Twitter data by thresholding a structural parameter of a graph constructed with hashtags co-occurrence. Here, hashtags are considered as string identifiers of sub-narratives related to COVID-19. For such sub-narrative, we found a polarized distribution of affective states, in which a radicalization of measured values emerged non-linearly for positive and negative valence of affective states, depending on the hashtag considered. We stress that our findings are not generalizable, but they suggest that such type of polarization was present online during social distancing, and that interactions may have non-linearly impacted the initial affective states of online groups’ constituent parts.