The temporal stability of ecological communities is shaped not only by environmental fluctuations but also by the structure of species interaction networks. In this study, we examine how interaction network structure (such as interaction strength and interaction symmetry) is reflected in community-level temporal variability using a generalized Lotka–Volterra model driven by temporally correlated environmental noise. We use dynamical mean-field theory to characterize species-level temporal means and variances in high-diversity communities. Building on these species-level statistics, we analyze community variability through a variance decomposition framework that separates contributions from individual species variance and from interspecific covariances, following the variance decomposition approach of Segrestin et al. (2024, Global Ecology and Biogeography, https://doi.org/10.1111/geb.13828). By using this decomposition to express how interaction network structure shapes community-level variability, we organize community-level temporal variability in terms of portfolio (averaging) effects, dominance effects, and interspecific synchrony. Our results provide a mechanistic null model for how different components of temporal community stability arise from interaction structure and environmental forcing, and clarify the mechanistic basis of the components identified by variance decomposition approaches in theoretical studies of ecological communities.