Functional Near-Infrared Spectroscopy (fNIRS) is growing increasingly popular as a neuroimaging tool for naturalistic settings due to its portability and cost-effectiveness. However, fNIRS is limited to measuring neural activity near the surface of the brain. Can we study neural activity from deeper brain regions in naturalistic settings with fNIRS? In this study, we built a signal prediction model with adapted principal component regression (aPCR) to predict moment-by-moment whole-brain neural dynamics from fNIRS signals during movie-watching. Using a stringent cross-participant and cross-stimulus approach, we were able to predict the neural dynamics across large areas of the brain using only prefrontal cortex fNIRS signals. Predicted neural dynamics recapitulated ground-truth functional connectivity patterns and captured the temporal accumulation of semantic information across temporal, parietal, and prefrontal cortex.