Brain-wide neuromodulation by behavioral variables, such as locomotion, pupil area, and face motion, have been observed in mice (Stringer et al., 2019; Musall et al., 2019). To study this mechanism at higher spatial resolution, we used two-photon imaging that allows recording of individual neuronal and synaptic bouton activity in mouse primary visual cortex (V1), while the animal's face was simultaneously videotaped. We aim to understand how face motion is related to the population activity of both cortical neurons and their lateral geniculate nucleus (LGN) afferents, during both visual stimuli evoked periods and their non-stimulus-evoked interval periods. To avoid ‘nonsense correlations’ during linear regression with variables that contain trends over time, we applied session permutation controls to ensure a nonsense-correlation-free, genuine relationship between neuronal and behavioral signals (Harris, 2020). We observed a robust correlation between face motion and neuronal population activity, which is higher for visually evoked response compared to non-evoked activity. In contrast, face motion does not correlate to LGN bouton activity during non-evoked periods, as boutons are almost silent. However, LGN activity surprisingly becomes significantly correlated with face motion during visually stimulation. This last observation can be only partially, but not completely, explained by eye movements, indicating a non-trivial mechanism of bottom-up source of behavioral modulation. Altogether, these results prompt the following hypothesis: improved encoding of face motion variables in V1 cortical neurons during visually evoked regimes is mainly due to face motion correlated bottom-up LGN inputs, rather than stronger top-down movement-related cortical inputs. In total, our work gives an unprecedented analysis of the circuit pathways that underlie the recent observations that mouse V1 activity is related, in part, to non-visual inputs.