Brain-scale neural circuits transform visual input into discrete movements with appropriate speed, vigor, and duration. Yet, single-cell resolution and mechanistic characterization of such circuits in vertebrates have been elusive, largely due to the vast number of neurons and complexity of the mammalian brain. The translucent larval zebrafish offers unique opportunities to study brain-wide cellular-level neural computations during visuomotor transformations. One such transformation is the optomotor response (OMR), where fish execute discrete locomotion bouts to stabilize their body position in response to optic flow. Previous studies [1] have functionally characterized OMR neural circuits across the brain, including the retinorecipient pretectum (Pt) and motor command neurons in the midbrain nucleus of the longitudinal fasciculus (nMLF) [2]. Recently, we implemented these neural circuits in a realistic, physics-based neuromechanical simulation, simZFish, replicating the OMR behavior of real zebrafish [3]. However, simZFish lacked biorealistic, functional connections between its artificial Pt and nMLF neurons, resulting in poor speed adaptations in the model. To address this, we combined volumetric two-photon microscopy with 3D holographic photostimulation [4] during visual stimulation and behavioral tail tracking. Using these all-optical methods, we mapped the functional connectivity between Pt and nMLF populations, revealing how visually characterized Pt neurons influence these nMLF neurons. Our results produced correlation-based Pt-nMLF functional connectomes, generating population-level models of how visual motion is decoded onto motor commands. To validate this connectivity data, we used these empirically measured connectivity weights to refine and constrain the neural architecture of simZFish. By comparing virtual and cell-specific nMLF-photostimulated behavior, we improved simZFish behavior when presented with novel speed-varying visual stimuli. Together, we demonstrate how all-optical methods at cellular resolution can empirically map functional connectivity to constrain neuromechanical simulations, ultimately offering critical insights into sensorimotor transformations.