Visibility forecasts produced using statistical approaches are not practical in remote locations having little or no observational record. This research examines the viability of making skillful, short-term (<20 h), stochastic visibility predictions in fog using uncalibrated output from a 4-km resolution ensemble of 10 WRF members. Each member of the ensemble obtains its initial conditions and lateral boundary conditions from a different member of Air Force Weather Agency's Joint Global Ensemble. Model error is sampled using unique combinations of planetary boundary layer and microphysics parameterizations, as well as by varying the land surface table, in each member. Two visibility parameterizations are tested to convert model output variables into a visibility forecast. The first, based purely on first principles of electromagnetic wave scattering, employs the scattering efficiency of spherical water droplets and the model-predicted total cross-sectional area of the droplets in a given path volume. The second employs an empirical relationship between liquid water content (LWC) and visibility. In addition to using LWC forecast values directly from the ensemble, an alternative approach is tested that utilizes other output variables and a conceptual model of fog dynamics to explicitly solve for LWC for each member in an offline environment. Finally, visibility forecasts are tested with an offline refinement to the droplet number concentration and size distribution, which impacts the total droplet cross-sectional area. A viable framework for producing physics-based, visibility-in-fog predictions is attractive for supporting aviation operations where little or no observational record exists.