Accurate photovoltaic (PV) production forecasting is essential for optimizing energy management in modern power systems. While existing methods often rely on time series analysis or meteorological data, this study explores the possibility of treating PV forecasting as a regression problem using sky images, without benefiting from information within temporal dependencies. We evaluate whether contrast and brightness features extracted from sky images captured simultaneously, but with different exposure settings, can provide sufficient predictive power for PV production. Three key scenarios are investigated: (1) regression on individual image types, (2) the impact of including the hour-of-day as a contextual feature, and (3) the fusion of contrast and brightness features across multiple image types. Using 5-fold cross-validation and a diverse set of 9 regression models (linear, tree-based, and ensemble methods), we demonstrate that multi-modal data integration, i.e., combining features from multiple images with the hour-of-day, significantly improves predictive accuracy.