An analytical model of sea optical properties has been developed in order to generate sea surface images, as seen by an infrared sensor. This model is based on a statistical approach and integrates the spatial variability of a wind-roughened sea surface whose variability ranges from a 1-m to a kilometer scale. It also takes into account submetric variability. A two-scale approach has been applied by superimposing small scale variability (smaller than the pixel footprint) to larger ones. Introducing multiresolution in the sensor field of view allows the requirement of any observational configuration, including nadir as well as grazing view geometry. The physical background of the methods has been tested against theoretical considerations. We also obtained a good agreement with dataset collections at our disposal and taken from the literature, such that a bias shows up at grazing angles, mainly explained by not taking into account multiple reflections. Applied to the generation of synthetic sea surface radiance images, our model leads to good quality ocean scenes, whatever the contextual conditions.