The increasing fidelity of visual simulators has created a requirement for realistic images of natural phenomena such as clouds. In the GBU15 part task trainer, cloud images were used to simulate the obscuration of the sensor's field of view during simulated training missions. Obtaining cloud images that met the fidelity requirements of the trainer was difficult. The cloud images had to look realistic, yet have a predefined distribution over a large area, and the images could not contain a background. Three methods for obtaining the cloud images were evaluated. The methods were photography, painting, and Computer Image Generation (CIG). The CIG method was chosen for the GBU15 part task trainer because it allowed control of cloud distribution, shape, and density, and it generated images at the required resolution and without a background.