Training requirements for a variety of platforms are quickly expanding to include larger and larger gaming areas in response to customer demand and the availability of data. However, there still remain several drawbacks to using worldwide high-resolution photo-specific data: size of the data, the ability to correlate data with sensor and SAF versions, the time required to validate and correct data. Instead using auto-generated simulation models coupled with real-world data to quickly and economically create training environments remains an attractive option. This paper describes two techniques recently developed to build realistic terrain texture that is pseudo-specific data (from low resolution data, i.e., Feature Identification Codes, or FICs). When using low resolution theme data resulting textures can appear "blocky" and unnatural. One way to improve this is to super-sample the boundaries between themes to a higher resolution in such a way that they appear more natural and less blocky when viewed up close. Stencils are defined for blending two or more theme types to create natural looking edges. Multiple stencils applied in specific ways are used to vary edges thereby avoiding repeating image patterns. Next, the super-sampled theme data is used with correlated templates of three-dimensional features to generate 3D content on-the-fly without the need of "pre-compiling" or "publishing" the database. The end result is the appearance of higher resolution terrain texture with accurately correlated 3D features.