In order to meet user requirements, tradeoffs are made in the implementation of the four functions (scene management, prioritization, geometric processing, and video processing) that comprise a digital image generation system of the sort used for flight training. This paper discusses how different approaches to image generator architecture affect the features apparent to the user. Among the architectural variations discussed are programmable versus pipelined geometric processing, and four variations of video processor (scanline, reverse-priority-ordered frame buffering, priority-ordered frame buffering, and distance buffering). The architectures are compared with respect to system cost, overload sensitivity, and the implementation of anti-aliasing, texture, and translucency features, among others. Understanding the tradeoffs involved will help designers and users better meet the requirements of a training task.