A variety of mission-dependent tasks are practiced in military flight simulators. A flexible way of meeting the diverse scene-content requirements employs a composite data base from which the appropriate feature models are displayed (and the less appropriate models may be excluded). Importance codes in the feature data are the basis for feature discrimination by the real time scene management software. This software monitors image generator loads, computes demand-variables based upon their deviations with respect to specified limits, and computes several control variables as functions of the maximum demand-variable. These control variables regulate the flow of feature data to or within the image generator. "Closed-loop" load optimization is thereby effected. Imagery and performance data from four example test flights (over the same path) contrast unmanaged, unselectively managed, and selectively managed image generation for two tasks. A given data base can be reoptimized by replacing importance code assignments. The prospects appear favorable for user-tuning of real time data bases to tailor them for particular training tasks.