Augmented Reality and Virtual Reality (AR/VR) are critical technologies for future training initiatives. However, creating sufficiently rich context and content for these platforms requires a significant cost in terms of time, labor, and technical expertise. Designing the environmental setting, implementing relevant actions to simulate real-world events and interactions, designing, and creating instructional content, and integrating it all within a digital form requires substantial effort. Creating multiple, varied, and useful training scenarios rapidly and repeatedly is therefore a significant barrier preventing widespread adoption and utility in DoD and commercial training programs. Recent advances in computer vision and artificial intelligence (AI) can be used to generate quality training scenarios automatically from 2D video. In this paper we describe the Generative Immersive Scenario Testbed (GIST), an AI driven approach to generate dynamic 3D virtual content from 2D videos. We outline GIST’s integrated system of technologies for generating scenario which recreate events from videos, including entity positions and high-level human behaviors. We describe the challenges and outcomes from our approach against 3 performance metrics: time to complete scenario generation, number of possible scenario permutations, and the efficacy of the generated content (ability to detect, track, and translate critical cues) for training.  Our results suggest that AI driven solutions hold great promise for the rapid creation of 3D training content.