Natural disaster response generates large amounts of unstructured visual and text data. This data is often underutilized due to the lack of structured metadata and automated indexing mechanisms. This presents a significant issue during an active disaster response, when time is severely limited and personnel must rapidly find relevant information to make time sensitive decisions. The rapid rise in the application of multi-modal artificial intelligence has led to a flourishing of many open-source models aimed at addressing this challenge. This paper evaluates the performance of four open-source generative models for generating text descriptions of disaster imagery collected over several years of U.S. disaster response. Through a comparative analysis across this dataset, this study highlights the strengths and limitations of each model and discusses their potential to enhance situational awareness, improve knowledge management, and support real-time decision-making in disaster environments.