Ever gotten lost in a complex chart, unsure of what story it was trying to tell? Or perhaps wondered if AI could interpret the nuances of a visualization? Recent advances in large language models (LLMs) might hold the answer. This emerging field, known as visualization captioning, empowers AI to describe the insights locked within charts and graphs. Much like a skilled analyst, these models can translate visual data into human-readable narratives, making information more accessible. Traditionally, visualizations offered powerful advantages over text, leveraging our natural visual processing abilities. Yet, their complexity sometimes hindered understanding, making textual support crucial. This is where LLMs come in. The secret sauce lies in their underlying architecture—neural networks and transformers—that enable them to process and generate language with remarkable fluency. These models learn from vast datasets of charts, captions, and associated data tables, deciphering the complex relationships between visual elements and their textual descriptions. However, these models are not without limitations. LLMs face challenges when dealing with subtle visual cues, complex chart types, and the nuances of human interpretation. Current research is actively addressing these challenges. Areas of active development include improving model accuracy in describing charts that contain both spatial and temporal information, and ensuring factual correctness of the generated text. From crafting more contextually relevant captions to handling complex visualizations involving both spatial and temporal features, the future of visualization captioning promises a more seamless blend of visual and textual information, unlocking new possibilities for data analysis and accessibility.