Library-specific flavors in MLflow are designed to bridge the gap between MLflow's general-purpose model registry and the specific requirements of various machine learning libraries. They enable the loading of models logged in MLflow by specifying the library or framework used for their creation. This ensures compatibility and allows for the seamless execution of models across different tools within the machine learning workflow. The other options do not accurately describe the primary purpose of flavors: updating model stages, logging metrics, and specifying version descriptions are all separate functionalities within MLflow that do not involve flavors.