This research explores the use of AI algorithms to generate lower body gait movements for underwater hydrotherapy exercises. Traditional hydrotherapy offers numerous health benefits, but monitoring patient movements underwater presents unique challenges. The study focuses on utilizing marker-less motion capture through advanced pose estimation models, such as YOLOv7, to accurately track and analyze lower body movements during hydrotherapy. The paper discusses the development of a tailored underwater dataset, annotated to focus on critical lower body joints, and evaluates the performance of different AI models under challenging underwater conditions. The results demonstrate the effectiveness of increasing dataset diversity in improving model accuracy and reducing false positives, highlighting the potential for AI-driven improvements in hydrotherapy practices.