<p>The quality of ultrasound diagnosis is significantly dependent on the skill and experience of the examiner. Therefore, the practical training of ultrasound diagnosis is essential for novice physicians. However, with the presence of lesions being pre-determined, existing training modalities, e.g., phantoms or simulated patients, fall short in providing training on the important “searching for lesions in an actual human body” process. This study aims to provide on-the-spot content for more flexible and accessible ultrasound training. Specifically, it involves tracking organs in ultrasound images and dynamically embedding realistic lesions. We developed a system capable of embedding virtual kidney stones and conducted experiments to evaluate whether these stones could be indistinguishable from real ones by physicians. In this study, we validated realism through two experiments. First, in a paired-comparison test with 23 physicians, virtual kidney stone images were evaluated as indicative of stones to a degree comparable with real cases, clearly surpassing no-finding images. Furthermore, in a video classification experiment with eight physicians, 54% of the videos containing virtual kidney stones were judged as real cases. The experimental results identified key points for the effectiveness of virtual kidney stones as educational tools: the natural placement of the stones, the representation of acoustic shadows, and the depiction of the stones’ impact on the body. Although our method focused on the first two points, many doctors found the virtual kidney stones to appear as real. We believe this technique can represent realistic virtual kidney stones and holds promising potential for providing extensive medical training scenarios in the future.</p>