Background: Realistic visualizations have been associated with benefits in retention performance, but also with disadvantages in transfer tasks. Realistic details are considered to be helpful for the initial learning of content, but not for the generalization to other domains and tasks. Objectives: The contradictory nature of previous evidence needed to be tested in a situation in which both the retention of concrete surfaces as well as the transfer to different shapes is necessary. Methods: Three experiments were conducted to assess whether transfer tasks in which knowledge of surfaces needed to be transferred to variations can benefit from realism, shape distinctness, and descriptive labels. In Experiment 1, participants learned skull anatomy either in a schematic or a realistic style and completed near and far transfer tests. In Experiment 2, the realism and shape distinctness of a fictional bone model was varied. Subjective cognitive load, retention performance, and transfer scores were measured. Experiment 3 compared how distinct and less distinct shapes of another fictional bone model affect learning in combination with label texts of differing semantic descriptiveness. Results and Conclusions: In Experiment 1, performance in an anatomy learning task was raised by learning with a realistic visualization across near and far transfer tests. Experiment 2 revealed that extraneous cognitive load is increased for realistic visualizations when the depicted shapes lack distinctness. Experiment 3 suggests that when realistic visualizations feature less distinct shapes, learners struggle with descriptive labels in retention tests and benefit from similarly unspecific labels.