We present an experimental comparison of three kinesiatric exercise instruction modalities: a live human instructor (human), recorded video (video) and a virtual figure displayed next to the representation of the users’ approximate skeleton (interactive). The results regarding user experience, preferences, and exercise accuracy indicate a preference for the human instructor across measures. A disparity exists between exercise accuracy and perceived ease of understanding when comparing the video with the interactive modality. Perception measures indicate a slight preference for the video modality, whilst performance data shows a significantly higher accuracy in the interactive condition. Our findings support the further investigation of digital interfaces to support physical therapy and rehabilitation as a cost-effective and potentially more efficiently customizable addition to traditional exercise instruction forms.