Diffusion models have emerged as a catalyst for remarkable advancements in video editing tasks. Despite these achievements, the majority of research in video editing has predominantly focused on style and content manipulation, leaving action editing as a relatively uncharted territory. Video editing of human poses entails multiple challenges. Firstly, the generation of action sequences is vulnerable to artifacts, primarily caused by the influence of reference videos. Secondly, maintaining a consistent background and character appearance throughout the edited footage poses a significant challenge. In addition, most action control methodologies are highly dependent on the availability of large volumes of high-quality training data. To address these issues, we introduce PoseMaster, an innovative one-shot framework that significantly improves the video editing process. Specifically, we utilize a temporal fusion method to mitigate the sticky effects, thus improving the smoothness of the synthesized videos. Complementing this approach is the Pose-Controller Module, which plays a crucial role in ensuring action continuity. Our experiments validate the effectiveness of PoseMaster, demonstrating its ability to generate more realistic videos with a higher level of action consistency compared to existing methods. This achievement demonstrates a substantial advancement in human-centric video editing, paving the way for more refined and sophisticated video editing capabilities.