In recent years, the increasing demand for long-term pedestrian retrieval has brought the cloth-changing person re-identification (CC-ReID) challenge into the spotlight. In scenarios spanning long periods, there are two main challenges: (1) clothing and background interference; (2) extraction of identity-sensitive information. To address these issues, we introduce a robust framework titled Mask-guided clothes-irrelevant and background-irrelevant Network (Magic-Net). Magic-Net employs knowledge distillation across two distinct streams: the outline stream and the exposed stream. The outline stream captures the pedestrians’ contour, minimizing the impact of clothing and background, while the exposed stream enriches identity-sensitive information from the pedestrian’s exposed areas. This dual-stream integration focuses the model on critical re-identification regions. Evaluations on several benchmark datasets demonstrate Magic-Net’s exceptional performance in tackling the CC-ReID challenge.