Accurate detection of malaria-infected red blood cells (iRBCs) in medical blood smear images is critical for clinical diagnosis. However, severe class imbalance, especially underrepresentation of late-stage gametocytes limits detection accuracy and increases the risk of missed infections, potentially compromising patient treatment outcomes. To address this challenge, in this work, we present a domain-aware, generative data augmentation pipeline that synthesizes realistic iRBCs within blood smear microscopy field-of-view (FoV) images. Our method integrates synthetic iRBC masks to real smear annotations, generating high-fidelity training data for segmentation models. We evaluate this approach across multiple segmentation models, including an optimized Cellpose variant, U-Net, and Mask R-CNN, and conduct a Zero-Shot evaluation using the recently released Cellpose-SAM. Our results demonstrate consistent improvements in iRBC detection, with the optimized Cellpose achieving a detection rate of 97.42%, outperforming conventional augmentation by over 4%, highlighting the effectiveness of targeted generative augmentation even on compact model sizes. This approach has the potential to enhance automated malaria diagnosis in resource-limited settings where accurate detection of all parasitic stages is critical for effective treatment and disease control.