Data augmentation remains a widely utilized technique in deep learning, particularly in tasks such as image classification, semantic segmentation, and object detection. Among them, Copy-Paste is a simple yet effective method and gain great attention recently. However, existing Copy-Paste often overlook contextual relevance between source and target images, resulting in inconsistencies in generated outputs. To address this challenge, we propose a context-aware approach that integrates Bidirectional Latent Information Propagation (BLIP) for content extraction from source images. By matching extracted content information with category information, our method ensures cohesive integration of target objects using Segment Anything Model (SAM) and You Only Look Once (YOLO). This approach eliminates the need for manual annotation, offering an automated and user-friendly solution. Experimental evaluations across diverse datasets demonstrate the effectiveness of our method in enhancing data diversity and generating high-quality pseudo-images across various computer vision tasks.