A novel method named PAFlow has been introduced to enhance structure-based drug design by generating three-dimensional molecules that bind effectively to target proteins. This approach specifically addresses challenges in previous models related to the stability and quality of generated molecules. By incorporating learnable atom numbers, PAFlow aims to improve the precision and reliability of molecule generation. The method holds promise for advancing drug discovery by producing target-aware molecules with better binding properties. Its development reflects ongoing efforts to refine computational techniques in molecular design, potentially accelerating the identification of effective drug candidates. The introduction of PAFlow represents a significant step toward more stable and high-quality molecule generation in the field of AI-driven drug design.