Abstract: Generative Artificial Intelligence (AI) has emerged as a key driver for the intelligent design of shear wall structures. However, the performance of generative AI models is heavily contingent upon the quality and scale of training data. Existing training datasets often prove insufficient when confronted with challenges such as updated design codes or unconventional design tasks, failing to meet the training demands for models geared towards these new requirements. To address this gap, a hybrid data augmentation method, driven by both automated and expert optimization, is proposed. The method aims to enhance the training data for intelligent design, specifically to tackle the challenges posed by new building design standards (e.g., a minimum floor height of 3 meters) and complex design tasks (e.g., functional requirements of the building may result in an irregular structural plan), thereby improving the learning efficacy and design quality of the generative AI model. The proposed methodology begins by evaluating the mechanical performance of an existing dataset of 650 shear wall structure designs against the updated standards. Samples that fail to comply with the new code are first subjected to an automated optimization process using parametric modeling. Designs that still do not meet the requirements after this stage are then manually adjusted by expert engineers. The final augmented dataset is subsequently used to retrain the generative AI model. An analysis based on a test set of 20 standard cases demonstrates that the hybrid automated-expert optimization method provides an efficient and cost-effective means of enhancing training data quality, with the design compliance rate increasing from 75% (pre-augmentation) to 85%. Consequently, the proposed method significantly reduces manual labor and time costs while effectively safeguarding and enhancing the design quality and reliability of generative AI when confronted with new design requirements.