Industrial image anomaly detection (IAD) is an important computer vision task in manufacturing, where identifying defects is crucial for quality assurance. Recent developments in vision-language models (VLMs), such as CLIP, have shown strong potential for zero-shot anomaly detection. However, a significant domain gap persists between pre-trained models and industrial image data. In this paper, we present SynCLIP-AD, a pipeline that extends WinCLIP by incorporating image-level and region-level adapters, along with cross-modal attention modules, to better align visual features with textual descriptions. To address the scarcity of abnormal samples, we introduce a synthetic anomaly generation pipeline that combines Natural Synthetic Anomalies (NSA) with generative AI, enabling controlled and diverse anomaly synthesis. Trained on the MVTec AD dataset, our approach achieves superior performance in both anomaly classification and segmentation tasks, with 93.4% AUROC at the image level and segmentation scores of 88.7%, 76.6%, and 43.8%, outperforming the zero-shot baseline. Extensive experiments and ablation studies validate the effectiveness of our feature adaptation strategy and synthetic data generation.