Detecting and localizing abnormal regions is fundamental in industrial vision. Recently, the frequent emergence of new products and unforeseen defects makes few-shot anomaly detection challenging and important. Existing CLIP-based few-shot methods offer strong cross-modal representations but lack deep interaction between prompt tokens and image features, hindering the precise modeling of anomaly semantics. To address this problem, we propose a few-shot industrial anomaly detection framework based on text-guided image feature synthesis. First, we introduce a pseudo-anomaly feature synthesis method guided by textual feature offsets. In the joint embedding space of CLIP, by computing the semantic offsets between normal and abnormal text embeddings, we guide the normal image features to generate diverse pseudo-anomaly features. Then we design a multi-objective contrastive learning strategy to achieve better semantic alignment between images and text. At the same time, the pseudo-anomaly image and text features can be jointly optimized. Thus, we construct a few-shot industrial defect detection framework that does not require real anomaly samples. Experiments on the MVTec and VisA datasets validate the effectiveness of our method. The results demonstrate that our method achieves strong performance in both image-level and pixel-level few-shot anomaly detection tasks.