<p>Currently, the defect detection capability of most deep learning networks heavily relies on high-quality training datasets, the collection and creation of which require significant time and resources. In particular, negative samples are scarce in industrial production, making it difficult to obtain enough for sufficient training. As new textiles are frequently introduced, a large number of new defect images are generated, and detection models often lack strong generalization ability, requiring specialized training for different scenarios. To address the limitations of deep learning in fabric defect detection and the challenges of expanding datasets for various textured fabrics, this paper proposes a multi-class controllable fabric defect synthesis and transfer method, named FLSC-DCCGAN, to fill this gap. This method, through an integrated design, is applied to the generation of fabric defect samples. First, a Generative Adversarial Network (GAN) is used to generate defect samples without backgrounds, expanding the existing defect dataset. The generated samples are then integrated with defect-free structures using image preprocessing and an improved Laplacian pyramid fusion technique. This approach enables the rapid generation of high-quality defect datasets. Experimental results demonstrate that the proposed method significantly improves the accuracy and robustness of fabric defect detection models.</p>