<p>Industrial visual inspection is fundamentally limited by the scarcity of defective samples and the high cost of pixel-level annotation, making anomaly synthesis a practical way to expand training data. However, for metallic components, anomalies are embedded in dense machining micro-textures and must adhere to strict geometric structures; under few-shot supervision, existing synthesis methods often suffer from mask–image misalignment, background structural drift, and over-smoothed high-frequency details, which undermines both realism and downstream utility. To address these challenges, we propose Joint Geometry and Texture Decoupled Diffusion (JGTDiff), a dual-stream, mask-conditioned cross-conditional diffusion framework that jointly synthesizes an anomaly image and its corresponding mask. JGTDiff stabilizes geometric conditioning with a structural prior module, enforces cross-scale consistency between the image and mask streams through multi-scale cross-conditional guidance (CCG), decouples geometry and texture manifolds via a dual-bottleneck reconstruction branch, and restores high-frequency details using a gradient-frequency post-compensation (GFPC) module to compensate for information loss during iterative denoising. Extensive experiments on the MVTec AD benchmark demonstrate that JGTDiff generates structurally consistent and texture-rich anomaly image–mask pairs, outperforming representative GAN- and diffusion-based baselines in fidelity, diversity, and alignment. Moreover, the synthetic data produced by JGTDiff leads to significant improvements in downstream anomaly detection, localization, and defect classification.</p>