The paper introduces CrackDiff, an innovative framework for pavement crack detection based on the generative diffusion model. Pavement cracks, which exhibit spatial continuity, pose significant challenges for traditional algorithms. CrackDiff leverages the learning capabilities of the generative diffusion model to capture the data distribution and latent spatial relationships of cracks, generating more accurate and continuous crack segmentation results. The model uses crack images as guidance and employs a multi-task UNet architecture to predict both the mask and noise simultaneously at each sampling step, enhancing robustness. Compared to other models, CrackDiff achieves superior performance on the Crack500 and DeepCrack datasets, with F1 scores of 0.818 and 0.841, and mIoU scores of 0.841 and 0.862, respectively. The paper discusses the contributions, methodology, experimental results, and limitations of CrackDiff, highlighting its effectiveness in capturing both shallow and deep image features. Future research directions include improving inference speed and exploring broader applications.The paper introduces CrackDiff, an innovative framework for pavement crack detection based on the generative diffusion model. Pavement cracks, which exhibit spatial continuity, pose significant challenges for traditional algorithms. CrackDiff leverages the learning capabilities of the generative diffusion model to capture the data distribution and latent spatial relationships of cracks, generating more accurate and continuous crack segmentation results. The model uses crack images as guidance and employs a multi-task UNet architecture to predict both the mask and noise simultaneously at each sampling step, enhancing robustness. Compared to other models, CrackDiff achieves superior performance on the Crack500 and DeepCrack datasets, with F1 scores of 0.818 and 0.841, and mIoU scores of 0.841 and 0.862, respectively. The paper discusses the contributions, methodology, experimental results, and limitations of CrackDiff, highlighting its effectiveness in capturing both shallow and deep image features. Future research directions include improving inference speed and exploring broader applications.