The Crack Diffusion Model (CrackDiff) is an innovative diffusion-based method for pavement crack detection. It leverages the learning capabilities of generative diffusion models to capture the data distribution and latent spatial relationships of cracks across different sample timesteps, generating more accurate and continuous crack segmentation results. CrackDiff uses crack images as guidance for the diffusion model and employs a multi-task UNet architecture to predict mask and noise simultaneously at each sampling step, enhancing the robustness of generations. Compared to other models, CrackDiff generates more accurate and stable results. Through experiments on the Crack500 and DeepCrack datasets, CrackDiff achieves the best performance (F1 = 0.818 and mIoU = 0.841 on Crack500, and F1 = 0.841 and mIoU = 0.862 on DeepCrack). The main contributions of this paper include introducing a novel framework for pavement crack detection based on the diffusion model, proposing a diffusion model structure based on multi-task UNet, and demonstrating the effectiveness of CrackDiff through experiments on two public datasets. The model excels in capturing both shallow and deep image features, achieving superior performance in crack segmentation tasks. However, the model faces challenges such as slow inference speed and difficulty in detecting cracks within shadows. Future research will focus on improving the inference speed of diffusion models and exploring broader applications. The code will be made publicly accessible.The Crack Diffusion Model (CrackDiff) is an innovative diffusion-based method for pavement crack detection. It leverages the learning capabilities of generative diffusion models to capture the data distribution and latent spatial relationships of cracks across different sample timesteps, generating more accurate and continuous crack segmentation results. CrackDiff uses crack images as guidance for the diffusion model and employs a multi-task UNet architecture to predict mask and noise simultaneously at each sampling step, enhancing the robustness of generations. Compared to other models, CrackDiff generates more accurate and stable results. Through experiments on the Crack500 and DeepCrack datasets, CrackDiff achieves the best performance (F1 = 0.818 and mIoU = 0.841 on Crack500, and F1 = 0.841 and mIoU = 0.862 on DeepCrack). The main contributions of this paper include introducing a novel framework for pavement crack detection based on the diffusion model, proposing a diffusion model structure based on multi-task UNet, and demonstrating the effectiveness of CrackDiff through experiments on two public datasets. The model excels in capturing both shallow and deep image features, achieving superior performance in crack segmentation tasks. However, the model faces challenges such as slow inference speed and difficulty in detecting cracks within shadows. Future research will focus on improving the inference speed of diffusion models and exploring broader applications. The code will be made publicly accessible.