In order to improve the current inpainting algorithm for natural art images, an artistic image inpainting model is proposed by integrating artificial intelligence and embedding multi-scale attention expansion convolution. According to the uniqueness of artistic images, the network structure of DMFB is improved, and its repair model combines two plug-and-play optimization modules: extended convolution block and coordinate attention mechanism. The extended convolution module is used to capture multi-scale context information, and the coordinate attention mechanism is used to improve the remote migration ability of features of the repaired network. The combination of the two makes the repair results meet the visual visibility and semantic rationality. The experimental results show that ablation experiments are carried out to evaluate the effectiveness of each module in this model. In this paper, the hybrid extended convolution and coordinated attention mechanism are used to train the network, and the effectiveness of MADC is verified. When the expansion rate changes to 2, 4, 6 and 8, the method in this paper will lead to moderately distorted structure and fuzzy texture. On the contrary, there is no coordination attention. In the branch of mechanism, the output of this method shows texture defects and discontinuities. By using these two branches, the method in this paper has achieved good results in structure and texture. The quantitative evaluation of branches without coordinated attention mechanism is given when the expansion rate is fixed at 2, 4, 6 and 8. Conclusion: The results restored by this method are more consistent with the real image in human senses, and the objective evaluation of PSNR, SSIM and MSE has also been improved to some extent.