Integration the need to apply these deep learning models in generating realistic landscapes similar to the Moon prompted the new approach based on the modified U-Net. In forming this piece of work, this paper takes advantage of the possibility of such a convolutional neural network to generate realistic fake landscapes of the Moon, similar to the details and texture as on the Moon’s surface. It is called as the U-Net architecture which has performed greatly on image segmentation but also on generation of images, just that is utilized here in this case to generate the information about the lunar landscape. The network is then created using a very large and varied set of pictures of the Moon and its landforms. Our approach embeds the encoder path and decoder path in the U-net, which maintains fine intricate details as well as context information of the generation image. Besides, we add to conditional generation for generating a lunar landscape aligned to a particular region and geological attribute. We conduct extensive and thorough quantitative and qualitative comparisons to real-world lunar imagery and terrain data for generated lunar landscapes’ realism. Results show that our improved U-Net model can generate images of the lunar landscape closely similar to the geological characteristics of the surface. We also discuss possible applications toward generating lunar landscapes through the U-net model for conducting better astronaut training in simulated environments, educational outreach, and easy creation and operation of virtual but highly realistic experiences on the moon for purposes of direct sciences and purely for entertainment. In conclusion, our study shows how the U-Net architecture can be used successfully to synthesize realistic artificial landscapes of the Moon, uncover possibly useful information in the geology of the Moon, and expand new opportunities in the exploration of space and education about the Moon in cyber space.