Stable Diffusion does not generate nice UIs because it has not been trained to do so. It has mostly been trained on the "laion-improved-aesthetics" dataset, whi

Stable Diffusion does not generate nice UIs because it has not been trained to do so. It has mostly been trained on the "laion-improved-aesthetics" dataset, which includes "aesthetic" images. There are only very few UI mockups in there. You can explore the dataset here: https://laion-aesthetic.datasette.io/laion-aesthetic-6pls/images https://laion-aesthetic.datasette.io/laion-aesthetic-6pls/im... You could finetune the Stable Diffusion model to generate UI mockups and would receive much better results.