Designen is an automatic pipeline for generating design templates, including background images and harmonious layout elements. The pipeline addresses the challenge of creating backgrounds with sufficient non-salient space for overlaying layout elements, which is crucial for effective design. To achieve this, Designen introduces two techniques: salient attention constraint and attention reduction. These techniques enhance spatial control in text-to-image (T2I) models, ensuring that the generated backgrounds have the necessary space for layout elements. The background is synthesized using a diffusion model, and the layout is generated using a Transformer-based autoregressive generator. An iterative inference strategy is proposed to refine the background and layout in multiple rounds, improving the overall harmony and visual accessibility of the design. Extensive experiments demonstrate that Designen generates high-quality templates comparable to human-designed ones, and it can also generate theme-consistent slides for presentation generation. The pipeline is evaluated using a dataset of over 40k advertisement banners, and the results show significant improvements over existing methods in terms of background quality, layout alignment, and occlusion.Designen is an automatic pipeline for generating design templates, including background images and harmonious layout elements. The pipeline addresses the challenge of creating backgrounds with sufficient non-salient space for overlaying layout elements, which is crucial for effective design. To achieve this, Designen introduces two techniques: salient attention constraint and attention reduction. These techniques enhance spatial control in text-to-image (T2I) models, ensuring that the generated backgrounds have the necessary space for layout elements. The background is synthesized using a diffusion model, and the layout is generated using a Transformer-based autoregressive generator. An iterative inference strategy is proposed to refine the background and layout in multiple rounds, improving the overall harmony and visual accessibility of the design. Extensive experiments demonstrate that Designen generates high-quality templates comparable to human-designed ones, and it can also generate theme-consistent slides for presentation generation. The pipeline is evaluated using a dataset of over 40k advertisement banners, and the results show significant improvements over existing methods in terms of background quality, layout alignment, and occlusion.