This paper introduces a pipeline called "Design" for automatically generating design templates, including background images and layout elements. The pipeline is designed to create high-quality templates that are suitable for use in banners, slides, and other design applications. The key challenge is to ensure that the background image has enough non-salient space for the layout elements to be overlaid. To achieve this, the paper proposes two techniques: salient attention constraint and attention reduction. These techniques help control the spatial distribution of salient objects in the background image, ensuring that there is sufficient space for layout elements. The pipeline also includes a layout generator that synthesizes layout elements based on the background image. To achieve a more harmonious composition, an iterative inference strategy is proposed, which allows the background and layout to be refined multiple times. The paper also presents a dataset of over 40,000 advertisement banners to evaluate the performance of the pipeline. The results show that the proposed pipeline generates high-quality templates that are comparable to those created by human designers. Additionally, the pipeline is extended to generate presentation slides that are theme-consistent. The paper also discusses related work in text-to-image generation and graphic layout generation. It highlights the importance of spatial control in generating desirable images and the need for content-aware layout generation. The paper proposes an iterative refinement strategy to improve the composition of the background and layout elements. The results show that the proposed pipeline outperforms existing methods in terms of background image quality, layout alignment, and occlusion rate. The paper concludes that the proposed pipeline has significant potential for design creation and can be used to generate a wide range of design templates.This paper introduces a pipeline called "Design" for automatically generating design templates, including background images and layout elements. The pipeline is designed to create high-quality templates that are suitable for use in banners, slides, and other design applications. The key challenge is to ensure that the background image has enough non-salient space for the layout elements to be overlaid. To achieve this, the paper proposes two techniques: salient attention constraint and attention reduction. These techniques help control the spatial distribution of salient objects in the background image, ensuring that there is sufficient space for layout elements. The pipeline also includes a layout generator that synthesizes layout elements based on the background image. To achieve a more harmonious composition, an iterative inference strategy is proposed, which allows the background and layout to be refined multiple times. The paper also presents a dataset of over 40,000 advertisement banners to evaluate the performance of the pipeline. The results show that the proposed pipeline generates high-quality templates that are comparable to those created by human designers. Additionally, the pipeline is extended to generate presentation slides that are theme-consistent. The paper also discusses related work in text-to-image generation and graphic layout generation. It highlights the importance of spatial control in generating desirable images and the need for content-aware layout generation. The paper proposes an iterative refinement strategy to improve the composition of the background and layout elements. The results show that the proposed pipeline outperforms existing methods in terms of background image quality, layout alignment, and occlusion rate. The paper concludes that the proposed pipeline has significant potential for design creation and can be used to generate a wide range of design templates.