Automating banner creation in e-commerce saves time, reduces costs, and enhances marketing efficiency. It enables the quick production of attractive, consistent banners tailored to various communication strategies, allowing businesses to swiftly adapt to market trends and customer needs without heavy manual input. In this study, we present a solution to this problem in the form of a framework that leverages the Latent Diffusion model to generate advertising banners from product descriptions autonomously. Our proposed Diffusion-Craft framework comprises two core modules: The Information Structure Reconfiguration module, which preprocesses and restructures data formats to optimize input for the Latent Diffusion model during both training and inference phases, and the Banner Generation module, which aims to produce banners that align with input descriptions while maintaining aesthetic appeal similar to human designs. Our framework achieves an impressive design time of approximately 10 s per banner. Our framework allows businesses to easily generate visually appealing banners that accurately represent their products, saving time and resources while ensuring effective customer communication.