A tokenized framework for personalized poster generation in visual communication arts

In the realm of visual communication arts, the ability to create impactful and visually appealing posters is crucial for effective message delivery. Given the rapid evolution of digital technologies and the increasing demand for personalized content, there is a significant need for advanced methods that streamline the poster design process. Addressing this need, this paper introduces a novel tokenized poster design system aimed at streamlining the automated creation of personalized visual content. In this context, a “tokenized framework” refers to an automated pipeline where extracted text keywords (tokens) act as the foundational semantic units driving the entire design process—from searching and retrieving image assets to guiding their contextual arrangement and blending. Utilizing both text-based and content-based image retrieval methods, the system accurately retrieves foreground images from the Internet and employs advanced image processing algorithms such as GrabCut and Poisson image manipulation for image segmentation and seamless blending to produce customized posters that meet user demands. Additionally, the system incorporates bidirectional regulations (i.e., positive and negative regulations) to provide figure arrangement recommendations based on artistic and compositional regulations of images, enhancing the visual appeal and practicality of the generated images. The experimental results demonstrate that the proposed system effectively utilizes image data extracted from large-scale databases to create posters with high personalization and visual appeal. Additionally, tests of the system show that integrating these technologies significantly reduces False Positive Rates (FPR) and generates images that meet user expectations across various application scenarios.