PaperDraw: Intelligent generation system for scientific publications pipeline diagrams

Pipeline diagrams in scientific publications effectively reduce readers’ cognitive load and enhance article readability. However, manually constructing these diagrams is complex and time-consuming, highlighting the need for efficient automated generation methods. While existing automation methods can encode logical relationships, their rigid layouts and limited visual styles lack the flexibility and aesthetic polish required for publication. In this work, we present PaperDraw, a system that generates publication-quality pipeline diagrams from natural language descriptions. To this end, we introduce a controllable layout optimization algorithm, which translates insights from design space analysis into actionable rules to enforce academic visual standards. Furthermore, we construct a style vector library from the dataset to recommend a diverse set of visual style candidates based on structural similarity. Flexible user adjustments and customization effectively eliminate the need for tedious manual post-processing, allowing researchers to rapidly iterate and finalize a professionally styled diagram directly from its logical structure. We demonstrate PaperDraw’s efficiency and effectiveness through case studies, user evaluations, and both qualitative and quantitative comparisons with existing methods. The results indicate that PaperDraw enables the efficient generation of pipeline diagrams that are both precise in semantics and compelling in visual design.