SlidesGen-Bench: Evaluating Slides Generation via Computational and Quantitative Metrics

The rapid evolution of Large Language Models (LLMs) has fostered diverse paradigms for automated slide generation, ranging from code-driven layouts to image-centric synthesis. However, evaluating these heterogeneous systems remains challenging, as existing protocols often struggle to provide comparable scores across architectures or rely on uncalibrated judgments. In this paper, we introduce SlidesGen-Bench, a benchmark designed to evaluate slide generation through a lens of three core principles: universality, quantification, and reliability. First, to establish a consistent evaluation framework, we ground our primary analysis in the visual domain, treating terminal outputs as visual renderings for consistent comparison, while acknowledging format-conditional ceilings in structural editability. Second, we propose a structured computational approach that quantitatively assesses slides across three distinct dimensions---Content, Aesthetics, and Editability. Finally, we ensure high correlation with human preference for the highly subjective aesthetic dimensions by constructing the Slides-Align1.5k dataset. Our experiments demonstrate that SlidesGen-Bench achieves a significantly higher degree of alignment with human judgment than existing evaluation pipelines. Our code and data are released to support further research.