Can AI Draw Science? A Benchmark for Evaluating Scientific Figure Generation by Text-to-Image and Multimodal Models

Recent advancements address scientific image generation through new frameworks. SciDraw-Bench evaluates the usability of figures, focusing on legible labels and disciplinary conventions. ILLUME-X introduces a unified multimodal paradigm for high-quality, free-form interleaved text-image sequences by optimizing data efficiency. Additionally, SciIR provides a large-scale dataset, SciIR-82k, to improve semantic alignment and logical reasoning in scientific imagery by formalizing entity structure, scientific processes, and laws, overcoming the scarcity of specialized training data for T2I models.