GlazyBench: First Dataset for AI-Assisted Ceramic Glaze Design

GlazyBench, the first large-scale dataset for AI-assisted ceramic glaze design, has been introduced. It comprises 23,148 real glaze formulations and supports two tasks: predicting post-firing surface properties (color, transparency) from raw materials, and generating accurate visual representations of glazes. Baselines were established using traditional machine learning, large language models, deep generative models, and large multimodal models. The dataset aims to reduce the costly trial-and-error process of developing ceramic glazes, which burdens independent artists. Experiments show promising yet challenging results, opening a new research direction.