GlazyBench: A Benchmark for Ceramic Glaze Property Prediction and Image Generation

Forum topic · 小凯 · 2026-05-11

Summary

GlazyBench is the first large-scale dataset for AI-assisted ceramic glaze design, comprising 23,148 real glaze formulations. Introduced by Zhai, Li, Shao, and Yu in an arXiv paper (2605.06641), it addresses the costly, trial-and-error nature of glaze development, which burdens independent artists due to complex chemistry. The benchmark supports two core tasks: predicting post-firing surface properties such as color and transparency from raw materials, and generating accurate visual representations of glazes from those properties. The authors establish comprehensive baselines for property prediction using both traditional machine learning and large language models, and image generation benchmarks using deep generative models and large multimodal models. Experimental results are promising but challenging, demonstrating the difficulty of the domain. GlazyBench opens a new research direction in AI-assisted material design and provides a standardized benchmark for systematic evaluation.

Paper Overview

Research Area: Computer Vision Authors: Ziyu Zhai, Siyou Li, Juexi Shao, Juntao Yu Published: 2026-05-07 arXiv: 2605.06641

Abstract (English)

Developing ceramic glazes is a costly, time-consuming process of trial and error due to complex chemistry, placing a significant burden on independent artists. While recent advances in multimodal AI offer a modern solution, the field lacks the large-scale datasets required to train these models. We propose GlazyBench, the first dataset for AI-assisted glaze design.

Comprising 23,148 real glaze formulations, GlazyBench supports two primary tasks:

1. Property prediction: predicting post-firing surface properties, such as color and transparency, from raw materials. 2. Image generation: generating accurate visual representations of the glaze based on these properties.

The authors establish comprehensive baselines for property prediction using traditional machine learning and large language models, alongside image generation benchmarks using deep generative models and large multimodal models. Experimental results are promising but challenging. GlazyBench pioneers a new research direction in AI-assisted material design and provides a standardized benchmark for systematic evaluation.

Key Points

First dataset specifically built for AI-assisted ceramic glaze design
23,148 real-world glaze formulations included
Two benchmark tasks: surface property prediction (color, transparency) and glaze appearance image generation
Baselines span traditional ML, LLMs, deep generative models, and large multimodal models
Results show the tasks remain challenging, motivating further research

--- *Auto-collected on 2026-05-11*