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, introduced in an arXiv paper (2605.06641) by Zhai, Li, Shao, and Yu. Developing ceramic glazes traditionally requires costly, time-consuming trial and error due to complex chemistry, a burden especially heavy for independent artists. The dataset contains 23,148 real glaze formulations and supports two core tasks: predicting post-firing surface properties such as color and transparency from raw material compositions, and generating accurate visual representations of glazes based on those properties. The authors establish comprehensive baselines using traditional machine learning and large language models for property prediction, plus image generation benchmarks using deep generative models and large multimodal models. Results are promising but challenging, showing the task is far from solved. GlazyBench opens a new research direction in AI-assisted material design and provides a standardized benchmark for systematic evaluation.

Overview

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

Abstract (translated)

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. The authors 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. Visual generation: generating accurate visual representations of the glaze based on these properties.

The paper establishes 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. Experiments show promising but challenging results. GlazyBench opens a new research direction in AI-assisted material design and provides a standardized benchmark for systematic evaluation.

Original Abstract (excerpt)

> 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: predicting post-firing surface properties, such as color and transparency, from raw materials, and generating accurate visual representations of the glaze based on these properties. We 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.

---

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