ProImage-Bench: Rubric-Based Evaluation
for Professional Image Generation

Minheng Ni*, Zhengyuan Yang*, Yaowen Zhang*, Linjie Li, Chung-Ching Lin, Kevin Lin, Zhendong Wang, Xiaofei Wang, Shujie Liu, Lei Zhang, Wangmeng Zuo, Lijuan Wang

📄 arXiv 💻 Code 💾 Dataset

We propose ProImage-Bench, a rubric-based evaluation benchmark for professional image generation. For 654 figures collected from real textbooks and technical reports, we construct detailed image instructions and a hierarchy of rubrics that decompose correctness into 6,076 criteria and 44,131 binary checks.

Dataset Domains

The biology domain contains illustrations sourced from biology textbooks, covering a wide range of scientific content, including cellular and organelle structures, physiological and metabolic processes, and ecological or evolutionary diagrams.

* Click on any image to view details.

Case Studies

Examples of model outputs on the benchmark

Case 1 of 30

Detailed Description

Case Studies of Evaluation Results

Select Model Output:

GPT-4o

Detailed Description

Evaluation Rubric

Overall Results on ProImage-Bench

Model	Biology		Engineering		General		Overall
Model	Acc	Score	Acc	Score	Acc	Score	Acc	Score
Nano Banana Pro	0.849	0.625	0.708	0.434	0.816	0.601	0.791	0.553
Wan2.5	0.714	0.433	0.606	0.309	0.755	0.519	0.692	0.420
GPT-4o	0.704	0.425	0.556	0.258	0.718	0.463	0.660	0.382
Nano Banana	0.697	0.400	0.579	0.276	0.716	0.468	0.664	0.381
Seedream	0.680	0.393	0.560	0.260	0.688	0.442	0.642	0.365
Imagen-3	0.600	0.288	0.492	0.195	0.638	0.377	0.577	0.287
FLUX-dev	0.592	0.286	0.444	0.167	0.616	0.359	0.551	0.270

Comparison of different models on ProImage-Bench across three domains.

Citation

If you find ProImage-Bench useful in your research, please consider citing our paper:

BibTeX

@article{ni2025proimage,
  title={ProImage-Bench: Rubric-Based Evaluation for Professional Image Generation},
  author={Ni, Minheng and Yang, Zhengyuan and Zhang, Yaowen and Li, Linjie and Lin, Chung-Ching and Lin, Kevin and Wang, Zhendong and Wang, Xiaofei and Liu, Shujie and Zhang, Lei and others},
  journal={arXiv preprint arXiv:2512.12220},
  year={2025}
}