GenArena

A unified evaluation framework for visual generation tasks using VLM-based pairwise comparison and Elo ranking.

Abstract

The rapid advancement of visual generation models has outpaced traditional evaluation approaches, necessitating the adoption of Vision-Language Models as surrogate judges. In this work, we systematically investigate the reliability of the prevailing absolute pointwise scoring standard, across a wide spectrum of visual generation tasks. Our analysis reveals that this paradigm is limited due to stochastic inconsistency and poor alignment with human perception. To resolve these limitations, we introduce GenArena, a unified evaluation framework that leverages a pairwise comparison paradigm to ensure stable and human-aligned evaluation. Crucially, our experiments uncover a transformative finding that simply adopting this pairwise protocol enables off-the-shelf open-source models to outperform top-tier proprietary models. Notably, our method boosts evaluation accuracy by over 20% and achieves a Spearman correlation of 0.86 with the authoritative LMArena leaderboard, drastically surpassing the 0.36 correlation of pointwise methods. Based on GenArena, we benchmark state-of-the-art visual generation models across diverse tasks, providing the community with a rigorous and automated evaluation standard for visual generation.

Quick Start

Installation

pip install genarena

Or install from source:

git clone https://github.com/ruihanglix/genarena.git
cd genarena
pip install -e .

Initialize Arena

Download benchmark data and official arena data with one command:

genarena init --arena_dir ./arena --data_dir ./data

This downloads:

• Benchmark Parquet data from rhli/genarena (HuggingFace)
• Official arena data (model outputs + battle logs) from rhli/genarena-battlefield

Environment Setup

Set your VLM API credentials:

export OPENAI_API_KEY="your-api-key"
export OPENAI_BASE_URL="https://api.example.com/v1"

For multi-endpoint support (load balancing and failover), use comma-separated values:

export OPENAI_BASE_URLS="https://api1.example.com/v1,https://api2.example.com/v1"
export OPENAI_API_KEYS="key1,key2,key3"

Run Evaluation

genarena run --arena_dir ./arena --data_dir ./data

View Leaderboard

genarena leaderboard --arena_dir ./arena --subset basic

Check Status

genarena status --arena_dir ./arena --data_dir ./data

Running Your Own Experiments

Directory Structure

To add your own model for evaluation, organize outputs in the following structure:

arena_dir/
└── <subset>/
    └── models/
        └── <GithubID>_<modelName>_<yyyymmdd>/
            └── <model_name>/
                ├── 000000.png
                ├── 000001.png
                └── ...

For example:

arena/basic/models/johndoe_MyNewModel_20260205/MyNewModel/

Generate Images with Diffgentor

Use Diffgentor to batch generate images for evaluation:

# Download benchmark data
hf download rhli/genarena --repo-type dataset --local-dir ./data

# Generate images with your model
diffgentor edit --backend diffusers \
    --model_name YourModel \
    --input ./data/basic/ \
    --output_dir ./arena/basic/models/yourname_YourModel_20260205/YourModel/

Run Battles for New Models

genarena run --arena_dir ./arena --data_dir ./data \
    --subset basic \
    --exp_name yourname_YourModel_20260205

GenArena automatically detects new models and schedules battles against existing models.

Submit to Official Leaderboard

Coming Soon: The genarena submit command will allow you to submit your evaluation results to the official GenArena leaderboard via GitHub PR.

The workflow will be:

1. Run evaluation locally with genarena run
2. Upload results to your HuggingFace repository
3. Submit via genarena submit which creates a PR for review

Documentation

Document	Description
Quick Start	Installation and basic usage guide
Architecture	System design and key concepts
CLI Reference	Complete command-line interface documentation
Experiment Management	How to organize and manage experiments
FAQ	Frequently asked questions

Citation

@article{li2026genarena,
  title={GenArena: How Can We Achieve Human-Aligned Evaluation for Visual Generation Tasks?},
  author={Li, Ruihang and Qu, Leigang and Zhang, Jingxu and Gui, Dongnan and Xu, Mengde and Zhang, Xiaosong and Hu, Han and Wang, Wenjie and Wang, Jiaqi},
  journal={arXiv preprint arXiv:2602.06013},
  year={2026}
}

License

Apache License 2.0