---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: metadata
    dtype: string
  - name: original_image
    dtype: image
  - name: condition_gray_background
    dtype: image
  - name: condition_rgba
    dtype: image
  - name: subject_mask
    dtype: image
  - name: condition_rgba_variants
    sequence: image
  - name: condition_white_variants
    sequence: image
  splits:
  - name: test
    num_bytes: 3658370666
    num_examples: 1000
  download_size: 3621645033
  dataset_size: 3658370666
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
license: apache-2.0
---

# CreatiDesign Benchmark

## Overview

CreatiDesign Benchmark is a comprehensive benchmark for evaluating multi-conditional graphic design generation models. It contains 1,000 carefully curated samples spanning real-world design scenarios, including movie posters, product advertisements, brand promotions, and social media content. The benchmark focuses on assessing models' fine-grained controllability over multiple heterogeneous conditions—such as primary visual elements (main images), secondary visual elements (decorative objects), and textual elements—while also measuring overall visual quality and strict adherence to user intent.

## Download and Usage

```python
from datasets import load_dataset
dataset_repo = 'HuiZhang0812/CreatiDesign_benchmark'
test_dataset = load_dataset(dataset_path, split='test')
```
To evaluate your model's graphic design generation capabilities using the CreatiDesign Benchmark, please follow [CreatiDesign](https://github.com/HuiZhang0812/CreatiDesign).


```
@article{zhang2025creatidesign,
  title={CreatiDesign: A Unified Multi-Conditional Diffusion Transformer for Creative Graphic Design},
  author={Zhang, Hui and Hong, Dexiang and Yang, Maoke and Chen, Yutao and Zhang, Zhao and Shao, Jie and Wu, Xinglong and Wu, Zuxuan and Jiang, Yu-Gang},
  journal={arXiv preprint arXiv:2505.19114},
  year={2025}
}
```