---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: height
    dtype: uint32
  - name: width
    dtype: uint32
  - name: label
    dtype:
      class_label:
        names:
          '0': real
          '1': fake
  - name: generator
    dtype: large_string
  - name: file_id
    dtype: large_string
  - name: description
    dtype: large_string
  - name: positive_prompt
    dtype: large_string
  - name: negative_prompt
    dtype: large_string
  - name: conditioning
    dtype: large_string
  - name: origin_dataset
    dtype: large_string
  - name: paired_real_images
    large_list: large_string
  splits:
  - name: train
    num_bytes: 28171529545
    num_examples: 144000
  - name: validation
    num_bytes: 7019695896
    num_examples: 36000
  download_size: 35178878540
  dataset_size: 35191225441
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
task_categories:
- image-classification
pretty_name: AI-GenBench (fake part)
---
# AI-GenBench: A New Ongoing Benchmark for AI-Generated Image Detection

Important: this is the fake part of the AI-GenBench dataset. To re-create the original benchmark, which includes real images, please check the [official repository](https://github.com/MI-BioLab/AI-GenBench).
Important 2: before using, please check the licensing terms of the images included!

## Details
The rapid advancement of generative AI has revolutionized image creation, enabling high-quality synthesis from text prompts while raising critical challenges for media authenticity. We present Ai-GenBench, a novel benchmark designed to address the urgent need for robust detection of AI-generated images in real-world scenarios. Unlike existing solutions that evaluate models on static datasets, AI-GenBench introduces a temporal evaluation framework where detection methods are incrementally trained on synthetic images, historically ordered by their generative models, to test their ability to generalize to new generative models, such as the transition from GANs to diffusion models.

More information can be found in our paper: [AI-GenBench: A New Ongoing Benchmark for AI-Generated Image Detection](https://arxiv.org/abs/2504.20865).

![ai_gen_bench_teaser.jpg](https://cdn-uploads.huggingface.co/production/uploads/630e4e44d818ed99de8cb869/oPeu0_0205U9ne0zv-U5b.jpeg)

## Content
This repository, as the name suggests, contains only the fake images of the AI-GenBench dataset. This is the part that gets automatically downloaded when using the simple dataset creation script

## License
The images contained in this dataset are obtained from multiple sources.

- [Aeroblade](https://github.com/jonasricker/aeroblade)
- [Artifact](https://github.com/awsaf49/artifact/)
- [DDMD (Towards the Detection of Diffusion Model Deepfakes)](https://github.com/jonasricker/diffusion-model-deepfake-detection)
- [DMimageDetection](https://github.com/grip-unina/DMimageDetection)
- [DRCT-2M](https://github.com/beibuwandeluori/DRCT)
- [ELSA_D3](https://huggingface.co/datasets/elsaEU/ELSA_D3)
- [SFHQ-T2I](https://github.com/SelfishGene/SFHQ-T2I-dataset)
- [Forensynths](https://peterwang512.github.io/CNNDetection)
- [GenImage](https://genimage-dataset.github.io)
- [Imaginet](https://github.com/delyan-boychev/imaginet)
- [Polardiffshield](https://github.com/qbammey/polardiffshield)
- [Synthbuster](https://www.veraai.eu/posts/dataset-synthbuster-towards-detection-of-diffusion-model-generated-images)

Please check the repository at [https://github.com/MI-BioLab/AI-GenBench](https://github.com/MI-BioLab/AI-GenBench) for more information on the sources of the data, the content breakdown, and exact filelists. Before using, please make sure you understand where the files are coming from and the related licensing terms.