This dataset is a multi-domain bennchmark for detecting AI-generated text-rich images from GPT-Image-2. The dataset consists of two complementary subsets:

Fake subset: Images generated by GPT-Image-2 using carefully designed prompts. The prompts are constructed to cover diverse domains and layouts while avoiding reproduction of specific real-world images.

Real subset: Images sampled from six publicly available datasets. These images are selected as… See the full description on the dataset page: https://huggingface.co/datasets/Shuyiww/TextRich.