GPT Image Edit-1.5M is an image generation dataset released by the University of California, Santa Cruz and the University of Edinburgh in 2025. The related paper results are "GPT-IMAGE-EDIT-1.5M: A Million-Scale, GPT-Generated Image Dataset", which aims to provide comprehensive multimodal data resources for the training and evaluation of image editing models. This dataset contains more than 1.5 million high-quality triplets (instructions, source images, edited images). It unifies and optimizes three popular image editing datasets: OmniEdit, HQ-Edit, and UltraEdit.

Dataset structure:

• HQ-Edit (183,182): A fully synthetic image editing dataset using high-quality input and output images
• UltraEdit (100,008): A comprehensive image editing dataset covering 9 editing tasks
• OmniEdit (1,270,385): A large-scale multi-task dataset containing original, rewritten, and Complex-Edit style commands

  

Citation

@misc{wang2025gptimageedit15mmillionscalegptgeneratedimage,
title={GPT-IMAGE-EDIT-1.5M: A Million-Scale, GPT-Generated Image Dataset},
author={Yuhan Wang and Siwei Yang and Bingchen Zhao and Letian Zhang and Qing Liu and Yuyin Zhou and Cihang Xie},
year={2025},
eprint={2507.21033},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2507.21033},
}