GPT-IMAGE-EDIT-1.5M: A Million-Scale, GPT-Generated Image Dataset
Yuhan Wang, Siwei Yang, Bingchen Zhao, Letian Zhang, Qing Liu, Yuyin Zhou, Cihang Xie
2025-07-29
Summary
This paper talks about GPT-IMAGE-EDIT-1.5M, which is a huge dataset made by an AI called GPT that helps teach other AI models how to edit images based on instructions. It is open for anyone to use and helps improve image editing with instructions.
What's the problem?
The problem is that teaching AI to edit images using instructions usually requires a lot of special data, and much of this data is owned by private companies, so other researchers don't get to use it. This limits how well open-source models can learn to do image editing.
What's the solution?
The paper created a very large public dataset using GPT-generated instructions and image edits, then used this dataset to fine-tune open-source AI models. This helped these models perform almost as well as private, closed models for instruction-guided image editing.
Why it matters?
This matters because it makes powerful image editing AI more accessible to everyone, not just big companies. Anyone can use this data and models to build better tools for editing images with simple instructions, which can help artists, designers, and everyday users.
Abstract
A publicly available dataset, GPT-IMAGE-EDIT-1.5M, enhances instruction-guided image editing by fine-tuning open-source models to achieve competitive performance with proprietary models.