InsightEdit: Towards Better Instruction Following for Image Editing

Overview

This repository contains the official implementation of InsightEdit. In this work, we first employed a novel data construction workflow to carefully curate the AdvancedEdit dataset, building a large‑scale, high‑quality dataset with complex instructions and consistent backgrounds. Subsequently, to further enrich image information, we introduced a dual‑stream bridging mechanism that leverages textual and visual features inferred by a multimodal large language model (MLLM) to more precisely guide the image editing process.

Dataset Construction Process

Dataset Example

Method

News

• [May 12, 2024] ⏰ I am rapidly reorganizing the corresponding code, materials, and dataset.
• [Nov 26, 2024] 🎉 The paper has been published on arXiv.

Citation

If you find this work useful for your research, please consider citing:

bibtex
@article{xu2024insightedit,
  title={InsightEdit: Towards Better Instruction Following for Image Editing},
  author={Xu, Yingjing and Kong, Jie and Wang, Jiazhi and Pan, Xiao and Lin, Bo and Liu, Qiang},
  journal={arXiv preprint arXiv:2411.17323},
  year={2024}
}