Reproduce “BrushEdit: All-In-One Image Inpainting and Editing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yaowei Li, Yuxuan Bian, Xuan Ju, Zhaoyang Zhang, Junhao Zhuang, Ying Shan, Yuexian Zou, Qiang Xu
BrushEdit introduces an all-in-one image editing and inpainting framework that leverages a Multimodal Large Language Model (MLLM) for interpreting natural language instructions and an enhanced dual-branch inpainting model to perform large-scale, interactive edits. The system achieves near-lossless background preservation and superior text alignment, significantly reducing inference time to 3.57 seconds for editing while maintaining high fidelity across diverse mask types.
Key insights
- Integrating MLLMs for instruction interpretation and control signal generation enables autonomous, interactive, and user-friendly free-form editing, moving beyond black-box systems.
- A dual-branch inpainting architecture, with a decoupled control branch and frozen main generation branch, efficiently preserves background fidelity while generating text-aligned foregrounds, allowing large-scale structural changes.
- Developing an 'all-in-one' inpainting model that generalizes to arbitrary mask shapes from a single unified training significantly enhances practical applicability and versatility.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.