---
title: "Edit Transfer: Learning Image Editing via Vision In-Context Relations"
canonical_url: "https://www.modelscope.cn/papers/127525"
md_url: "https://www.modelscope.cn/papers/127525.md"
arxiv_id: 2503.13327
published: 2025-07-01
last_updated: 2025-07-01
authors:
  - "Lan Chen"
  - "Qi Mao"
  - "Yuchao Gu"
  - "Mike Zheng Shou"
model_name: "Edit Transfer"
model_developer: "中国传媒大学MIPG实验室，新加坡国立大学Show Lab"
domain:
  - "计算机视觉"
  - "图像编辑"
  - "扩散模型"
  - "上下文学习"
  - "生成式AI"
type:
  - "计算机视觉"
  - "图像编辑"
  - "扩散模型"
  - "上下文学习"
  - "生成式AI"
  - "Computer Vision and Pattern Recognition"
arxiv_url: "https://arxiv.org/abs/2503.13327"
pdf_url: "https://arxiv.org/pdf/2503.13327"
code_link: "https://github.com/black-forest-labs/flux"
---

# Edit Transfer: Learning Image Editing via Vision In-Context Relations

> We introduce a new setting, Edit Transfer, where a model learns a transformation from just a single source-target example and applies it to a new query image. While text-based methods excel at semantic manipulations through textual prompts, they often…

「Edit Transfer: Learning Image Editing via Vision In-Context Relations」是 ModelScope 魔搭社区收录的论文，arXiv 2503.13327，作者为 Lan Chen, Qi Mao, Yuchao Gu et al.，发表于 2025-07-01，属于 计算机视觉、图像编辑、扩散模型 领域。

- **ArXiv**: 2503.13327
- **Published**: 2025-07-01
- **Authors**: Lan Chen, Qi Mao, Yuchao Gu, Mike Zheng Shou
- **Model**: Edit Transfer
- **Developer**: 中国传媒大学MIPG实验室，新加坡国立大学Show Lab
- **Domain**: 计算机视觉, 图像编辑, 扩散模型, 上下文学习, 生成式AI
- **ArXiv URL**: https://arxiv.org/abs/2503.13327
- **PDF**: https://arxiv.org/pdf/2503.13327
- **Code**: https://github.com/black-forest-labs/flux

Source: https://www.modelscope.cn/papers/127525

---

> Edit Transfer：通过视觉关系上下文学习实现图像编辑迁移

## 摘要

本文提出了 Edit Transfer，一种全新的图像编辑范式。该方法从给定的源-目标编辑示例对中学习视觉变换关系，并将其应用到新的查询图像上。与基于文本的图像编辑（TIE）和基于参考的图像编辑（RIE）方法不同，Edit Transfer 基于 FLUX.1-dev（DiT架构）模型，通过提出的视觉关系上下文学习框架，将示例图像和查询图像嵌入到四宫格复合布局中，利用多模态注意力（MMA）机制处理，并通过轻量级 LoRA 微调捕获视觉关系。该方法仅需42张训练图像即可有效处理复杂的非刚性空间变换和组合编辑任务，并在多项指标上优于现有最先进方法。

## Abstract

We introduce a new setting, Edit Transfer, where a model learns a transformation from just a single source-target example and applies it to a new query image. While text-based methods excel at semantic manipulations through textual prompts, they often struggle with precise geometric details (e.g., poses and viewpoint changes). Reference-based editing, on the other hand, typically focuses on style or appearance and fails at non-rigid transformations. By explicitly learning the editing transformation from a source-target pair, Edit Transfer mitigates the limitations of both text-only and appearance-centric references. Drawing inspiration from in-context learning in large language models, we propose a visual relation in-context learning paradigm, building upon a DiT-based text-to-image model. We arrange the edited example and the query image into a unified four-panel composite, then apply lightweight LoRA fine-tuning to capture complex spatial transformations from minimal examples. Despite using only 42 training samples, Edit Transfer substantially outperforms state-of-the-art TIE and RIE methods on diverse non-rigid scenarios, demonstrating the effectiveness of few-shot visual relation learning.