Edit Transfer:

Learning Image Editing via Vision In-Context Relations

Lan Chen1,
Qi Mao1,
Yuchao Gu2,
Mike Zheng Shou2,
1MIPG, Communication University of China
2Show Lab, National University of Singapore
arXiv Code

Edit Transfer Examples

Fixed Image
Image 1
Image 2
Image 3
Image 4
Image 5
Image 6
Image 7
Image 8
Image 9
Image 10
Image 11
Image 12
Image 13
Image 14
Image 15
Image 16
Image 17
Image 18
Image 19
Image 20
Image 21
Image 23
Image 24
Image 25
Image 26

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.

Pipeline


Visual relation in-context learning for Edit Transfer. (a) We arrange in-context examples in a four-panel layout: the top row (an editing pair \((I_s, I_t)\)) and the bottom row (the query pair \(( \hat{I}_s, \hat{I}_t )\)). Our goal is to to learn the transformation from \(I_s\) \(\rightarrow\) \(I_t\), and apply it to the bottom-left image \(\hat{I}_s\), producing the target \(\hat{I}_t\), in the bottom-right. (b) We fine-tune a lightweight LoRA in the MMA to better capture visual relations. Noise addition and removal are applied only to \(z_t\), while the conditional tokens \(c_T\) ( derived from \((I_s, I_t, \hat{I}_s)\) ) remain noise-free. (c) Finally, we cast Edit Transfer as an image generation task by initializing the bottom-right latent token \(z_T\) with random noise and concatenating it with the clean tokens \(c_I\). Leveraging the enhanced in-context capability of the fine-tuned DiT blocks, the model generates \(I_t\), effectively transferring the same edits from the top row to the bottom-left image.



Traning data samples.

To encode a clear "source image \(\rightarrow\) edited image" order, we arrange each example-request pair into a 2 × 2 composite: the first row contains the example pair (Is, It), and the second row contains the request pair \((\hat{I}_s, \hat{I}_t)\).



Generalization Performance

Generalization performance of Edit Transfer. Our model demonstrates remarkable generalization by: (b) Generating novel pose variations within a given editing type, even if such variations were unseen during training; (c) Flexibly combining different editing types; (d) Transferring its capabilities across other species

BibTeX

@misc{chen2025edittransferlearningimage,
      title={Edit Transfer: Learning Image Editing via Vision In-Context Relations}, 
      author={Lan Chen and Qi Mao and Yuchao Gu and Mike Zheng Shou},
      year={2025},
      eprint={2503.13327},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.13327}, 
}, 
}