In-Context Brush: Zero-shot Customized Subject Insertion with Context-Aware Latent Space Manipulation
Yu Xu, Fan Tang, You Wu, Lin Gao, Oliver Deussen, Hongbin Yan, Jintao Li, Juan Cao, Tong-Yee Lee · 2025
Read In-Context Brush: Zero-shot Customized Subject Insertion with Context-Aware Latent Space Manipulation by Yu Xu, Fan Tang, You Wu, Lin Gao, Oliver Deussen, Hongbin Yan, Jintao Li, Juan Cao, Tong-Yee Lee (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-20271.txt.
Description
Pasting a chosen object into a photo usually needs model retraining or fiddly controls. Yu Xu and co-authors proposed in 2025 In-Context Brush, a zero-shot insertion method. It treats the object photo and text prompt as demonstrations, and the masked target image as the question.
Built on a pretrained inpainting network, the method tweaks the latent space at test time on two levels. Features shift inside each attention head toward the subject. Heads are also reweighted to strengthen prompt control. Experiments showed better identity preservation, prompt alignment, and image quality than rivals, with no extra training or data.
Catalog record
| CREATOR | Yu Xu, Fan Tang, You Wu, Lin Gao, Oliver Deussen, Hongbin Yan, Jintao Li, Juan Cao, Tong-Yee Lee |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-20271