---
title: "BlobCtrl: Taming Controllable Blob for Element-level Image Editing"
canonical_url: "https://www.modelscope.cn/papers/127301"
md_url: "https://www.modelscope.cn/papers/127301.md"
arxiv_id: 2503.13434
published: 2025-10-01
last_updated: 2025-10-01
authors:
  - "Yaowei Li"
  - "Lingen Li"
  - "Zhaoyang Zhang"
  - "Xiaoyu Li"
  - "Guangzhi Wang"
  - "Hongxiang Li"
  - "Xiaodong Cun"
  - "Ying Shan"
  - "Yuexian Zou"
model_name: BlobCtrl
model_developer: "北京大学、香港中文大学、腾讯 ARC Lab、香港科技大学、大湾区大学"
domain:
  - "计算机视觉"
  - "人工智能"
  - "多模态"
  - "图像编辑"
  - "生成式模型"
type:
  - "计算机视觉"
  - "人工智能"
  - "多模态"
  - "图像编辑"
  - "生成式模型"
  - "Computer Vision and Pattern Recognition"
  - "Artificial Intelligence"
  - Multimedia
arxiv_url: "https://arxiv.org/abs/2503.13434"
pdf_url: "https://arxiv.org/pdf/2503.13434"
code_link: "https://liyaowei-stu.github.io/project/BlobCtrl/"
---

# BlobCtrl: Taming Controllable Blob for Element-level Image Editing

> As user expectations for image editing continue to rise, the demand for flexible, fine-grained manipulation of specific visual elements presents a challenge for current diffusion-based methods. In this work, we present BlobCtrl, a framework for element-level…

「BlobCtrl: Taming Controllable Blob for Element-level Image Editing」是 ModelScope 魔搭社区收录的论文，arXiv 2503.13434，作者为 Yaowei Li, Lingen Li, Zhaoyang Zhang et al.，发表于 2025-10-01，属于 计算机视觉、人工智能、多模态 领域。

- **ArXiv**: 2503.13434
- **Published**: 2025-10-01
- **Authors**: Yaowei Li, Lingen Li, Zhaoyang Zhang, Xiaoyu Li, Guangzhi Wang, Hongxiang Li, Xiaodong Cun, Ying Shan, Yuexian Zou
- **Model**: BlobCtrl
- **Developer**: 北京大学、香港中文大学、腾讯 ARC Lab、香港科技大学、大湾区大学
- **Domain**: 计算机视觉, 人工智能, 多模态, 图像编辑, 生成式模型
- **ArXiv URL**: https://arxiv.org/abs/2503.13434
- **PDF**: https://arxiv.org/pdf/2503.13434
- **Code**: https://liyaowei-stu.github.io/project/BlobCtrl/

Source: https://www.modelscope.cn/papers/127301

---

> BlobCtrl：驯服可控 Blob 实现元素级图像编辑

## 摘要

BlobCtrl 是一个基于概率 Blob（二维高斯分布）表示的元素级图像编辑框架。它将 Blob 作为视觉基元，通过解耦布局与外观，实现对图像中单个对象的添加、平移、缩放、移除和替换等细粒度可控操作。该框架采用上下文内双分支扩散模型架构（前景分支用于身份编码，背景分支用于场景保持与融合），并结合自监督的“解耦-重建”训练范式与身份保持损失函数，在多种元素级编辑任务上取得了领先性能。此外，论文还提出了大规模训练数据集 BlobData 和系统评估基准 BlobBench。

## Abstract

As user expectations for image editing continue to rise, the demand for flexible, fine-grained manipulation of specific visual elements presents a challenge for current diffusion-based methods. In this work, we present BlobCtrl, a framework for element-level image editing based on a probabilistic blob-based representation. Treating blobs as visual primitives, BlobCtrl disentangles layout from appearance, affording fine-grained, controllable object-level manipulation. Our key contributions are twofold: (1) an in-context dual-branch diffusion model that separates foreground and background processing, incorporating blob representations to explicitly decouple layout and appearance, and (2) a self-supervised disentangle-then-reconstruct training paradigm with an identity-preserving loss function, along with tailored strategies to efficiently leverage blob-image pairs. To foster further research, we introduce BlobData for large-scale training and BlobBench, a benchmark for systematic evaluation. Experimental results demonstrate that BlobCtrl achieves state-of-the-art performance in a variety of element-level editing tasks, such as object addition, removal, scaling, and replacement, while maintaining computational efficiency. Project Webpage: https://liyaowei-stu.github.io/project/BlobCtrl/