AnyMS: Bottom-up Attention Decoupling for Layout-guided and Training-free Multi-subject Customization

Reading time: 1 minute
...

📝 Original Info

  • Title: AnyMS: Bottom-up Attention Decoupling for Layout-guided and Training-free Multi-subject Customization
  • ArXiv ID: 2512.23537
  • Date: 2025-12-29
  • Authors: Binhe Yu, Zhen Wang, Kexin Li, Yuqian Yuan, Wenqiao Zhang, Long Chen, Juncheng Li, Jun Xiao, Yueting Zhuang

📝 Abstract

Figure 1. AnyMS enables training-free layout-guided multi-subject customization, supporting diverse subject combinations and scaling to larger numbers while maintaining a balance among layout control, text alignment, and identity preservation. See more visualization details and layout configurations in the Appendix.

📄 Full Content

Recent advances in large-scale pre-trained diffusion models [4,29,33,35,38] have enabled the novel application of customized image generation [1,7,37,42], allowing users to generate images containing specific subjects of interest. While single-subject customization has achieved remarkable success [7,8,24,37], recent research has advanced toward the more challenging task of multi-subject customization [10,19,26]. This paradigm focuses on synthesizing multiple custom subjects into coherent scenes guided by textual prompts, thereby offering greater flexibility and personalization in user-driven content creation.

…(본문이 길어 일부가 생략되었습니다.)

Reference

This content is AI-processed based on open access ArXiv data.

Start searching

Enter keywords to search articles

↑↓
ESC
⌘K Shortcut