Rethinking attention modules of SDXL in a flow perspective

Zhongpai Gao, Juxiang Zhou, Jianhou Gan, Jiatian MeiMathematical 2025 · DOI

Abstract

We reinterpreted the attention mechanism in the training-free methods of diffusion models using flow matching theory. Based on this, we proposed the FlowAttnXL training-free framework for text-to-image tasks. It introduces two modules: Flow Matching Cross-Attention (FMXA) and Flow Arithmetic Perturbation Guidance (FAPG). These modules operate plug-and-play, requiring no retraining and compatible with pipelines like Stable Diffusion XL (SDXL). Experiments on MS COCO show significant improvements over state-of-the-art methods. Ablation studies validate the flow-based design, with FMXA improving text-image alignment and FAPG enhancing visual fidelity. This work establishes a unified framework for semantic alignment and feature propagation in diffusion models.
BibTeX
@inproceedings{gao2025rethinking,
  author = {Zhongpai Gao and Juxiang Zhou and Jianhou Gan and Jiatian Mei},
  title = {{Rethinking attention modules of SDXL in a flow perspective}},
  booktitle = {Mathematical},
  year = {2025},
  doi = {10.3934/mfc.2025015},
}

Loading related content...

Figure