Self-Attention Decomposition For Training Free Diffusion Editing
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911442921324544 |
|---|---|
| author | Anand, Tharun Vali, Mohammad Hassan Solin, Arno Rosh, Green Prasad, BH Pawan |
| author_facet | Anand, Tharun Vali, Mohammad Hassan Solin, Arno Rosh, Green Prasad, BH Pawan |
| contents | Diffusion models achieve remarkable fidelity in image synthesis, yet precise control over their outputs for targeted editing remains challenging. A key step toward controllability is to identify interpretable directions in the model's latent representations that correspond to semantic attributes. Existing approaches for finding interpretable directions typically rely on sampling large sets of images or training auxiliary networks, which limits efficiency. We propose an analytical method that derives semantic editing directions directly from the pretrained parameters of diffusion models, requiring neither additional data nor fine-tuning. Our insight is that self-attention weight matrices encode rich structural information about the data distribution learned during training. By computing the eigenvectors of these weight matrices, we obtain robust and interpretable editing directions. Experiments demonstrate that our method produces high-quality edits across multiple datasets while reducing editing time significantly by 60% over current benchmarks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_22650 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Self-Attention Decomposition For Training Free Diffusion Editing Anand, Tharun Vali, Mohammad Hassan Solin, Arno Rosh, Green Prasad, BH Pawan Computer Vision and Pattern Recognition Diffusion models achieve remarkable fidelity in image synthesis, yet precise control over their outputs for targeted editing remains challenging. A key step toward controllability is to identify interpretable directions in the model's latent representations that correspond to semantic attributes. Existing approaches for finding interpretable directions typically rely on sampling large sets of images or training auxiliary networks, which limits efficiency. We propose an analytical method that derives semantic editing directions directly from the pretrained parameters of diffusion models, requiring neither additional data nor fine-tuning. Our insight is that self-attention weight matrices encode rich structural information about the data distribution learned during training. By computing the eigenvectors of these weight matrices, we obtain robust and interpretable editing directions. Experiments demonstrate that our method produces high-quality edits across multiple datasets while reducing editing time significantly by 60% over current benchmarks. |
| title | Self-Attention Decomposition For Training Free Diffusion Editing |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2510.22650 |