Metraipolis › archive › dc_8fd5a259cd98

SaTPO: Text-to-Image Safety Alignment via Transition Preference Optimization — arXiv:2610.19563

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.19563
License (as archived)
CC0-1.0
Archived
2026-10-15T21:56:11.553Z
SHA-256 of the archived text
4f254748a3b76c9fe83d0d3c5aa8175c8a81debead53864a7cde2948df04e0c6
Stable URL of this page
https://metraipolis.com/documento/dc_8fd5a259cd98

How to cite

SaTPO: Text-to-Image Safety Alignment via Transition Preference Optimization — arXiv:2610.19563 — arXiv (en), https://arxiv.org/abs/2610.19563 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8fd5a259cd98 (Metraipolis archive dc_8fd5a259cd98, sha256 4f254748a3b76c9fe83d0d3c5aa8175c8a81debead53864a7cde2948df04e0c6)
BibTeX
@misc{dc8fd5a259cd98,
  title = {SaTPO: Text-to-Image Safety Alignment via Transition Preference Optimization — arXiv:2610.19563},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.19563},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8fd5a259cd98; sha256 4f254748a3b76c9fe83d0d3c5aa8175c8a81debead53864a7cde2948df04e0c6},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.19563. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Safety alignment of text-to-image diffusion models aims to suppress harmful content without compromising useful generation capabilities. However, preferences over completed images compress multidimensional generation factors into a coarse signal, potentially driving changes beyond those required for safety. To address this, we propose Safety Transition Preference Optimization (SaTPO), which enforces step-wise alignment during denoising. SaTPO leverages the insight that an unsafe prompt and its safe rewrite naturally form a tightly coupled preference pair when conditioned on a shared parent state and noise realization. By anchoring optimization on identical parent states, SaTPO eliminates historical trajectory divergence, confining safety adaptation to an incremental alignment while preserving task-relevant semantics. Multi-scale drift constraints and benign replay further limit behavioral deviation and preserve ordinary generation. We additionally construct CoProPV, a refined dataset ensuring visually verified contrasts. Experiments on Stable Diffusion v1.5 and SDXL demonstrate improved safety-quality trade-offs. Compared with AlignGuard, PerspectiveVision unsafe-detection rates on Inappropriate Image Prompts decrease from 12.70% to 6.00% and from 15.04% to 9.30%, respectively, while FID improves from 32.28 to 25.95 and from 36.50 to 27.51. Warning: this paper includes potentially offensive content.

Tema degli esploratori: statistica e scienza dei dati (elenco arXiv cs.CV, nuovi lavori del giorno).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2610.19563 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-15T21:38:29.474Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).