Metraipolis › archive › dc_ced077ed46d6

When Scene Text Hijacks the Scene: Uncovering, Exploiting, and Mitigating Rendered-Text Semantic Leakage in Image Generation Models — arXiv:2610.11286

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.11286
License (as archived)
CC0-1.0
Archived
2026-10-09T06:30:26.098Z
SHA-256 of the archived text
749ec8b198d2e7950333f10c0a801dcedbbf501d6aa12d6d75d19a082fbafdbc
Stable URL of this page
https://metraipolis.com/documento/dc_ced077ed46d6

How to cite

When Scene Text Hijacks the Scene: Uncovering, Exploiting, and Mitigating Rendered-Text Semantic Leakage in Image Generation Models — arXiv:2610.11286 — arXiv (en), https://arxiv.org/abs/2610.11286 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ced077ed46d6 (Metraipolis archive dc_ced077ed46d6, sha256 749ec8b198d2e7950333f10c0a801dcedbbf501d6aa12d6d75d19a082fbafdbc)
BibTeX
@misc{dcced077ed46d6,
  title = {When Scene Text Hijacks the Scene: Uncovering, Exploiting, and Mitigating Rendered-Text Semantic Leakage in Image Generation Models — arXiv:2610.11286},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.11286},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ced077ed46d6; sha256 749ec8b198d2e7950333f10c0a801dcedbbf501d6aa12d6d75d19a082fbafdbc},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.11286. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

The reliability and accountability of image generative models (IGMs) are essential for building responsible and trustworthy AI systems. Recent IGMs, such as Nano Banana and GPT-Image, now support complex instruction following, realistic image synthesis, and controllable scene-text rendering. As these capabilities expand, safety analysis must also account for new control channels introduced by complex prompts. In this work, we study rendered-text semantic leakage, a largely overlooked phenomenon in open-domain text rendering. Although rendered text is intended to serve as a local visual constraint that should be reproduced verbatim in the generated image, it also carries linguistic semantics that may be interpreted by the model as part of the input instruction. This makes rendered text a potential semantic control channel whose safety implications remain insufficiently understood. We systematically characterize this phenomenon by decoupling the main visual prompt from the rendered text and measuring their individual and compositional effects on generated images. We quantify semantic leakage and rendering fidelity, and further analyze how leakage emerges from intermediate model evidence. We then show that harmful semantics embedded in scene text can persist through LLM-based prompt enhancement pipelines and steer non-text image regions, even when the main visual prompt remains benign. Finally, we propose a preliminary mitigation approach that reduces unsafe semantic transfer from rendered text to non-text regions while preserving the intended text-rendering behavior on FLUX-2-dev. Our findings reveal rendered text as a dual-use carrier of visible data and latent semantics, exposing a text-centric cross-modal attack surface in modern IGMs.

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/2610.11286 — 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 2026-10-09T05:55:07.290Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).