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

Key points

• The reliability and accountability of image generative models (IGMs) are essential for building responsible and trustworthy AI systems.1
• 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.1
• 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.1
• 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.1

Sources (1)

• [1]When Scene Text Hijacks the Scene: Uncovering, Exploiting, and Mitigating Rendered-Text Semantic Leakage in Image Generation Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 05:50 AM

  In this work, we study rendered-text semantic leakage, a largely overlooked phenomenon in open-domain text rendering.

  The reliability and accountability of image generative models (IGMs) are essential for building responsible and trustworthy AI systems.

Extractive summary: sentences quoted from the sources.