Latent Watermarks under Generative Editing: A Benchmark and Analysis of Detection Survival

Key points

• We benchmark eight watermark methods against five editors across four generative backbones, four editing strengths, and five semantic categories, with edit-validity and threshold checks.1
• Separating editing from seven subsequent distortions reveals that editing alone primarily distinguishes Tree-Ring, while added distortions expose a broader spectrum of detection survival.1
• Sequential edits reveal a second hidden difference: score separation can decline while detection rates remain near their ceiling.1
• Embedding-strength interventions in two methods link higher clean separation to higher post-edit separation.1

Sources (1)

• [1]Latent Watermarks under Generative Editing: A Benchmark and Analysis of Detection Survival
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 09:01 AM

  Ordinary prompt-based editing can cause latent watermark detection to fail without explicitly targeting the watermark.

  We benchmark eight watermark methods against five editors across four generative backbones, four editing strengths, and five semantic categories, with edit-validity and threshold checks.

Extractive summary: sentences quoted from the sources.