Research map / Image generation / Autoregressive and latent generation
SEAL: Semantic Aware Image Watermarking
Kasra Arabi, R. Teal Witter, Chinmay Hegde, Niv Cohen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add spatial test to SEAL detection (ablation) | SEAL · Cat Attack tamper detection | ROC-AUC: 0.551 -> 0.982 | Improved |
| condition watermark detection on image semantics (SEAL vs initial-noise baselines) | SEAL · Private-model forgery attack | AUC: 0.000 -> 0.708 | Improved |
| increase pasted object scale from 10% to 40% of image (ablation) | SEAL + Spatial Test · Cat Attack | ROC-AUC (%): 95.4 -> 98.0 | Improved |
| use SEAL instead of post-hoc/model-based watermarks under regeneration attack | SEAL · Regeneration attack | detection accuracy (%): 51 -> 98 | Improved |
| use caption embedding instead of visual feature vector of proxy image (ablation) | SEAL · related vs unrelated separation | separation | Improved |
| fine-tune embedding model on 10k caption pairs (ablation) | SEAL · related vs unrelated separation | separation | Improved |
| apply SEAL watermark to generation | Stable Diffusion v2 · COCO prompts | CLIP score: 31.365 -> 31.499 | No clear effect |
| use SEAL against image transformation removal attacks | SEAL · Image transformations | ROC-AUC: 0.896 average | Mixed |
Closest papers by meaning
- Black-Box Forgery Attacks on Semantic Watermarks for Diffusion Models (CVPR 2025 · similarity 0.92)
- Invisible Image Watermarks Are Provably Removable Using Generative AI (NeurIPS 2024 · similarity 0.91)
- An Undetectable Watermark for Generative Image Models (ICLR 2025 · similarity 0.90)
- SleeperMark: Towards Robust Watermark against Fine-Tuning Text-to-image Diffusion Models (CVPR 2025 · similarity 0.90)
- Hidden in the Noise: Two-Stage Robust Watermarking for Images (ICLR 2025 · similarity 0.90)
- Watermarks in the Sand: Impossibility of Strong Watermarking for Language Models (ICML 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
