Research map / Image generation / Autoregressive and latent generation
Spherical Watermark: Encryption-Free, Lossless Watermarking for Diffusion Models
Xiaoxiao Hu, Jiaqi Jin, Sheng Li, Wanli Peng, Xinpeng Zhang, Zhenxing Qian
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 |
|---|---|---|---|
| use spherical-mapping lossless watermark instead of Gaussian Shading with fixed keys | Spherical Watermark · COCO FID (SD v2.1) | FID: 49.4379 -> 46.8132 | Improved |
| use spherical watermark instead of Gaussian Shading under adversarial attack | Spherical Watermark · COCO adversarial (SD v2.1) | bit accuracy: 88.06 -> 98.12 | Improved |
| use spherical watermark instead of PRC Watermark under post-processing | Spherical Watermark · COCO post-processing (SD v2.1) | TPR@1%FPR: 87.03 -> 97.50 | Improved |
| use spherical watermark instead of Gaussian Shading under post-processing | Spherical Watermark · COCO post-processing (SD v2.1) | bit accuracy: 98.43 -> 95.02 | Hurt |
| remove binary embedding module (ablation) | Spherical Watermark · COCO (SD v2.1) | latent classifier accuracy | Improved |
| add spherical mapping instead of Gaussian Shading transform (ablation) | Spherical Watermark · COCO brightness | bit accuracy | Improved |
| increase row sparsity s from 1 to 4 (ablation) | Spherical Watermark · COCO brightness x2 | TPR: 99.72 -> 83.68 | Hurt |
| increase repetition count N from 1 to 31 (ablation) | Spherical Watermark · COCO brightness x2 | TPR: 95.10 -> 99.72 | Improved |
| switch ODE solver among DDIM, PNDM, DPM-Solver++ (ablation) | Spherical Watermark · COCO brightness | bit accuracy: 96.06 / 96.17 / 96.02 | No clear effect |
| vary inversion timesteps from 10 to 50 (ablation) | Spherical Watermark · COCO PNG | bit accuracy: 99.97 -> 99.99 | No clear effect |
| replace PRC extraction with spherical extraction | Spherical Watermark · extraction time | time | Improved |
Closest papers by meaning
- Shallow Diffuse: Robust and Invisible Watermarking through Low-Dim Subspaces in Diffusion Models (NeurIPS 2025 · similarity 0.89)
- T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Gaussian Shading: Provable Performance-Lossless Image Watermarking for Diffusion Models (CVPR 2024 · similarity 0.88)
- GaussMarker: Robust Dual-Domain Watermark for Diffusion Models (ICML 2025 · similarity 0.87)
- Hidden in the Noise: Two-Stage Robust Watermarking for Images (ICLR 2025 · similarity 0.87)
- AquaLoRA: Toward White-box Protection for Customized Stable Diffusion Models via Watermark LoRA (ICML 2024 · similarity 0.87)
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.
