Research map / Video / Video generation
WaTeRFlow: Watermark Temporal Robustness via Flow Consistency
Utae Jeong, Sumin In, Hyunju Ryu, Jaewan Choi, Feng Yang, Jongheon Jeong, Seungryong Kim, Sangpil Kim
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 FUSE video generation branch and TCL (ablation) | WaTeRFlow · SVD-XT frames | avg bit accuracy: 66.32 -> 84.96 | Improved |
| add temporal consistency loss (TCL) (ablation) | WaTeRFlow · SVD-XT frames | avg bit accuracy: 79.11 -> 84.96 | Improved |
| add semantic preservation loss (ablation) | WaTeRFlow · SVD-XT first frame | first-frame bit accuracy: 89.63 -> 96.93 | Improved |
| add FUSE image editing branch (ablation) | WaTeRFlow · SVD-XT after editing | avg bit accuracy: 77.47 -> 80.20 | Improved |
| use AnimateLCM proxy instead of SVD with CFG (ablation) | FUSE video proxy · SVD-XT frames | bit accuracy / cost: 83.83 -> 84.96; VRAM 93.80 -> 44.51 GB | Mixed |
| use SVD proxy without CFG (ablation) | FUSE video proxy · SVD-XT frames | avg bit accuracy: 84.96 -> 72.93 | Hurt |
| use DINOv2 instead of CLIP for semantic loss (ablation) | semantic preservation · SVD-XT | avg bit accuracy: 84.96 -> 83.49 | Hurt |
| use WaTeRFlow vs VINE | watermarking · SVD-XT I2V | avg bit accuracy: 73.92 -> 84.96 | Improved |
| remove optical-flow estimator in training (ablation) | FUSE · training stability | BCE loss | Hurt |
Closest papers by meaning
- Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances (ICLR 2025 · similarity 0.88)
- A Locally Tokenized Generative Model for Robust Time-Series Watermarking (NeurIPS 2026 · similarity 0.85)
- FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis (CVPR 2024 · similarity 0.85)
- Safe-Sora: Safe Text-to-Video Generation via Graphical Watermarking (NeurIPS 2025 · similarity 0.85)
- T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models (NeurIPS 2025 · similarity 0.85)
- RAW: A Robust and Agile Plug-and-Play Watermark Framework for AI-Generated Images with Provable Guarantees (NeurIPS 2024 · similarity 0.85)
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.
