Research map / Video / Video generation
TEAR: Temporal-aware Automated Red-teaming for Text-to-Video Models
Jiaming He, Guanyu Hou, Hongwei Li, Zhicong Huang, Kangjie Chen, Yi Yu, Wenbo Jiang, Guowen Xu, Tianwei Zhang
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 |
|---|---|---|---|
| increase refinement rounds from direct (0) to 8 (ablation) | TEAR refine model · five T2V models | ASR: 57%-71% -> 83%-95% | Improved |
| increase refinement rounds from direct to 8 (ablation) | TEAR refine model · five T2V models | NSFW filter pass rate: 79%-86% -> above 95% | Improved |
| use temporal-aware generator instead of adapted T2I red-teaming (FLIRT) | automated red-teaming · Hunyuan-Video | ASR: 57.2% -> 82.3% | Improved |
| use temporal-aware generator instead of FLIRT | automated red-teaming · Wan 2.2 | ASR: 56.4% -> 80.5% | Improved |
| remove seed prompts (seed-free generation) (ablation) | TEAR · Hunyuan-Video | ASR: 82.3% -> 79.2% | No clear effect |
| remove seed prompts (seed-free generation) (ablation) | TEAR · Wan 2.2 | ASR: 80.5% -> 76.9% | No clear effect |
| raise inference steps up to about 50 (ablation) | target T2V video generation · Hunyuan-Video | ASR: 76.41 -> 82.79 | Improved |
| vary embedded CFG scale (ablation) | target T2V video generation · Hunyuan-Video | ASR | Mixed |
| transfer optimized prompts to unseen T2V models | TEAR prompts · five T2V models | ASR: average 76.4% | Improved |
| add temporal rewriting vs naive prompts | prompting · Hunyuan-Video | ASR: 2.6% -> 82.3% | Improved |
Closest papers by meaning
- T2VSafetyBench: Evaluating the Safety of Text-to-Video Generative Models (NeurIPS 2024 · similarity 0.89)
- RISE: Red-teaming via Iterative Strategy Evolution for Modern Text-to-Image Models (NeurIPS 2026 · similarity 0.88)
- Red-Teaming Text-to-Image Systems by Rule-based Preference Modeling (NeurIPS 2025 · similarity 0.88)
- Video-SafetyBench: A Benchmark for Safety Evaluation of Video LVLMs (NeurIPS 2025 · similarity 0.87)
- T2V-OptJail: Discrete Prompt Optimization for Text-to-Video Jailbreak Attacks (NeurIPS 2025 · similarity 0.86)
- ART: Automatic Red-teaming for Text-to-Image Models to Protect Benign Users (NeurIPS 2024 · similarity 0.86)
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.
