Research map / Video / Video-language understanding
Skyra: AI-Generated Video Detection via Grounded Artifact Reasoning
Yifei Li, Wenzhao Zheng, Yanran Zhang, Runze Sun, Yu Zheng, Lei Chen, Jie Zhou, Jiwen Lu
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 chain-of-thought grounded explanation to the SFT answer (ablation) | Skyra-SFT · ViF-Bench | accuracy: 54.04 -> 90.11 | Improved |
| add SFT cold-start before RL (ablation) | Skyra-RL · ViF-Bench | accuracy: 50.09 -> 91.02 | Improved |
| add GRPO RL stage after SFT (ablation) | Skyra-SFT · ViF-Bench | accuracy: 90.11 -> 91.02 | Improved |
| add RL stage after SFT, especially for hard I2V samples (ablation) | Skyra-SFT · ViF-Bench | recall: 84.65 -> 88.35 | Improved |
| use asymmetric accuracy reward with heavier false-positive penalty (ablation) | Skyra-RL · ViF-Bench | accuracy: 76.24 -> 91.02 | Improved |
| use inspection (check-block) reward instead of plain format reward (ablation) | Skyra-RL · ViF-Bench | accuracy: 90.05 -> 91.02 | Improved |
| RL adaptation on 2.2K unlabeled-annotation GenVideo samples from Skyra-SFT | Skyra · GenVideo (out-of-domain) | accuracy: 71.78 -> 91.00 | Improved |
| add RL stage after SFT on out-of-domain GenVideo (ablation) | Skyra-SFT · GenVideo (out-of-domain) | accuracy: 63.98 -> 71.78 | Improved |
| apply image degradations (JPEG compression, noise, zoom) at test time | Skyra-RL · ViF-Bench | accuracy: 91.02 -> 80.80 (compression) | Hurt |
| prompt off-the-shelf MLLMs with CoT instead of fine-tuning | Qwen2.5-VL, InternVL-3, GPT-4.1-mini, Gemini-2.5-flash · ViF-Bench | accuracy: below 60% | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BusterX++ — BusterX: MLLM-Powered AI-Generated Video Forgery Detection and Explanation (NeurIPS 2026) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Sanity Check for AI-generated Image Detection (ICLR 2025 · similarity 0.86)
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark (CVPR 2024 · similarity 0.85)
- AI-Generated Video Detection via Perceptual Straightening (NeurIPS 2025 · similarity 0.85)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.85)
- VERA: Explainable Video Anomaly Detection via Verbalized Learning of Vision-Language Models (CVPR 2025 · similarity 0.85)
- Needle In A Video Haystack: A Scalable Synthetic Evaluator for Video MLLMs (ICLR 2025 · 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.
