Research map / Video / Video generation
MEV: A Multi-Event Video Dataset for Long-Take Generation
Peiyuan Zhu, Shaoan Xie, Yifan Shen, Wen Tian, Zijian Li, Guangyi Chen, Kun Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces MEV, a multi-event video dataset intended to supervise long-take text-to-video generation with temporally consistent multi-event descriptions.
Problem
Existing text-to-video datasets mostly contain short single-event clips or coarse captions, giving little supervision for long-take temporal dependencies and causing inconsistencies when multiple events are described.
Approach
A dataset of multi-event videos with event-level annotations aimed at training and evaluating long-take generation; the abstract mentions only the dataset's motivation, not its specific construction.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fine-tune Wan2.1-1.3B on MEV (verified events, grounded boundaries) instead of an UltraVideo-Long subset with uniform boundaries (ablation) | MinT-style multi-event Wan adaptation · MEV-Eval | EOA: 0.33 -> 0.45 | Improved |
| fine-tune on MEV with grounded event conditioning instead of Ultra-Wan data (ablation) | MinT-style multi-event Wan adaptation · MEV-Eval | TGA: 0.42 -> 0.60 | Improved |
| fine-tune on MEV with grounded event conditioning instead of Ultra-Wan data (ablation) | MinT-style multi-event Wan adaptation · MEV-Eval | BTC: 0.50 -> 0.64 | Improved |
| multi-event fine-tuning on MEV added to base Wan2.1-1.3B | Wan2.1-1.3B · MEV-Eval | EEA: 0.42 -> 0.61 | Improved |
| multi-event fine-tuning on MEV added to base Wan2.1-1.3B | Wan2.1-1.3B · MEV-Eval | EOA: 0.20 -> 0.45 | Improved |
| fine-tune on MEV instead of Ultra-Wan data (ablation) | MinT-style multi-event Wan adaptation · MEV test set VBench | Imaging Quality: 63.328 -> 64.924 | Improved |
| fine-tune on MEV instead of Ultra-Wan data (ablation) | MinT-style multi-event Wan adaptation · MEV test set VBench | Subject Consistency: 0.936 -> 0.960 | Improved |
| multi-event fine-tuning on MEV added to base Wan2.1-1.3B | Wan2.1-1.3B · MEV test set VBench | Motion Smoothness: 0.991 -> 0.991 | No clear effect |
| multi-event fine-tuning on MEV added to base Wan2.1-1.3B | Wan2.1-1.3B · MEV test set VBench | Aesthetic Quality: 0.585 -> 0.590 | No clear effect |
| use Gemini 2.5 Pro instead of Qwen3-VL for camera-motion annotation (ablation) | MEV annotation pipeline · camera-motion labels, manual evaluation | exact-set accuracy: 10% -> 22% | Improved |
| autoregressive prompt-switching (LongLive) instead of MEV-grounded fine-tuning | Wan2.1-1.3B · MEV-Eval and VBench | Aesthetic Quality: 0.585 -> 0.504 | Hurt |
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.
- LongLive — LongLive: Real-time Interactive Long Video Generation (ICLR 2026) · better on 5 of 5 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
- Mind the Time: Temporally-Controlled Multi-Event Video Generation (CVPR 2025 · similarity 0.89)
- LVD-2M: A Long-take Video Dataset with Temporally Dense Captions (NeurIPS 2024 · similarity 0.89)
- VidEvent: A Large Dataset for Understanding Dynamic Evolution of Events in Videos (AAAI 2025 · similarity 0.87)
- MovieBench: A Hierarchical Movie Level Dataset for Long Video Generation (CVPR 2025 · similarity 0.87)
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025 · similarity 0.86)
- Beyond Grounding: Extracting Fine-Grained Event Hierarchies across Modalities (AAAI 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.
