Research map / Video / Video generation
Mode Seeking meets Mean Seeking for Fast Long Video Generation
Shengqu Cai, Weili Nie, Chao Liu, Julius Berner, Lvmin Zhang, Nanye Ma, Hansheng Chen, Maneesh Agrawala, Leonidas Guibas, Gordon Wetzstein, Arash Vahdat
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 decoupled DDT dual heads instead of a single velocity head (ablation) | Ours · Ablation (30s) | quality: 0.5298 -> 0.6359 | Improved |
| add sliding-window teacher DMD alignment (ablation) | Ours · Ablation (30s) | quality: 0.6075 -> 0.6359 | Improved |
| add long-video SFT flow-matching anchor (ablation) | Ours · Ablation (30s) | consistency: 0.9579 -> 0.9615 | Improved |
| add long-video SFT flow-matching anchor (ablation) | Ours · Ablation (30s) | motion: 0.9690 -> 0.9685 | Mixed |
| use mode-seeking teacher alignment instead of long-context SFT | Ours vs Long-context SFT · 30-second VBench-Long | aesthetic quality: 0.4973 -> 0.5735 | Improved |
| train on mixed-length videos instead of long videos only | Mixed-length SFT vs Long-context SFT · 30-second VBench-Long | aesthetic quality: 0.4973 -> 0.5467 | Improved |
| use ours instead of InfinityRoPE | Ours · 30-second VBench-Long | dynamic degree: 0.7188 -> 0.9453 | Improved |
| use ours instead of rollout baselines at 60 seconds | Ours vs LongLive · 60-second stress test | dynamic degree: 0.8203 -> 0.9297 | Improved |
| use ours vs CausVid for subject consistency | Ours vs CausVid · 30-second VBench-Long | subject consistency: 0.9736 vs 0.9682 | Hurt |
| use ours instead of other methods in user study | Ours · User study 30s | overall quality: 3.68 vs 3.23 (InfinityRoPE) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- LongLive — LongLive: Real-time Interactive Long Video Generation (ICLR 2026) · better on 4 of 4 shared results
- InfinityRoPE — Infinity-RoPE: Action-Controllable Infinite Video Generation Emerges From Autoregressive Self-Rollout (CVPR 2026) · better on 4 of 4 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
- DynaMem: Consistent Long Video Generation via Hierarchical Memory and Motion Priors (ICML 2026 · similarity 0.89)
- FreeSwim: Revisiting Sliding-Window Attention Mechanisms for Training-Free Ultra-High-Resolution Video Generation (ECCV 2026 · similarity 0.88)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.88)
- Surprise Forcing: What to Remember, When to Skip in Long Video Generation (ECCV 2026 · similarity 0.88)
- LongDiff: Training-Free Long Video Generation in One Go (CVPR 2025 · similarity 0.88)
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 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.
