Research map / Video / Video generation
Rolling Forcing: Autoregressive Long Video Diffusion in Real Time
Kunhao Liu, Wenbo Hu, Jiale Xu, Ying Shan, Shijian 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 |
|---|---|---|---|
| use rolling-window joint denoising at inference instead of frame-by-frame (ablation) | Rolling Forcing · MovieGen 30s videos | quality drift: 5.53 -> 0.01 | Improved |
| train with rolling windows instead of frame-by-frame (ablation) | Rolling Forcing · MovieGen 30s videos | quality drift: 0.89 -> 0.01 | Improved |
| add Self Forcing objective to mixed training (ablation) | Rolling Forcing · MovieGen 30s videos | imaging quality: 62.00 -> 70.75 | Improved |
| add attention sink (global context frames) (ablation) | Rolling Forcing · MovieGen 30s videos | subject consistency: 83.22 -> 92.80 | Improved |
| use Rolling Forcing instead of Self Forcing | Wan2.1-T2V-1.3B · MovieGen 30s videos | quality drift: 1.66 -> 0.01 | Improved |
| compare Rolling Forcing to Self Forcing on dynamic degree | Wan2.1-T2V-1.3B · Full VBench | dynamic degree: 68.05 -> 60.14 | Hurt |
| compare throughput to Self Forcing | Wan2.1-T2V-1.3B · MovieGen | throughput (FPS): 15.38 -> 15.79 | Improved |
| place global-context RoPE indices at fixed 0:L_glo instead of just before temporal context (ablation) | Rolling Forcing · qualitative | video artifacts | Hurt |
| place global-context frames within the denoising window (ablation) | Rolling Forcing · qualitative | video dynamics | Hurt |
Later papers that compared with this method
19 later accepted papers report Rolling-Forcing as a baseline in their results tables; 17 of them report a better number on most of the results they share with it.
- Infinity-RoPE: Action-Controllable Infinite Video Generation Emerges From Autoregressive Self-Rollout (CVPR 2026 · better on 7 of 7 shared results)
- Deep Forcing: Training-Free Long Video Generation with Deep Sink and Participative Compression (ICML 2026 · better on 3 of 6 shared results)
- PackForcing: Short Video Training Suffices for Long Video Sampling and Long Context Inference (ECCV 2026 · better on 5 of 6 shared results)
- OmniMem: Scalable and Adaptive Memory Retrieval for Long Video Generation (NeurIPS 2026 · better on 3 of 3 shared results)
- Learning-to-Memorize: Dynamic Context Management for Long-Horizon Autoregressive Video Generation (NeurIPS 2026 · better on 3 of 3 shared results)
- AdaState: Self-Evolving Anchors for Streaming Video Generation (NeurIPS 2026 · better on 3 of 3 shared results)
- Surprise Forcing: What to Remember, When to Skip in Long Video Generation (ECCV 2026 · better on 3 of 3 shared results)
- Test-Time Noise Guided Adaptation for Realistic Autoregressive Video Generation (ECCV 2026 · better on 3 of 3 shared results)
11 more not listed.
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
- Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion (NeurIPS 2025 · similarity 0.90)
- Equilibrium Forcing: Adaptive Video Generation Without Noise Conditioning (NeurIPS 2026 · similarity 0.89)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.89)
- AdaState: Self-Evolving Anchors for Streaming Video Generation (NeurIPS 2026 · similarity 0.88)
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.88)
- RAVEN: Real-time Autoregressive Video Extrapolation with Consistency-model GRPO (NeurIPS 2026 · 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.
