Research map / Video / Video generation
Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillation
Yunhong Lu, Yanhong Zeng, Haobo Li, Hao Ouyang, Qiuyu Wang, Ka Leong Cheng, Jiapeng Zhu, Hengyuan Cao, Zhipeng Zhang, Xing Zhu, Yujun Shen, Min 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 |
|---|---|---|---|
| add Re-DMD training (ablation) | Reward Forcing · Ablation (long video) | dynamic score: 43.75 -> 64.06 | Improved |
| add Re-DMD training (ablation) | Reward Forcing · Ablation (long video) | drift: 1.77 -> 2.51 | Hurt |
| add Re-DMD training (ablation) | Reward Forcing · Ablation (long video) | background consistency: 95.85 -> 95.07 | Hurt |
| add EMA Sink (ablation) | Reward Forcing · Ablation (long video) | dynamic score: 35.15 -> 64.06 | Improved |
| add EMA Sink (ablation) | Reward Forcing · Ablation (long video) | motion smoothness: 98.64 -> 98.82 | Improved |
| add EMA Sink (ablation) | Reward Forcing · Ablation (long video) | drift: 2.65 -> 2.51 | Improved |
| add sink token (ablation) | Reward Forcing · Ablation (long video) | drift: 5.08 -> 2.51 | Improved |
| add sink token (ablation) | Reward Forcing · Ablation (long video) | dynamic score: 51.56 -> 64.06 | Improved |
| lower EMA coefficient alpha from 0.99 to 0.9 (ablation) | Reward Forcing · Ablation (long video) | motion smoothness / drift: smoothness 98.96 -> 99.09, drift 2.52 -> 3.23 | Mixed |
| lower EMA coefficient alpha from 0.99 to 0.5 (ablation) | Reward Forcing · Ablation (long video) | drift: 2.52 -> 3.78 | Hurt |
| lower reward temperature beta to 1/5 (higher reward weight) (ablation) | Reward Forcing · Ablation (long video) | dynamic / background / smoothness: dynamic 94.53, background 92.40, smoothness 96.40 | Mixed |
| lower beta from 1 to 1/2 (higher reward weight) (ablation) | Reward Forcing · Ablation (long video) | dynamic score: 54.68 -> 64.06 | Improved |
Later papers that compared with this method
5 later accepted papers report Reward Forcing as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.
- VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion (NeurIPS 2026 · better on 2 of 3 shared results)
- AdaState: Self-Evolving Anchors for Streaming Video Generation (NeurIPS 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)
- RAVEN: Real-time Autoregressive Video Extrapolation with Consistency-model GRPO (NeurIPS 2026 · better on 2 of 2 shared results)
- Helios: Real Real-Time Long Video Generation Model (TPAMI 2026 · better on 2 of 2 shared results)
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 5 of 5 shared results
- Rolling Forcing — Rolling Forcing: Autoregressive Long Video Diffusion in Real Time (ICLR 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
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.91)
- AdaState: Self-Evolving Anchors for Streaming Video Generation (NeurIPS 2026 · similarity 0.89)
- CoDMD: Copula-aware Distribution Matching Distillation for Fast Video Generation (NeurIPS 2026 · similarity 0.87)
- Improving Video Generation with Human Feedback (NeurIPS 2025 · similarity 0.87)
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025 · similarity 0.87)
- Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion (NeurIPS 2025 · 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.
