mlsanityData to 7 Oct 2026

Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillation

arXiv 2512.04678 ↗First posted 4 Dec 2025cs.CVcited by 75

Not on the map. Why?

Cited by 35 in Diffusion models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Efficient streaming video generation is critical for simulating interactive and dynamic worlds. Existing methods distill few-step video diffusion models with sliding window attention, using initial frames as sink tokens to maintain attention performance and reduce error accumulation. However, video frames become overly dependent on these static tokens, resulting in copied initial frames and diminished motion dynamics. To address this, we introduce Reward Forcing, a novel framework with two key designs. First, we propose EMA-Sink, which maintains fixed-size tokens initialized from initial frames and continuously updated by fusing evicted tokens via exponential moving average as they exit the sliding window. Without additional computation cost, EMA-Sink tokens capture both long-term context and recent dynamics, preventing initial frame copying while maintaining long-horizon consistency. Second, to better distill motion dynamics from teacher models, we propose a novel Rewarded Distribution Matching Distillation (Re-DMD). Vanilla distribution matching treats every training sample equally, limiting the model's ability to prioritize dynamic content. Instead, Re-DMD biases the model's output distribution toward high-reward regions by prioritizing samples with greater dynamics rated by a vision-language model. Re-DMD significantly enhances motion quality while preserving data fidelity. We include both quantitative and qualitative experiments to show that Reward Forcing achieves state-of-the-art performance on standard benchmarks while enabling high-quality streaming video generation at 23.1 FPS on a single H100 GPU.

Key facts. Cited by 75 papers in the map (data through 7 Oct 2026). Builds on 85 of its 94 references that are in the map. First posted on arXiv 4 Dec 2025.

Builds on

85 of 94 references.

References · 85

  1. 2017Attention Is All You Needcited by 59,932
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  3. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  4. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
  5. 2025Qwen3 Technical Reportcited by 8,195
  6. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  7. 2022Scalable Diffusion Models with Transformerscited by 5,931
  8. 2025Qwen2.5-VL Technical Reportcited by 5,418
  9. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,581
  10. 2021Improved Denoising Diffusion Probabilistic Modelscited by 3,736
  11. 2022Elucidating the Design Space of Diffusion-Based Generative Modelscited by 2,994
  12. 2025Wan: Open and Advanced Large-Scale Video Generative Modelscited by 2,536
38 more
  1. 2023Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasetscited by 2,526
  2. 2022Video Diffusion Modelscited by 2,280
  3. 2024CogVideoX: Text-to-Video Diffusion Models with An Expert Transformercited by 2,198
  4. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,115
  5. 2022Imagen Video: High Definition Video Generation with Diffusion Modelscited by 1,765
  6. 2023VBench: Comprehensive Benchmark Suite for Video Generative Modelscited by 1,755
  7. 2022Make-A-Video: Text-to-Video Generation without Text-Video Datacited by 1,725
  8. 2024HunyuanVideo: A Systematic Framework For Large Video Generative Modelscited by 1,597
  9. 2023One-step Diffusion with Distribution Matching Distillationcited by 978
  10. 2022CogVideo: Large-scale Pretraining for Text-to-Video Generation via Transformerscited by 973
  11. 2024WorldSimBench: Towards Video Generation Models as World Simulatorscited by 867
  12. 2024Improved Distribution Matching Distillation for Fast Image Synthesiscited by 648
  13. 2024Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusioncited by 634
  14. 2024Movie Gen: A Cast of Media Foundation Modelscited by 614
  15. 2025Flow-GRPO: Training Flow Matching Models via Online RLcited by 605
  16. 2025Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusioncited by 569
  17. 2022Phenaki: Variable Length Video Generation From Open Domain Textual Descriptioncited by 475
  18. 2023Aligning Text-to-Image Models using Human Feedbackcited by 418
  19. 2024From Slow Bidirectional to Fast Autoregressive Video Diffusion Modelscited by 401
  20. 2025DanceGRPO: Unleashing GRPO on Visual Generationcited by 388
  21. 2022Latent Video Diffusion Models for High-Fidelity Long Video Generationcited by 385
  22. 2024Pyramidal Flow Matching for Efficient Video Generative Modelingcited by 354
  23. 2023LAVIE: High-Quality Video Generation with Cascaded Latent Diffusion Modelscited by 349
  24. 2024Open-Sora Plan: Open-Source Large Video Generation Modelcited by 286
  25. 2025Improving Video Generation with Human Feedbackcited by 267
  26. 2025MAGI-1: Autoregressive Video Generation at Scalecited by 262
  27. 2025Seedance 1.0: Exploring the Boundaries of Video Generation Modelscited by 247
  28. 2024StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Textcited by 242
  29. 2025LongLive: Real-time Interactive Long Video Generationcited by 237
  30. 2023SEINE: Short-to-Long Video Diffusion Model for Generative Transition and Predictioncited by 231
  31. 2025SkyReels-V2: Infinite-length Film Generative Modelcited by 218
  32. 2025Rolling Forcing: Autoregressive Long Video Diffusion in Real Timecited by 209
  33. 2025Video models are zero-shot learners and reasonerscited by 208
  34. 2025Unified Reward Model for Multimodal Understanding and Generationcited by 194
  35. 2024Autoregressive Video Generation without Vector Quantizationcited by 180
  36. 2025Self-Forcing++: Towards Minute-Scale High-Quality Video Generationcited by 176
  37. 2023FreeNoise: Tuning-Free Longer Video Diffusion via Noise Reschedulingcited by 173
  38. 2023NUWA-XL: Diffusion over Diffusion for eXtremely Long Video Generationcited by 170

50 most cited of 85 All 85 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 1; 2026: 74.

75 papers.

Sort:

By area:

  1. 2026Advancing Open-source World Modelscited by 144
  2. 2026Helios: Real Real-Time Long Video Generation Modelcited by 56
  3. 2026WoVR: World Models as Reliable Simulators for Post-Training VLA Policies with RLcited by 46
  4. 2026Causal Forcing++: Scalable Few-Step Autoregressive Diffusion Distillation for Real-Time Interactive Video Generationcited by 44
  5. 2026LoL: Longer than Longer, Scaling Video Generation to Hourcited by 29
  6. 2026ShotStream: Streaming Multi-Shot Video Generation for Interactive Storytellingcited by 22
  7. 2026Relax Forcing: Relaxed KV-Memory for Consistent Long Video Generationcited by 20
  8. 2026Manifold-Aware Exploration for Reinforcement Learning in Video Generationcited by 14
  9. 2026Past- and Future-Informed KV Cache Policy with Salience Estimation in Autoregressive Video Diffusioncited by 13
  10. 2026HiAR: Efficient Autoregressive Long Video Generation via Hierarchical Denoisingcited by 13
  11. 2026Hallo-Live: Real-Time Streaming Joint Audio-Video Avatar Generation with Asynchronous Dual-Stream and Human-Centric Preference Distillationcited by 13
  12. 2026One-Forcing: Towards Stable One-Step Autoregressive Video Generationcited by 13
38 more
  1. 2026Pathwise Test-Time Correction for Autoregressive Long Video Generationcited by 12
  2. 2026Mode Seeking meets Mean Seeking for Fast Long Video Generationcited by 12
  3. 2026LongLive-RAG: A General Retrieval-Augmented Framework for Long Video Generationcited by 12
  4. 2026Salt: Self-Consistent Distribution Matching with Cache-Aware Training for Fast Video Generationcited by 11
  5. 2026Forcing-KV: Hybrid KV Cache Compression for Efficient Autoregressive Video Diffusion Modelscited by 10
  6. 2025LongVie 2: Multimodal Controllable Ultra-Long Video World Modelcited by 9
  7. 2026Light Forcing: Accelerating Autoregressive Video Diffusion via Sparse Attentioncited by 9
  8. 2026KVPO: ODE-Native GRPO for Autoregressive Video Alignment via KV Semantic Explorationcited by 9
  9. 2026Long-Horizon Streaming Video Generation via Hybrid Attention with Decoupled Distillationcited by 7
  10. 2026Efficient Video Diffusion Models: Advancements and Challengescited by 6
  11. 2026RAVEN: Real-time Autoregressive Video Extrapolation with Consistency-model GRPOcited by 6
  12. 2026Advancing Narrative Long Video Generation via Training-Free Identity-Aware Memorycited by 6
  13. 2026Beyond VLM-Based Rewards: Diffusion-Native Latent Reward Modelingcited by 5
  14. 2026Video Generation Models as World Models: Efficient Paradigms, Architectures and Algorithmscited by 5
  15. 2026Grounded Forcing: Bridging Time-Independent Semantics and Proximal Dynamics in Autoregressive Video Synthesiscited by 5
  16. 2026Human Cognition in Machines: A Unified Perspective of World Modelscited by 5
  17. 2026Head Forcing: Long Autoregressive Video Generation via Head Heterogeneitycited by 5
  18. 2026DySink: Dynamic Frame Sinks for Autoregressive Long Video Generationcited by 5
  19. 2026AAD-1: Asymmetric Adversarial Distillation for One-Step Autoregressive Video Generationcited by 5
  20. 2026Adapting VACE for Real-Time Autoregressive Video Diffusioncited by 4
  21. 2026VIGOR: VIdeo Geometry-Oriented Reward for Temporal Generative Alignmentcited by 4
  22. 2026UniTemp: Unlocking Video Generation in Any Temporal Order via Bidirectional Distillationcited by 4
  23. 2026Exploring Data-Free LoRA Transferability for Video Diffusion Modelscited by 3
  24. 2026DMGD: Train-Free Dataset Distillation with Semantic-Distribution Matching in Diffusion Modelscited by 3
  25. 2026Stream-T1: Test-Time Scaling for Streaming Video Generationcited by 3
  26. 2026FashionChameleon: Towards Real-Time and Interactive Human-Garment Video Customizationcited by 3
  27. 2026VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusioncited by 3
  28. 2026HeadCast: Casting Attention Heads for Efficient Autoregressive Video Generationcited by 3
  29. 2026Stream-R1: Reliability-Perplexity Aware Reward Distillation for Streaming Video Generationcited by 2
  30. 2026Goodbye Drift: Anchored Tree Sampling for Long-Horizon Video-to-Video Generationcited by 2
  31. 2026AdaState: Self-Evolving Anchors for Streaming Video Generationcited by 2
  32. 2026DisCo: World Models with Discrete Camera Motion Controlcited by 2
  33. 2026Lip Forcing: Few-Step Autoregressive Diffusion for Real-time Lip Synchronizationcited by 2
  34. 2026Ms. Forcing: Efficient Streaming Video Generation with Multi-Scale Patchification and Attentioncited by 2
  35. 2026RECAP-Forcing: Retaining Content Appearances for Long Video Generationcited by 2
  36. 2026Advances in GRPO for Generation Models: A Surveycited by 1
  37. 2026Offline Preference Optimization for Rectified Flow with Noise-Tracked Pairscited by 1
  38. 2026PhyMotion: Structured 3D Motion Reward for Physics-Grounded Human Video Generationcited by 1

50 most cited of 75 All 75 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Causal-rCM: A Unified Teacher-Forcing and Self-Forcing Open Recipe for Autoregressive Diffusion Distillation in Streaming Video Generation and Interactive World Models
  2. 2026Stream Forcing: Constructing Unified Training Trajectory for Robust Streaming Video Generation
  3. 2024DOLLAR: Few-Step Video Generation via Distillation and Latent Reward Optimization
  4. 2026Data-Forcing Distillation: Restoring Diversity and Fidelity in Few-Step Video Generation
  5. 2026Reward Lightning: Fast Video Generation via Homologous Preference Distillation
  6. 2026Memory Forcing: Attendable Mid-Horizon History for Streaming Video Generation
  7. 2025Playing with Transformer at 30+ FPS via Next-Frame Diffusion
  8. 2026VDOT++: Unified Few-Step Video Generation via Unbalanced Optimal Transport Distillation