| On-Policy Diffusion Reinforcement Learning Meets Off-Policy Quality Anchoring | ✓ Link | 87.16 | | DiffusionCompass (Wan2.1-14B)+RP | 2026-09-08 |
| PhyRPR: Training-Free Physics-Constrained Video Generation | | 86.36 | | PhyRPR | 2026-01-14 |
| Test-Time Noise Guided Adaptation for Realistic Autoregressive Video Generation | ✓ Link | 86.3 | 86.4 | TANGO | 2026-07-17 |
| M4V: Multimodal Mamba for Efficient Text-to-Video Generation | ✓ Link | 86.14 | 87.56 | M4V* (Wan2.1) (extra data) | 2025-06-12 |
| GRACE: Generation-aware latent compression for efficient video generation | ✓ Link | 85.81 | 86.02 | GRACE | 2026-10-07 |
| Lance: Unified Multimodal Modeling by Multi-Task Synergy | ✓ Link | 85.11 | | Lance (3B) | 2026-05-18 |
| SALT: Self-Consistent Distribution Matching with Cache-Aware Training for Few-Step Video Generation | ✓ Link | 85.08 | 85.96 | Salt + Causal Forcing | 2026-09-08 |
| LongLive-2.0: An NVFP4 Parallel Infrastructure for Long Video Generation | ✓ Link | 85.06 | | LongLive-2.0 (5B, BF16, 4 steps) | 2026-05-18 |
| Astrolabe: Steering Forward-Process Reinforcement Learning for Distilled Autoregressive Video Models | ✓ Link | 84.93 | | LongLive (Astrolabe) | 2026-03-17 |
| LongLive: Real-time Interactive Long Video Generation | ✓ Link | 84.87 | 86.97 | LongLive | 2025-09-26 |
| ViRDM: Taming Representation Distribution Matching for Few-Step Causal Video Generation | ✓ Link | 84.87 | 85.82 | ViRDM | 2026-09-24 |
| Transition Matching Distillation for Fast Video Generation | | 84.68 | | TMD-N2H5 (TMD) | 2026-01-14 |
| RepForcing: Representation-Space Initialization for Autoregressive Video Distillation | ✓ Link | 84.66 | | RepForcing (chunk-wise, 4-step) | 2026-10-14 |
| Bernini: Latent Semantic Planning for Video Diffusion | | 84.64 | | Bernini | 2026-05-21 |
| EcoVideo: Entropy-Orchestrated Video Generation Paradigm in Cloud-Edge Dynamics | ✓ Link | 84.6 | | EcoVideo | 2026-06-29 |
| Light Forcing: Accelerating Autoregressive Video Diffusion via Sparse Attention | ✓ Link | 84.5 | | Light Forcing | 2026-02-04 |
| Streaming Autoregressive Video Generation via Diagonal Distillation | ✓ Link | 84.48 | | DiagDistill | 2026-03-10 |
| Cycle-World: Mitigating Error Accumulation in Long-term Video World Models via Reverse-Prediction Cycle Consistency | | 84.36 | 86.14 | Cycle-World | 2026-07-13 |
| Glance Forcing: Making Bidirectional Diffusion Autoregressive at a Glance | ✓ Link | 84.33 | 85.01 | Glance Forcing (25k) | 2026-10-14 |
| HiStream: Efficient High-Resolution Video Generation via Redundancy-Eliminated Streaming | ✓ Link | 84.2 | | HiStream | 2025-12-24 |
| Causal Forcing++: Scalable Few-Step Autoregressive Diffusion Distillation for Real-Time Interactive Video Generation | ✓ Link | 84.14 | 84.89 | Causal Forcing++ (2-step) | 2026-05-14 |
| Nexus-Vid: Efficient Frequency Bridging with Homogeneous Latent Space for Video Unified Models | ✓ Link | 84.12 | 85.03 | Lumos-Nexus | 2026-09-08 |
| Causal Forcing: Autoregressive Diffusion Distillation Done Right for High-Quality Real-Time Interactive Video Generation | ✓ Link | 84.04 | | Causal Forcing | 2026-02-02 |
| MHLA: Restoring Expressivity of Linear Attention via Token-Level Multi-Head | ✓ Link | 83.82 | | Wan-MHLA-H (Wan2.1-1.3B) | 2026-01-12 |
| In-Flight KV Cache with Clean Anchors for Faster Autoregressive Video Diffusion | | 83.8 | 85.9 | FlashForward | 2026-09-26 |
| ReHyAt: Recurrent Hybrid Attention for Video Diffusion Transformers | | 83.79 | | Wan2.1 1.3B + ReHyAt (15×Tc=3, To=1) | 2026-01-07 |
| Motif-Video 2B: Technical Report | | 83.76 | | Motif-Video 2B | 2026-04-14 |
| OSP-Next: Efficient High-Quality Video Generation with Sparse Sequence Parallelism, HiF8 Quantization, and Reinforcement Learning | ✓ Link | 83.73 | 85.19 | OSP-Next | 2026-05-27 |
| SANA-Video: Efficient Video Generation with Block Linear Diffusion Transformer | ✓ Link | 83.71 | 84.35 | SANA-Video | 2025-09-29 |
| Content-Aware Dynamic Patchification for Efficient Video Diffusion | ✓ Link | 83.56 | 84.79 | DynaPatch (20% token reduction) | 2026-06-03 |
| SSBP: Stage-Specialized Block Pruning for Video Diffusion Models | | 83.52 | 85.02 | SSBP-Wan2.1-1.3B | 2026-09-08 |
| Surprise Forcing: What to Remember, When to Skip in Long Video Generation | | 83.4 | | Surprise Forcing | 2026-07-20 |
| PyramidalWan: On Making Pretrained Video Model Pyramidal for Efficient Inference | | 83.33 | | Wan-DMD (4 steps) | 2026-01-08 |
| ForcingBooster: Adaptive Patch-wise Denoising for Fast Autoregressive Video Diffusion | ✓ Link | 83.32 | 84.73 | Wan2.1 14B + ForcingBooster | 2026-10-08 |
| PixelUMM: Encoder-Free Unified Image and Video Understanding and Generation | ✓ Link | 83.24 | 84.1 | PixelUMM 8B MoT | 2026-09-29 |
| Speculative Forcing for Efficient Autoregressive Video Generation | ✓ Link | 83.21 | 83.54 | Speculative Forcing (learned verifier) | 2026-10-15 |
| FastSTAR: Spatiotemporal Token Pruning for Efficient Autoregressive Video Synthesis | | 83.18 | | FastSTAR (20%, 30%, 40%, 70%) | 2026-03-07 |
| VINO: A Unified Visual Generator with Interleaved OmniModal Context | ✓ Link | 83.17 | | VINO (with LLM rewriter) | 2026-01-05 |
| OmniWeaving: Towards Unified Video Generation with Free-form Composition and Reasoning | ✓ Link | 83.1 | 84.37 | OmniWeaving | 2026-03-25 |
| Motion-Aware Caching for Efficient Autoregressive Video Generation | ✓ Link | 82.84 | | MotionCache (on SkyReels-V2-1.3B) | 2026-05-03 |
| Mask Forcing: Improving Autoregressive Video Diffusion Distillation via Dual-Noise Masking Rollout | ✓ Link | 82.76 | | Causal Forcing + Mask Forcing (chunk-wise) | 2026-09-08 |
| LinCa: Accelerating Diffusion Models via Learnable Decomposed Feature Caching | ✓ Link | 82.56 | | LinCa (Wan2.1-1.3B, N=6) | 2026-08-18 |
| Cross-Resolution Distribution Matching for Diffusion Distillation | | 82.51 | 84.37 | RMD (Wan2.1-T2V-14B) | 2026-03-06 |
| FIRM-Video: Check Before You Score for Reliable Text-to-Video Reward Modeling | ✓ Link | 82.36 | 84.91 | FIRM-Video-8B (Qwen3-VL-8B), Best-of-8 on Wan2.1-T2V-1.3B | 2026-08-22 |
| Routing Policy Optimization for Online Diffusion Reinforcement Learning | | 82.0 | | RPO (Wan2.2-TI2V-5B) | 2026-10-15 |
| DisRM: Reward Modeling as Discriminative Prediction | | 81.38 | | Ours-DPO (extra data) | 2026-09-08 |
| TimeRipples: Accelerating vDiTs by Understanding the Spatio-Temporal Correlations in Latent Space | | 81.16 | | TIMERIPPLE (CogVideoX1.5-5B-T2V, 80% reuse) | 2025-11-15 |
| DreamWorld: Unified World Modeling in Video Generation | ✓ Link | 80.97 | 83.49 | DreamWorld | 2026-02-28 |
| D2Cache: Second-Order Delta Caching for Higher Video Diffusion Acceleration | ✓ Link | 80.74 | 84.36 | D2Cache (Wan2.1, slow) | 2026-06-03 |
| VDE: Training-Free Accelerating Rectified Flow Model via Velocity Decomposition and Estimation | ✓ Link | 80.43 | | VDE-fast (Wan2.1) | 2026-05-22 |
| ResilPhase: Plug-and-Play Phase Mapping and Noise-Resilient Macro-Trajectory Extrapolation for Diffusion Acceleration | ✓ Link | 80.42 | | ResilPhase | 2026-06-25 |
| Accelerating Text-to-Video Generation with Calibrated Sparse Attention | | 80.4 | 82.3 | CalibAtt (Wan 2.1 14B, 480p) | 2026-03-05 |
| ResCa: Residual Caching for Diffusion Transformers Acceleration | ✓ Link | 79.98 | | ResCa-IE (N=6, K=32) | 2026-06-03 |
| PhysRVG: Physics-Aware Unified Reinforcement Learning for Video Generative Models | | 78.89 | | PhysRVG | 2026-01-16 |
| ProPhy: Progressive Physical Alignment for Dynamic World Simulation | ✓ Link | | 81.0 | CogVideoX with ProPhy | 2025-12-05 |
| FastSTAR: Spatiotemporal Token Pruning for Efficient Autoregressive Video Synthesis | | | 83.71 | FastSTAR (20%, 30%, 50%, 60%) | 2026-03-07 |
| MosaicMem: Hybrid Spatial Memory for Controllable Video World Models | | | 81.11 | MosaicMem (full) | 2026-03-17 |
| Reward Lightning: Fast Video Generation via Homologous Preference Distillation | | | 85.83 | Reward Lightning (HPD, 4-NFEs) | 2026-07-04 |
| Mask Forcing: Improving Autoregressive Video Diffusion Distillation via Dual-Noise Masking Rollout | ✓ Link | | 83.71 | LongLive + Mask Forcing (chunk-wise) | 2026-09-08 |
| PhysAlign: Learning Physical Priors for Dynamical Event-Driven Video Generation via Representation Alignment | | | 79.05 | PhysAlign (extra data) | 2026-09-08 |
| Simplifying and Improving On-Policy Distillation for Few-Step Diffusion | | | 85.72 | SMD (Wan2.1-T2V-14B, 4-NFE) | 2026-10-13 |
| RepForcing: Representation-Space Initialization for Autoregressive Video Distillation | ✓ Link | | 85.67 | RepForcing (frame-wise, 4-step) | 2026-10-14 |
| SpecKCast: Spectral Kalman Forecasting for Diffusion Acceleration | | | 80.86 | SpecKCast (Wan2.1-14B, approximately 14 NFE) | 2026-10-19 |