World Models' Last Exam in Physics

World Models' Last Exam in Physics benchmarks video models via 40 measurement-based physics tasks, finding the best model scores 57.76/100 with widespread inconsistencies.

Mingju Gao, Qingle Liu, Yuzhao Peng, Xinjie Lin and 7 more

Published Oct 6, 2026 · ▲ 2 on Hugging Face

ALIVE: Interaction-Aligned Object Insertion for First-Frame-Guided Video Editing

ALIVE inserts objects that interact with video contents via first-frame editing and interaction guidance, outperforming baselines on interaction and insertion benchmarks.

Zhenghong Zhou, Zhe Lin, Jiebo Luo, Yuqian Zhou

Published Oct 6, 2026 · ▲ 3 on Hugging Face · Code

S2PD: Serial-to-Parallel Diffusion for Physically and Logically Consistent Video Generation

S2PD switches from autoregressive to parallel diffusion during denoising to enforce physical and logical consistency with faster sampling than fully serial methods.

Jeffrey Hu, Daniel Olmeda Reino, Ayush Tewari

Published Oct 5, 2026 · 0 citations · ▲ 1 on Hugging Face · Code ★ 4

HLA-WM: Hybrid Linear Attention for Long-Horizon Video World Models

HLA-WM combines geometry-guided retrieval with recurrent linear attention to fix long-range forgetting in video world models, improving 60-second consistency metrics by up to 28.5% with 12× lower memory and no retraining.

Zhuokun Chen, Feng Chen, Xi Lin, Xiyu Wu and 3 more

Published Oct 5, 2026 · ▲ 6 on Hugging Face · Code ★ 2

DuoMatching: Joint-Marginal Distribution Matching for Few-Step Video Generation

DuoMatching improves few-step video generation by jointly matching frame distributions and adding frame-level supervision via an image teacher, boosting visual quality and semantic alignment over 80%.

Jiahao Zhan, Yan Wang, Yongrui Ma, Qunliang Xing and 4 more

Published Oct 2, 2026 · 0 citations · ▲ 17 on Hugging Face · Code ★ 4

In-Distribution Forcing for Long Video Generation at Test Time

In-Distribution Forcing prevents out-of-distribution key-value drift via self-caching to extend short video models to minute-scale generation.

Jeongwoo Shin, Youngyoon Choi, Sangwoo Jo, Hyunmog Kim and 4 more

Published Oct 2, 2026 · 0 citations · ▲ 29 on Hugging Face · Code ★ 3

ROWBench: Do Video Models Render What the Program Specifies?

PROWBench evaluates video models' fidelity to program-specified world events via replayable world records and VLM-based logic-render and interaction metrics.

Zheng-Hui Huang, Guixu Lin, Yu-Ju Tsai, Jian-Kai Zhu and 5 more

Published Oct 1, 2026 · 0 citations · ▲ 70 on Hugging Face · Code ★ 31

World Observer: Joint Actor-Observer Generation for Persistent World Modeling

World Observer jointly generates actor and panoramic observer views to continuously model out-of-view dynamics via shared geometric warping and observer sinks.

Hyunwook Choi, Dahyun Chung, Hyunsung Kim, Siyoon Jin and 3 more

Published Oct 1, 2026 · 0 citations · ▲ 85 on Hugging Face · Code ★ 30

Spatial Memory Intelligence: Endowing World Models with Understanding-Driven Long-Term Memory

Spatial Memory Intelligence introduces understanding-driven atomic operations for spatial-memory management in long-video world models, improving sparsity, stability, and spatial consistency.

Ying Yang, Guiyu Zhang, Lianghua Huang, Chang Nie and 4 more

Published Oct 1, 2026 · 0 citations · ▲ 49 on Hugging Face · Code ★ 19

Ego2Act: Evaluating Goal-Directed Manipulation in Egocentric Video Generation

Ego2Act evaluates egocentric video generation on multi-step goal-directed manipulation, showing models skip steps and fail at fine-grained physical dynamics.

Patrick Amadeus Irawan, Iskandar Muda Rizky Parlambang, Rava Maulana, Qinrong Cui and 9 more

Published Oct 1, 2026 · 0 citations · ▲ 32 on Hugging Face · Code ★ 3

FrameMorrow: Future-guided Frame Selection with Prospective Tokens for Long-Horizon Video Generation

FrameMorrow guides historical frame selection via prospective tokens representing future needs, improving consistency and quality across diverse long-horizon video generators.

Bo Yin, Xiaobin Hu, Jiaqi Zhao, Shuicheng Yan

Published Sep 30, 2026 · 0 citations · ▲ 92 on Hugging Face · Code ★ 26

Video Generation Models: A Survey of Post-Training and Alignment

This survey reviews post-training and alignment strategies for video generation, framing them as implicit or explicit alignment across four methodological categories to improve controllability and reliability.

Chaoyu Li, Xiaoyi Gu, Yogesh Kulkarni, Eun Woo Im and 9 more

Published Sep 30, 2026 · ▲ 60 on Hugging Face · Code ★ 208

Rollout-Marginal Distillation for Long-Horizon Autoregressive Video Generation

Rollout-Marginal Distillation scores autoregressive video chunks independently against a chunk teacher to prevent error accumulation, then applies video-level distillation to restore temporal coherence.

Chenjian Gao, Zhihao Hu, Jianqi Ma, Jun Zhang and 2 more

Published Sep 29, 2026 · 0 citations · ▲ 18 on Hugging Face · Code ★ 2

QuantWM: Temporally Consistent 2-Bit KV Cache Quantization for Video World Models

QuantWM is a training-free 2-bit KV cache quantization framework for video world models that preserves attention logits and token selection to eliminate temporal flickering while achieving up to 6.20x memory compression.

Jiaqi Zhao, Xiaobin Hu, Bo Yin, Junpeng Jiang and 2 more

Published Sep 22, 2026 · 0 citations · ▲ 13 on Hugging Face

Vidu S2: Real-Time Interactive, Editable, and Spatial Video Generation

Vidu S2 enables real-time interactive avatar and editing video generation at 720p with dynamic references and spatial capabilities, outperforming all baselines.

Jintao Zhang, Kai Jiang, Jintao Chen, Xu Wang and 31 more

Published Sep 10, 2026 · 0 citations · ▲ 706 on Hugging Face · Code ★ 501

VGI-Bench: Probing Visual Intelligence in Video Generation Models

VGI-Bench evaluates video generation models via 27 visual reasoning tasks, finding top models achieve only 51% accuracy with limited self-correction.

Xuan He, Cong Wei, Yuhao Cheng, Linrui Ma and 19 more

Published Aug 20, 2026 · 0 citations · ▲ 336 on Hugging Face · Code ★ 14

Kandinsky 5.0: A Family of Foundation Models for Image and Video Generation

Kandinsky 5.0 introduces state-of-the-art 6B image and 2B/19B video generation models with optimized training and inference for high-speed, high-quality synthesis.

Arkhipkin, Vladimir, Korviakov, Vladimir, Gerasimenko, Nikolai, Parkhomenko, Denis and 21 more

Published Nov 19, 2025 · 0 citations · ▲ 236 on Hugging Face · Code ★ 834

LongCat-Video Technical Report

LongCat-Video is a 13.6B DiT video generation model that efficiently produces high-quality minutes-long 720p videos via coarse-to-fine generation, block sparse attention, and multi-reward RLHF.

Meituan LongCat Team, Xunliang Cai, Qilong Huang, Zhuoliang Kang and 7 more

Published Oct 25, 2025 · ▲ 43 on Hugging Face · Code ★ 8,986

MilliVid: Adaptive Latents for Long-Range Consistency in Video Generation

Ishaan Chandratreya, David Charatan, Basile Van Hoorick, Sergey Zakharov and 3 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

Bidirectional Sparse Attention for Faster Video Diffusion Training

Chenlu Zhan, Wen Li, Jun Zhang, chuyu shen and 1 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026