pith. sign in

arxiv: 2605.30349 · v1 · pith:7O4SG57Snew · submitted 2026-05-28 · 💻 cs.CV

AdaState: Self-Evolving Anchors for Streaming Video Generation

Yusuf Dalva , Pinar Yanardag This is my paper

Pith reviewed 2026-06-29 08:21 UTC · model grok-4.3

classification 💻 cs.CV
keywords adaptive statestreaming video generationautoregressive video diffusionself-evolving anchorsvideo dynamicsKV cacherelative time
0
0 comments X

The pith

Replacing the static first-frame anchor with a self-evolving hidden state allows autoregressive video models to produce richer motion and natural scene changes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Autoregressive video diffusion models generate frames one chunk at a time while keeping the first frame's key-value representation fixed in the attention cache. This fixed anchor draws excessive attention, which dampens motion, camera movement, and scene evolution in favor of static consistency. The paper replaces that anchor with an adaptive state: a hidden latent that the model denoises at every step alongside the visible content but never renders as output. By making the state evolve through attention to both the prior state and current content, and by treating time as relative so every step uses the same positional structure, the process turns denoising into a recurrence carried only by the existing KV cache. If the claim holds, generated videos gain substantially better dynamics without extra losses, modules, or supervision.

Core claim

The paper claims that the adaptive state, a hidden latent denoised but never rendered, generates its own scene reference at each chunk by attending to the previous state and current content. Because the formulation makes every generation step see identical relative positional structure, the state transition becomes the same recurrence at every step, carried by the KV cache and trained solely with the standard diffusion objective.

What carries the argument

The adaptive state, a hidden latent that the model denoises alongside visible content but never renders, serving as an evolving scene anchor through recurrence in the KV cache.

If this is right

  • Generated videos exhibit richer motion and natural scene progression instead of being locked to the initial viewpoint.
  • Every generation step uses the same positional structure, so the state transition remains identical regardless of how far generation has progressed.
  • The recurrence is carried entirely by the KV cache and standard diffusion training, requiring no external module.
  • Scene references evolve at each step by attending to both the prior state and current content rather than a frozen first frame.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same recurrence pattern could be tested in autoregressive generation of other modalities where a fixed initial token limits later variation.
  • Longer sequences might maintain coherence longer if the state continues to evolve without accumulating drift toward the opening frame.
  • The approach could be combined with explicit camera or motion controls to see whether the evolving anchor amplifies or interferes with those signals.

Load-bearing premise

The model can learn an effective denoising transition for the hidden state using only the standard diffusion objective and the existing KV cache, without any auxiliary loss or external supervision on the state itself.

What would settle it

Side-by-side generation of the same prompts with and without the adaptive state, scored on motion magnitude and scene-change metrics, would show no measurable increase in dynamics if the central claim is false.

Figures

Figures reproduced from arXiv: 2605.30349 by Pinar Yanardag, Yusuf Dalva.

Figure 1
Figure 1. Figure 1: AdaState. Colored markers highlight the scene at each timestamp; dashed lines trace their progression (red: baselines, teal: AdaState). Top, t=30s (6× training horizon): Infinity-RoPE’s static anchor cannot adapt to the evolving scene, forcing the model to realize all implied content, schools of fish, sea turtles, within the initial layout, producing hallucinated duplications by t=30. AdaState’s markers dr… view at source ↗
Figure 2
Figure 2. Figure 2: The anchor-recency structure of streaming video attention. (a) Off-diagonal attention in Self-Forcing across chunk depths. The anchor at position 0 (squares) and the freshest chunk frame (triangles) consistently dominate; remaining positions receive roughly uniform mass. (b) 5-second generation on the same prompt. Without a persistent reference, coherence degrades over time. A static reference preserves id… view at source ↗
Figure 3
Figure 3. Figure 3: AdaState Framework. The adaptive state (green) is denoised alongside content at each chunk but never rendered. Its clean KV is written to position 0 and carried to subsequent chunks via the state recurrence (green dashed). Decoded state previews (middle, green-bordered, matching the state tokens) visualize the hidden state in image space; the zoom insets reveal the model’s denoising errors, which the archi… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison across anchor categories. Top block: 12-second generation sampled at 3-second intervals. Bottom block: 30-second generation sampled at 7.5-second intervals. Each block pairs AdaState against one exemplar per baseline category (no reference, EMA reference, static reference); the six exemplars across the two blocks cover all baseline groups. Methods without a persistent anchor accumula… view at source ↗
Figure 5
Figure 5. Figure 5: Subject consistency vs. dynamic de￾gree at 30 seconds. The dynamics distribution becomes bimodal: most baselines collapse to the left as motion stops, while AdaState alone occupies the shaded upper-right region where high dynam￾ics and high consistency coexist. To confirm the perceptual ranking, we conduct a user study with 40 raters with Prolific plat￾form1 . Each rater views videos from AdaState and four… view at source ↗
Figure 6
Figure 6. Figure 6: User study (5-point Likert, N=40 raters). Methods are ordered by coherent￾progression score. AdaState gets the highest rat￾ings on both coherent progression and prompt fol￾lowing. The horizon weighting ablation motivates our two-regime training. At α=2, dynamics and total score peak, the right choice within the training horizon. At α=4, the optimizer con￾centrates more gradient on late frames, trading with… view at source ↗
Figure 7
Figure 7. Figure 7: User study evaluation interface. Each rater views a video generated from a given prompt and scores it on two dimensions using a 5-point Likert scale. Method identity is hidden; video order is randomized. Identifying information has been redacted for anonymity. C Evaluation Details Detailed Quantitative Results. Tables 4 and 5 report the per-dimension VBench scores at 5 and 30 seconds, extending [PITH_FULL… view at source ↗
read the original abstract

Autoregressive video diffusion models generate streaming video by producing frames sequentially, conditioning each chunk on previously generated content. These models are structurally anchored to the first frame: its key-value representation occupies a privileged position in the attention cache and serves as the primary scene reference throughout generation. As the cleanest and most error-free position in the cache, this anchor draws disproportionate attention, suppressing video dynamics, and locking scene composition to the initial viewpoint even as the scene naturally evolves. The result is a temporally shallow video in which motion, camera movement, and scene progression are dampened in favor of static consistency. To address this, we replace the static anchor with an adaptive state, a hidden latent that the model denoises alongside content at every chunk but never renders. Rather than referencing a frozen first frame, the model generates its own scene anchor at each step by attending to both the previous state and the current content, producing a reference that evolves with the generated content. Unlike standard video generation, which encodes an absolute notion of time, our formulation treats time as relative: every generation step sees the same positional structure regardless of how far generation has progressed, and the state transition is identical at every chunk. Together, these properties introduce a recurrence into the generation process, where denoising serves as the transition function, and the KV cache serves as the carrier, requiring no external module. Experiments demonstrate that the adaptive state substantially improves video dynamics, enabling richer motion and natural scene progression within generated videos.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes AdaState for autoregressive video diffusion models in streaming generation. It identifies the fixed first-frame KV anchor as causing suppressed dynamics and static scene locking, and replaces it with a hidden adaptive state that is denoised at each chunk (but never rendered) by attending to the prior state and current content inside the existing KV cache. The approach treats time as relative with identical positional structure at every step, turning denoising into a recurrence transition carried by the KV cache without external modules. The abstract asserts that experiments show the adaptive state yields richer motion and natural scene progression.

Significance. If the empirical claims hold, the method would offer a lightweight architectural change that introduces recurrence into streaming video diffusion without auxiliary losses or new modules, potentially addressing a structural limitation in temporal dynamics. The absence of any reported metrics, baselines, ablations, or state analysis, however, prevents assessment of whether the claimed gains materialize or whether the state carries meaningful scene-evolution information.

major comments (2)
  1. [Abstract] Abstract: the central claim that 'experiments demonstrate that the adaptive state substantially improves video dynamics' is unsupported by any quantitative metrics, baseline comparisons, dataset details, ablation results, or state-trajectory analysis, rendering the empirical contribution unevaluable.
  2. [Abstract] Abstract (method description): the state transition is trained solely via the standard diffusion objective on visible frames with no auxiliary loss, reconstruction target, or consistency regularizer on the hidden state itself; no evidence is supplied that the state avoids collapse to a constant representation or carries scene-evolution information, leaving the recurrence mechanism and claimed motion gains unverified.
minor comments (1)
  1. [Abstract] The description of 'time as relative' and identical positional structure across chunks would benefit from an explicit diagram or pseudocode showing the KV-cache layout and attention pattern at successive steps.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and for identifying the need to strengthen the empirical grounding of the abstract. We respond to each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that 'experiments demonstrate that the adaptive state substantially improves video dynamics' is unsupported by any quantitative metrics, baseline comparisons, dataset details, ablation results, or state-trajectory analysis, rendering the empirical contribution unevaluable.

    Authors: The referee is correct that the current abstract asserts empirical gains without accompanying quantitative details, baselines, or analysis. We will revise the abstract to remove or qualify the unsupported claim and will add a concise summary of key metrics, datasets, and references to the experimental sections in the revised version. revision: yes

  2. Referee: [Abstract] Abstract (method description): the state transition is trained solely via the standard diffusion objective on visible frames with no auxiliary loss, reconstruction target, or consistency regularizer on the hidden state itself; no evidence is supplied that the state avoids collapse to a constant representation or carries scene-evolution information, leaving the recurrence mechanism and claimed motion gains unverified.

    Authors: The description is accurate: training uses only the standard diffusion loss with no auxiliary terms on the hidden state. We acknowledge that this leaves open the possibility of collapse and that no direct verification is currently provided. In revision we will add state-trajectory visualizations and simple quantitative checks (e.g., state variance across chunks) to demonstrate that the hidden state evolves meaningfully rather than collapsing. revision: yes

Circularity Check

0 steps flagged

No circularity: architectural proposal validated empirically

full rationale

The paper proposes an architectural replacement of the static first-frame KV anchor with a hidden adaptive state that is denoised jointly but never rendered. The claimed benefit in video dynamics is presented solely as an experimental outcome from applying the standard diffusion objective. No equations, derivations, fitted parameters renamed as predictions, self-citations, or ansatzes appear in the provided text. The construction is self-contained as a modeling change whose effectiveness is asserted via results rather than reduced to its inputs by definition.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the introduction of one new invented entity (the adaptive state) and one domain assumption about diffusion models being able to learn its transition without extra supervision. No free parameters are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Diffusion models can jointly denoise content and an auxiliary hidden state using the standard noise-prediction objective.
    Invoked when the paper states that the model denoises the state alongside content at every chunk.
invented entities (1)
  • adaptive state (hidden latent) no independent evidence
    purpose: Evolving scene reference that replaces the static first-frame anchor and is never rendered to the viewer.
    The paper introduces this entity to solve the static-anchor problem; no independent evidence (e.g., predicted observable quantity) is provided in the abstract.

pith-pipeline@v0.9.1-grok · 5791 in / 1357 out tokens · 24640 ms · 2026-06-29T08:21:21.822826+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 19 canonical work pages · 11 internal anchors

  1. [1]

    Titans: Learning to Memorize at Test Time

    Ali Behrouz, Peilin Zhong, and Vahab Mirrokni. Titans: Learning to memorize at test time. arXiv preprint arXiv:2501.00663, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  2. [2]

    Mode seeking meets mean seeking for fast long video generation.arXiv preprint arXiv:2602.24289, 2026

    Shengqu Cai, Weili Nie, Chao Liu, Julius Berner, Lvmin Zhang, Nanye Ma, Hansheng Chen, Maneesh Agrawala, Leonidas Guibas, Gordon Wetzstein, et al. Mode seeking meets mean seeking for fast long video generation.arXiv preprint arXiv:2602.24289, 2026

    work page arXiv 2026

  3. [3]

    Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling

    Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, and Yoshua Bengio. Empirical evaluation of gated recurrent neural networks on sequence modeling.arXiv preprint arXiv:1412.3555, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  4. [4]

    Self-forcing++: Towards minute-scale high-quality video generation

    Justin Cui, Jie Wu, Ming Li, Tao Yang, Xiaojie Li, Rui Wang, Andrew Bai, Yuanhao Ban, and Cho-Jui Hsieh. Self-forcing++: Towards minute-scale high-quality video generation. In The Fourteenth International Conference on Learning Representations, 2026. URL https: //openreview.net/forum?id=DzvPiqh23f

    2026

  5. [5]

    Scaling rectified flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, Dustin Podell, Tim Dockhorn, Zion English, and Robin Rombach. Scaling rectified flow transformers for high-resolution image synthesis. InForty-first International Conference on Machine Learning, 2024. URL https:...

    2024

  6. [6]

    Bartoldson, Bhavya Kailkhura, Abhinav Bhatele, and Tom Goldstein

    Jonas Geiping, Sean Michael McLeish, Neel Jain, John Kirchenbauer, Siddharth Singh, Brian R. Bartoldson, Bhavya Kailkhura, Abhinav Bhatele, and Tom Goldstein. Scaling up test-time compute with latent reasoning: A recurrent depth approach. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2026. URL https://openreview. net/forum...

    2026

  7. [7]

    Think before you speak: Training language models with pause tokens

    Sachin Goyal, Ziwei Ji, Ankit Singh Rawat, Aditya Krishna Menon, Sanjiv Kumar, and Vaishnavh Nagarajan. Think before you speak: Training language models with pause tokens. InThe Twelfth International Conference on Learning Representations, 2024. URL https: //openreview.net/forum?id=ph04CRkPdC

    2024

  8. [8]

    Mamba: Linear-Time Sequence Modeling with Selective State Spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  9. [9]

    Training large language models to reason in a continuous latent space

    Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason E Weston, and Yuandong Tian. Training large language models to reason in a continuous latent space. InSecond Conference on Language Modeling, 2025. URL https://openreview.net/forum?id= Itxz7S4Ip3

    2025

  10. [10]

    Reasoning with latent tokens in diffusion language models.arXiv preprint arXiv:2602.03769, 2026

    Andre He, Sean Welleck, and Daniel Fried. Reasoning with latent tokens in diffusion language models.arXiv preprint arXiv:2602.03769, 2026

    work page arXiv 2026

  11. [11]

    Long short-term memory.Neural computation, 9(8): 1735–1780, 1997

    Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory.Neural computation, 9(8): 1735–1780, 1997. 10

    1997

  12. [12]

    Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

    Xun Huang, Zhengqi Li, Guande He, Mingyuan Zhou, and Eli Shechtman. Self forcing: Bridging the train-test gap in autoregressive video diffusion.arXiv preprint arXiv:2506.08009, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  13. [13]

    VBench++: Comprehensive and versatile benchmark suite for video generative models.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    Ziqi Huang, Fan Zhang, Xiaojie Xu, Yinan He, Jiashuo Yu, Ziyue Dong, Qianli Ma, Nattapol Chanpaisit, Chenyang Si, Yuming Jiang, Yaohui Wang, Xinyuan Chen, Ying-Cong Chen, Limin Wang, Dahua Lin, Yu Qiao, and Ziwei Liu. VBench++: Comprehensive and versatile benchmark suite for video generative models.IEEE Transactions on Pattern Analysis and Machine Intelli...

    work page doi:10.1109/tpami.2025.3633890 2025

  14. [14]

    Memrope: Training-free infinite video generation via evolving memory tokens.arXiv preprint arXiv:2603.12513, 2026

    Youngrae Kim, Qixin Hu, C-C Jay Kuo, and Peter A Beerel. Memrope: Training-free infinite video generation via evolving memory tokens.arXiv preprint arXiv:2603.12513, 2026

    work page arXiv 2026

  15. [15]

    Rolling Sink: Bridging Limited-Horizon Training and Open-Ended Testing in Autoregressive Video Diffusion

    Haodong Li, Shaoteng Liu, Zhe Lin, and Manmohan Chandraker. Rolling sink: Bridging limited-horizon training and open-ended testing in autoregressive video diffusion.arXiv preprint arXiv:2602.07775, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  16. [16]

    Rolling Forcing: Autoregressive Long Video Diffusion in Real Time

    Kunhao Liu, Wenbo Hu, Jiale Xu, Ying Shan, and Shijian Lu. Rolling forcing: Autoregressive long video diffusion in real time.arXiv preprint arXiv:2509.25161, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  17. [17]

    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, et al. Reward forcing: Efficient streaming video generation with rewarded distribution matching distillation.arXiv preprint arXiv:2512.04678, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  18. [18]

    Packforcing: Short video training suffices for long video sampling and long context inference.arXiv preprint arXiv:2603.25730, 2026

    Xiaofeng Mao, Shaohao Rui, Kaining Ying, Bo Zheng, Chuanhao Li, Mingmin Chi, and Kaipeng Zhang. Packforcing: Short video training suffices for long video sampling and long context inference.arXiv preprint arXiv:2603.25730, 2026

    work page arXiv 2026

  19. [19]

    Show your work: Scratchpads for intermediate computation with language models

    Maxwell Nye, Anders Johan Andreassen, Guy Gur-Ari, Henryk Michalewski, Jacob Austin, David Bieber, David Dohan, Aitor Lewkowycz, Maarten Bosma, David Luan, Charles Sutton, and Augustus Odena. Show your work: Scratchpads for intermediate computation with language models. InDeep Learning for Code Workshop, 2022. URL https://openreview.net/ forum?id=HBlx2idbkbq

    2022

  20. [20]

    Jacob Pfau, William Merrill, and Samuel R. Bowman. Let’s think dot by dot: Hidden computa- tion in transformer language models. InFirst Conference on Language Modeling, 2024. URL https://openreview.net/forum?id=NikbrdtYvG

    2024

  21. [21]

    Movie Gen: A Cast of Media Foundation Models

    Adam Polyak, Amit Zohar, Andrew Brown, Andros Tjandra, Animesh Sinha, Ann Lee, Apoorv Vyas, Bowen Shi, Chih-Yao Ma, Ching-Yao Chuang, et al. Movie gen: A cast of media foundation models.arXiv preprint arXiv:2410.13720, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  22. [22]

    Learning internal representa- tions by error propagation

    David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. Learning internal representa- tions by error propagation. Technical report, 1985

    1985

  23. [23]

    Codi: Com- pressing chain-of-thought into continuous space via self-distillation

    Zhenyi Shen, Hanqi Yan, Linhai Zhang, Zhanghao Hu, Yali Du, and Yulan He. Codi: Com- pressing chain-of-thought into continuous space via self-distillation. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 677–693, 2025

    2025

  24. [24]

    Learning to (learn at test time): RNNs with expressive hidden states

    Yu Sun, Xinhao Li, Karan Dalal, Jiarui Xu, Arjun Vikram, Genghan Zhang, Yann Dubois, Xinlei Chen, Xiaolong Wang, Sanmi Koyejo, Tatsunori Hashimoto, and Carlos Guestrin. Learning to (learn at test time): RNNs with expressive hidden states. InForty-second Inter- national Conference on Machine Learning, 2025. URLhttps://openreview.net/forum? id=wXfuOj9C7L

    2025

  25. [25]

    Wan: Open and Advanced Large-Scale Video Generative Models

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianxiao Yang, et al. Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  26. [26]

    Efficient streaming language models with attention sinks

    Guangxuan Xiao, Yuandong Tian, Beidi Chen, Song Han, and Mike Lewis. Efficient streaming language models with attention sinks. InThe Twelfth International Conference on Learning Representations, 2024. URLhttps://openreview.net/forum?id=NG7sS51zVF. 11

    2024

  27. [27]

    Visionreward: Fine-grained multi-dimensional human preference learning for image and video generation

    Jiazheng Xu, Yu Huang, Jiale Cheng, Yuanming Yang, Jiajun Xu, Yuan Wang, Wenbo Duan, Shen Yang, Qunlin Jin, Shurun Li, et al. Visionreward: Fine-grained multi-dimensional human preference learning for image and video generation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 11269–11277, 2026

    2026

  28. [28]

    Longlive: Real-time interactive long video generation

    Shuai Yang, Wei Huang, Ruihang Chu, Yicheng Xiao, Yuyang Zhao, Xianbang Wang, Muyang Li, Enze Xie, Ying-Cong Chen, Yao Lu, Song Han, and Yukang Chen. Longlive: Real-time interactive long video generation. InThe Fourteenth International Conference on Learning Representations, 2026. URLhttps://openreview.net/forum?id=nCAODkpsPJ

    2026

  29. [29]

    Infinity-rope: Action-controllable infinite video generation emerges from autoregressive self- rollout.arXiv preprint arXiv:2511.20649, 2025

    Hidir Yesiltepe, Tuna Han Salih Meral, Adil Kaan Akan, Kaan Oktay, and Pinar Yanardag. Infinity-rope: Action-controllable infinite video generation emerges from autoregressive self- rollout.arXiv preprint arXiv:2511.20649, 2025

    work page arXiv 2025

  30. [30]

    Deep forcing: Training-free long video generation with deep sink and participative compression

    Jung Yi, Wooseok Jang, Paul Hyunbin Cho, Jisu Nam, Heeji Yoon, and Seungryong Kim. Deep forcing: Training-free long video generation with deep sink and participative compression. arXiv preprint arXiv:2512.05081, 2025

    work page arXiv 2025

  31. [31]

    One-step diffusion with distribution matching distillation

    Tianwei Yin, Michaël Gharbi, Richard Zhang, Eli Shechtman, Fredo Durand, William T Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6613–6623, 2024

    2024

  32. [32]

    From slow bidirectional to fast autoregressive video diffusion models

    Tianwei Yin, Qiang Zhang, Richard Zhang, William T Freeman, Fredo Durand, Eli Shechtman, and Xun Huang. From slow bidirectional to fast autoregressive video diffusion models. In CVPR, 2025

    2025

  33. [33]

    Videossm: Autoregressive long video generation with hybrid state-space memory.arXiv preprint arXiv:2512.04519, 2025

    Yifei Yu, Xiaoshan Wu, Xinting Hu, Tao Hu, Yangtian Sun, Xiaoyang Lyu, Bo Wang, Lin Ma, Yuewen Ma, Zhongrui Wang, et al. Videossm: Autoregressive long video generation with hybrid state-space memory.arXiv preprint arXiv:2512.04519, 2025

    work page arXiv 2025

  34. [34]

    Test-Time Training Done Right

    Tianyuan Zhang, Sai Bi, Yicong Hong, Kai Zhang, Fujun Luan, Songlin Yang, Kalyan Sunkavalli, William T Freeman, and Hao Tan. Test-time training done right.arXiv preprint arXiv:2505.23884, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  35. [35]

    Causal Forcing: Autoregressive Diffusion Distillation Done Right for High-Quality Real-Time Interactive Video Generation

    Hongzhou Zhu, Min Zhao, Guande He, Hang Su, Chongxuan Li, and Jun Zhu. Causal forcing: Autoregressive diffusion distillation done right for high-quality real-time interactive video generation.arXiv preprint arXiv:2602.02214, 2026. 12 Table of Contents A Implementation Details 14 B User Study Details 14 C Evaluation Details 15 D Supplementary Video Results...

    work page internal anchor Pith review Pith/arXiv arXiv 2026