Pith. sign in

REVIEW 4 major objections 6 minor 37 references

Alaya-EVOKE: From Linear-Scaling Supervision to Endless World

T0 review · 4 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Evoke recasts interactive video generation as a bounded recurrent process: persistent scene geometry lives in a camera-indexed world state bank, and a 31.4-second teacher supervision budget transfers long-horizon stability and timed text…

desk verdict A well-built bounded-recurrent world model with a genuinely new external-geometry design, but the memory-consistency evidence is thinner than the architecture claims and the depth/pose calibration on generated rollouts is the load-bearing soft spot. read the letter →

arxiv 2608.13546 v1 pith:BL4QQFWH submitted 2026-08-13 cs.CV

Yuanyang Yin , Gongxuan Wang , Yifan Zhan , Chuanhao Li , Kaipeng Zhang , Feng Zhao This is my paper · ORCID
classification cs.CV
keywords interactiveworldmodelsstatebanklong-horizonsupervisiondistributionmatchingdistillationsparseattentionfew-stepgenerationcameracontrolexternalgeometricmemory
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper aims to remove the two bottlenecks that keep interactive video world models from running for hours: the denoiser context must not grow with session history, and a few-step student must inherit capabilities its teacher never showed it. Evoke's first move is to externalize persistent state: generated frames are unprojected into a camera-indexed world state bank, and each recurrent step reads back only the geometry visible from the current camera pose, so per-chunk context, positional span, and compute stay fixed as the session lengthens. Its second move is to treat the teacher as part of the design: chunk-wise sparse attention makes long-window scoring affordable, per-chunk text conditions put prompt changes inside a single rollout, and a 30-second distribution-matching objective under self-forced rollouts transfers long-horizon photometric stability and timed text control to a three-step student that runs without classifier-free guidance. If the claims hold, interactive generation becomes a bounded recurrence: long sessions cost more calls, not bigger calls, and the reported 2.11 s per 1.5 s chunk at 384x640 on one H200 is a latency, not a limit.

What carries the argument

The load-bearing object is the bounded recurrent loop formed by the student and the external world state bank. The bank is built by taking monocular depth estimates for twelve frames of each generated chunk and unprojecting them with the known camera intrinsics and extrinsics; a read ranks stored views by co-visibility with the target camera, renders up to eight sources with z-buffering, and returns a view-aligned warp plus a per-pixel visibility mask. That mask decides how much geometry enters the denoiser: unsupported regions are assigned full noise and pruned from the token sequence, so the memory contributes exactly where stored coverage exists and nowhere else. Making long-horizon supervision affordable is the teacher's chunk-wise sparse attention, where each chunk sees a first-frame sink, local context with one-frame overlap, compressed nearby frames, a few selected distant frames, and a linear-attention global state, keeping attention cost approximately linear in sequence length. The 30-second distribution-matching objective is applied to the full self-forced rollout with per-chunk detached backward graphs, so the supervision horizon exceeds the gradient horizon and long-range signals reach the student without backpropagating through the whole trajectory.

What would settle it

Take a synthetic scene with ground-truth depth, run a leave-and-return camera trajectory, and compare the bank-rendered warp with the true geometry under the same camera pose. If revisit PSNR or geometric alignment does not rise once the retention window covers the time away, or if depth bias causes systematic misalignment on generated imagery, the claim that external geometric memory provides persistent recall fails. A second check is whether per-chunk latency on a single H200 stays at the reported 2.11 s after many thousands of chunks with the bank at full retention.

Watch

Extended reading notes

Core claim

Evoke's central claim is that persistent memory and interactive latency do not have to be carried by the same mechanism. Scene geometry is held in an explicit, camera-indexed world state bank; at each step the bank is read by rendering stored surfaces into the current view, the render is injected at the coarsest stage of a three-step pyramid denoiser alongside a short local history, and the emitted chunk is written back to the bank with a bounded retention budget. The paper's formal statement is that extending a session increases only the number of recurrent calls, without increasing the context length, positional span, or computational footprint of an individual call. To give this bounded student the long-horizon behavior it needs, the teacher is redesigned for supervision: chunk-wise sparse attention with bounded local, distant, and linear-attention sources makes a 31.4-second full-window objective practical, per-chunk conditioning makes mid-session prompt changes expressible, and distribution matching over the complete self-forced rollout transfers those capabilities. The reported experimental outcome is a three-step, CFG-free model that leads WBench navigation, stays competitive on VBench-Long and VBench-2.0, keeps photometric statistics flat over 65.5-minute rollouts, and revisits stored geometry with 2.3-3.2 dB higher PSNR once retention covers the time away.

Load-bearing premise

The load-bearing premise is that the monocular depth estimates used to build the world state bank are accurate enough that geometry rendered on a return visit aligns with what the session actually generated; if depth is systematically biased on synthetic or generated imagery, the external-memory consistency benefit fails.

Editorial extensions

If this is right

  • If the bounded-recurrence claim holds, session length can grow for hours without changing the per-chunk cost, so open-ended interactive generation becomes a repeated fixed-size call rather than a context-management problem.
  • The three-step, CFG-free student inherits long-horizon photometric stability and mid-session prompt responsiveness from the teacher, meaning low latency need not cap long-term consistency.
  • Because the world state bank is read through camera pose, revisiting a location restores stored geometry instead of relying on the denoiser to reinvent it; the paper reports 2.3-3.2 dB higher revisit PSNR once retention covers the time away.
  • Per-chunk text conditioning introduces a timed evocation capability: new elements can appear in unanchored space, with a 67% realization rate in the controlled test, while geometry-supported content resists overwriting at 4%, so users can stage events without destabilizing the scene.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • One implication the paper leaves implicit: if the 30-second supervision horizon saturates, the same teacher recipe could supervise much longer rollouts without recomputing full-trajectory objectives; the paper's own window-length sweep suggests diminishing returns once the relevant temporal scale is covered, and an explicit test would measure drift on hour-scale rollouts after training with differ
  • The world state bank's fidelity is only as good as the monocular depth used to build it; a testable extension is to feed the depth estimator's calibrated uncertainty into the visibility mask so unreliable geometry is suppressed rather than injected, which could improve revisit consistency on synthetic or stylized content where depth priors are weakest.
  • The evocation asymmetry, where text can add unanchored content but not overwrite anchored geometry, suggests a practical authoring rule: introduce controllable elements before their anchoring surfaces are generated, or extend the bank with an explicit erase-and-rewrite channel that the paper does not claim.
  • Connecting to embodied navigation, camera-pose-indexed geometry could serve as a general episodic memory for agents that must re-identify locations; Evoke demonstrates the mechanism for generation, and the same read-write pattern could be evaluated on navigation metrics.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. Evoke is an interactive video world model that decouples persistent scene state from the denoiser context. A camera-indexed external world state bank stores geometry obtained by unprojecting monocular depth estimates from generated chunks and renders view-relevant stored content back into the current view, while the teacher is redesigned with chunk-wise sparse attention and per-chunk text conditioning to provide roughly 30-second supervision via distribution matching under self-forced rollouts; this supervision is transferred to a three-step, CFG-free student. The paper claims that extending a session increases only the number of recurrent calls, that each 1.5 s chunk is generated in 2.11 s on one H200 at 384x640, that the method reaches state-of-the-art performance on WBench while remaining competitive on VBench-Long and VBench-2.0, and that hour-scale rollouts remain stable.

Significance. If the bounded-recurrence claim holds, this is a valuable systems contribution: it cleanly separates persistent geometric state from transient denoiser context, and the per-call cost bound follows by construction from fixed budgets (19-frame history, 90 s geometry retention, at most 8 retrieved sources, local positional indices). The paper is unusually explicit about the scope of its evidence: Figure 4 is labeled n=1, the long-session content descriptor is a self-consistency measure rather than an identity metric, VBench protocol deviations are declared, and the WBench row is labeled as the authors' own evaluation. The controlled short- vs long-horizon teacher ablation and the wall-clock decomposition (Fig. 10) are well designed and genuinely informative. However, the geometric-memory benefit is not exercised at novel viewpoints, the headline SOTA claim rests on a 0.1-point margin on a self-run evaluation, and the demonstrated teacher-horizon benefit is photometric only.

major comments (4)
  1. [Sec. 3.4, Sec. 4.4] The quantitative recall test in Sec. 4.4 measures "revisit PSNR between two 12 s windows rendered at identical camera poses." At an identical pose, the stored source frame itself can satisfy the query, so the test does not exercise alignment at nearby or novel viewpoints, which is exactly where a systematically biased monocular depth estimate (the bank is built by unprojecting depth from a monocular model, Sec. 3.4) or drift between the commanded trajectory P_k and the motion the model actually generates would return wrong geometry. No calibration of depth error on generated rollouts is provided, and the Navigation (78.63) and Perspective (69.74) scores in Table 1 indicate that camera adherence is imperfect, so pose drift is not merely hypothetical. I recommend an additional recall evaluation at nearby non-identical poses and a depth-error measurement on generated frames before the general claim of geometric recall is accepted.
  2. [Abstract, Sec. 4.1, Table 7] The abstract's "state-of-the-art performance on WBench" is based on Table 7, whose footnote states "Our own evaluation of the released student, not a leaderboard submission," and the margin over the second-ranked system is 0.1 in Average. The paper itself says this margin is "of the same order as the ones Sec. 4.1 declines to read as wins." The SOTA claim should either be softened or accompanied by the margin and the self-run status in the abstract, since a 0.1-point lead on a self-evaluated row does not support the unqualified wording used in the headline.
  3. [Sec. 3.2, Sec. 4.3, Appendix C] Section 3.2 motivates long-horizon supervision primarily by content drift ("scene identity, object appearance, or spatial layout may evolve gradually while every short window remains individually plausible"), yet the matched short- vs long-teacher comparison shows no significant separation on the content descriptor and no improvement on sharpness; the demonstrated benefit is restricted to photometric stability (Sec. 4.3, Appendix C). The paper is candid about this in the body, but the abstract's "improving resistance to long-term drift" and contribution (iii)'s "strong long-horizon consistency" are broader than the evidence. Either qualify these claims to photometric stability or provide additional evidence of content-level consistency transfer.
  4. [Sec. 4.1, Appendix E, Table 2] The VBench-2.0 row (66.77, leading Veo 3 by 0.05) and VBench-Long row (85.11) use one sample per prompt instead of the official five, different clip lengths, and prompt augmentation; the paper itself states that "Single-sampling inflates variance but not the expectation, and the 0.05 margins reported in Table 2 are of that order." Because the rank-1 VBench-2.0 result and the "competitive" claim depend on margins at or below the declared protocol-induced variance, the main text should carry the same caveat that the appendix expresses, rather than presenting the leaderboard positions without it.
minor comments (6)
  1. [Fig. 8] The caption's phrase "shading marks assertion" is unclear; please specify what the shaded region denotes (for example, the time interval during which the clause is in force).
  2. [Eq. (3)] The role of the detach and the normalization constant nu is described in prose, but the equation alone does not make clear which term is detached; a one-line clarification in the notation would help reproducibility.
  3. [Sec. 3.4] The sentence "because every third frame is ingested" is ambiguous given the 12 depth-estimated frames per chunk and the 9 latent / 36 pixel frames; please clarify the mapping between these counts and the 2160-frame retention budget.
  4. [Sec. 4.4, Fig. 8] The anchored-floor result (4% realization for anchored clauses) means the world state bank actively resists text-driven modification of stored content; this is presented as a property, but it is also an interaction limitation that should be stated explicitly in the conclusions alongside the future-work items.
  5. [Fig. 10] The caption sentence "The geometric path costs 38% of the denoiser and 93% of a recurrent step that skips it" is confusing; rephrasing to state that geometry adds 1.84 s to an approximately 2.08 s denoiser step would make the decomposition easier to read.
  6. [References] References [Wan et al., 2025a] and [Wan et al., 2025b] appear to be the same paper and should be deduplicated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the design claims are architectural, and the key results are supported by external benchmarks and controlled ablations rather than by definitional reuse of their own inputs.

full rationale

Evoke's derivation chain is architectural and empirical rather than self-referential. The bounded recurrent formulation (Sec. 3.1, Eq. 1) is a design definition; its stated property that per-call context and cost are session-independent follows from the fixed budgets of local history, retrieved sources, and positional indices, and it is corroborated by measured wall-clock behavior (Sec. 4.4 and App. B). The teacher and student objectives (Eqs. 2-3) build on the authors' earlier DMD formulation, but the paper's novel claims are tested by controlled comparisons: the long-horizon teacher is compared with a short-horizon teacher under matched recipes (Sec. 4.3), and per-chunk conditioning is ablated against a global prompt (Sec. 4.4). The paper also reports negative or partial results explicitly, e.g. the content descriptor does not significantly separate the two students and anchored prompt changes are realized in only 4% of cases, which shows the evaluation is not rigged to confirm the method. The world-state-bank recall test uses identical camera poses and therefore largely verifies retention and eviction bookkeeping rather than novel-view geometric accuracy, but the paper itself limits the claim ('recognizable rather than pixel-faithful reconstruction' and 'Coverage bounds recall from above; it is not a fidelity measure'), so this is a bounded sanity check rather than a prediction equivalent to its input. External leaderboards (WBench, VBench-2.0, VBench-Long) provide independent evaluation, and the self-citations (Helios, Sekai, scaling-law analysis) are implementation choices rather than load-bearing justifications. No equation is defined in terms of the hypothesis it supports, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The core engineering claims rest on accurate monocular depth, valid distribution-matching gradients, reliable benchmark metrics, exact camera poses, and training-data coverage. None of these are derived in the paper; they are assumptions imported from prior tools and datasets. There are no fitted physical constants and no newly postulated physical entities.

free parameters (4)
  • supervision horizon W = 20 chunks, approx. 31.4 seconds
    Chosen via a sweep over W; authors report large gains while it exposes new rollout perturbations, then diminishing returns. The exact value is a hand-set design parameter, not fitted to benchmark scores.
  • world-state retention budget = 2160 pixel frames / 90 seconds
    Hand-set to cover typical revisit intervals. Boundedness of per-step cost is independent of the value, but the 2.11 s runtime and the revisit-recall experiments use this setting.
  • chunk length = 9 latent frames = 1.5 s at 24 fps
    Architectural choice shared by teacher and student. It defines the number of recurrent calls and the linear-attention cost, but is not fitted to a target.
  • history tier sizes and retrieval limits = long/mid/short = 16/2/1 frames; up to 8 source views; visibility threshold 0.5
    Hand-set engineering parameters that bound denoiser context and geometric-conditioning cost. They influence the measured wall clock and the reported coverage values.
assumptions (5)
  • domain assumption Monocular depth estimates are sufficiently accurate for world-state lifting and rendering.
    Sec. 3.4 unprojects depth maps to build the bank; if depth is biased, rendered memory misaligns with true geometry and the revisit-consistency claim fails.
  • domain assumption The teacher and critic distribution-matching scores are a valid training signal for the student.
    Sec. 3.3 and Eq. (3) rely on DMD gradients from the teacher and critic; this assumes flow-matching teacher scores are meaningful for the 3-step student.
  • domain assumption Official benchmark answer keys and metrics measure what the paper claims.
    WBench and VBench rows are self-run or leaderboard-based; the SOTA and competitiveness claims depend on metric fidelity.
  • domain assumption Camera trajectories are known exactly and can be used as conditioning.
    Eq. (1) and the camera-control training stages assume exact pose as input; pose error enters both the bank write and bank read paths.
  • domain assumption The training distribution (Sekai plus internal videos) covers the deployment conditions.
    Claims of hour-scale open-ended generation assume the model generalizes beyond its training videos; no distribution-shift test is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Alaya-EVOKE: From Linear-Scaling Supervision to Endless World." pith.science (2026). https://pith.science/paper/BL4QQFWH

@misc{pith2026260813546,
  author       = {Pith},
  title        = {Pith review of: Alaya-EVOKE: From Linear-Scaling Supervision to Endless World},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BL4QQFWH}},
  note         = {Machine review of arXiv:2608.13546}
}
abstract

Interactive world models must support persistent memory, responsive interaction, and long-horizon generation, yet these requirements place conflicting demands on the model. Maintaining history in the denoiser context or key-value cache incurs growing cost, forcing a trade-off between session length and retained memory, while low-latency interaction relies on few-step generation whose capabilities are bounded by its teacher. Evoke addresses both limitations by externalizing persistent world state and redesigning the teacher for long-horizon interactive generation. Scene geometry is maintained in an external, camera-indexed world state bank, from which only view-relevant information is retrieved, keeping the denoiser context bounded as the session grows. Rather than treating the teacher as a fixed generator, we design it for long-horizon supervision: its sparse attention combines chunk-wise grouping, retrieval of selected distant frames, and a linear-attention global state, yielding linear growth in memory and compute while enabling supervision over long horizons. Such supervision exposes content drift that stays locally plausible within short windows, while per-chunk conditioning enables prompt changes and event control throughout the sequence. A 30-second distribution-matching objective, applied under self-forced rollouts, transfers both capabilities to a three-step student that uses no classifier-free guidance, improving resistance to long-term drift while preserving responsive conditioning. With bounded context and recurrent external memory, Evoke supports open-ended, continuously evolving generation; on a single H200 at $384\times 640$, each $1.5\,\mathrm{s}$ chunk is generated in $2.11\,\mathrm{s}$. As a three-step world model, Evoke achieves state-of-the-art performance on WBench while remaining competitive on VBench-Long and VBench-2.0.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 2 canonical work pages

  1. [2]

    Skyreels-v2: Infinite-length film generative model.arXiv preprint arXiv:2504.13074, 2025a

    Guibin Chen, Dixuan Lin, Jiangping Yang, Chunze Lin, Junchen Zhu, Mingyuan Fan, Hao Zhang, Sheng Chen, Zheng Chen, Chengcheng Ma, et al. Skyreels-v2: Infinite-length film generative model.arXiv preprint arXiv:2504.13074, 2025a. Junsong Chen, Yuyang Zhao, Jincheng Yu, Ruihang Chu, Junyu Chen, Shuai Yang, Xianbang Wang, Yicheng Pan, Daquan Zhou, Huan Ling, ...

    arXiv

  2. [3]

    Creativity Commonsense Controllability Human Fidelity Physics Model Composition Diversity Instance Preserv

    Raw scores in [0, 100], higher is better;boldis best, underline second best. Creativity Commonsense Controllability Human Fidelity Physics Model Composition Diversity Instance Preserv. Motion Rational. Dyn. Spatial Rel. Dyn. Attribute Motion Order Human Interaction Complex Landscape Complex Plot Camera Motion Human Anatomy Human Identity Human Clothes Mec...

    work page arXiv 2025

  3. [5]

    happyoyster.com

    https://www. happyoyster.com. Hao He, Yinghao Xu, Yuwei Guo, Gordon Wetzstein, Bo Dai, Hongsheng Li, and Ceyuan Yang. Cameractrl: Enabling camera control for text-to-video generation.arXiv preprint arXiv:2404.02101,

    arXiv

  4. [6]

    Cameractrl ii: Dynamic scene exploration via camera-controlled video diffusion models

    Hao He, Ceyuan Yang, Shanchuan Lin, Yinghao Xu, Meng Wei, Liangke Gui, Qi Zhao, Gordon Wetzstein, Lu Jiang, and Hongsheng Li. Cameractrl ii: Dynamic scene exploration via camera-controlled video diffusion models. InCVPR, 2025a. Xianglong He, Chunli Peng, Zexiang Liu, Boyang Wang, Yifan Zhang, Qi Cui, Fei Kang, Biao Jiang, Mengyin An, Yangyang Ren, et al. ...

    arXiv

  5. [8]

    Lora: Low-rank adaptation of large language models.arXiv preprint arXiv:2106.09685,

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models.arXiv preprint arXiv:2106.09685,

    arXiv

  6. [10]

    Deepspeed ulysses: System optimizations for enabling training of extreme long sequence transformer models.arXiv preprint arXiv:2309.14509,

    Sam Ade Jacobs, Masahiro Tanaka, Chengming Zhang, Minjia Zhang, Shuaiwen Leon Song, Samyam Rajbhandari, and Yuxiong He. Deepspeed ulysses: System optimizations for enabling training of extreme long sequence transformer models.arXiv preprint arXiv:2309.14509,

    arXiv

  7. [12]

    Pyramidal flow matching for efficient video generative modeling.arXiv preprint arXiv:2410.05954,

    Yang Jin, Zhicheng Sun, Ningyuan Li, Kun Xu, Kun Xu, Hao Jiang, Nan Zhuang, Quzhe Huang, Yang Song, Yadong Mu, and Zhouchen Lin. Pyramidal flow matching for efficient video generative modeling.arXiv preprint arXiv:2410.05954,

    arXiv

  8. [13]

    Hunyuan-gamecraft: High-dynamic interactive game video generation with hybrid history condition.arXiv preprint arXiv:2506.17201, 2025a

    Jiaqi Li, Junshu Tang, Zhiyong Xu, Longhuang Wu, Yuan Zhou, Shuai Shao, Tianbao Yu, Zhiguo Cao, and Qinglin Lu. Hunyuan-gamecraft: High-dynamic interactive game video generation with hybrid history condition.arXiv preprint arXiv:2506.17201, 2025a. Wuyang Li, Wentao Pan, Po-Chien Luan, Yang Gao, and Alexandre Alahi. Stable video infinity: Infinite- length ...

    arXiv

Show all 37 references
  1. [14]

    Depth anything 3: Recovering the visual space from any views.arXiv preprint arXiv:2511.10647, 2025a

    Haotong Lin, Sili Chen, Junhao Liew, Donny Y Chen, Zhenyu Li, Guang Shi, Jiashi Feng, and Bingyi Kang. Depth anything 3: Recovering the visual space from any views.arXiv preprint arXiv:2511.10647, 2025a. Shanchuan Lin, Xin Xia, Yuxi Ren, Ceyuan Yang, Xuefeng Xiao, and Lu Jiang...

    arXiv

  2. [15]

    Latent consistency models: Synthesizing high-resolution images with few-step inference.arXiv preprint arXiv:2310.04378,

    Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthesizing high-resolution images with few-step inference.arXiv preprint arXiv:2310.04378,

    arXiv

  3. [16]

    Yume-1.5: A text-controlled interactive world generation model.arXiv preprint arXiv:2512.22096,

    16 Xiaofeng Mao, Zhen Li, Chuanhao Li, Xiaojie Xu, Kaining Ying, Tong He, Jiangmiao Pang, Yu Qiao, and Kaipeng Zhang. Yume-1.5: A text-controlled interactive world generation model.arXiv preprint arXiv:2512.22096,

  4. [17]

    Cosmos world foundation model platform for physical ai.arXiv preprint arXiv:2501.03575,

    NVIDIA. Cosmos world foundation model platform for physical ai.arXiv preprint arXiv:2501.03575,

    arXiv

  5. [18]

    Advancing open-source world models.arXiv preprint arXiv:2601.20540,

    Robbyant Team, Zelin Gao, Qiuyu Wang, Yanhong Zeng, Jiapeng Zhu, Ka Leong Cheng, Yixuan Li, Hanlin Wang, Yinghao Xu, Shuailei Ma, Yihang Chen, Jie Liu, Yansong Cheng, Yao Yao, Jiayi Zhu, Yihao Meng, Kecheng Zheng, Qingyan Bai, Jingye Chen, Zehong Shen, Yue Yu, Xing Zhu, Yujun ...

    arXiv

  6. [19]

    Progressive distillation for fast sampling of diffusion models.arXiv preprint arXiv:2202.00512,

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models.arXiv preprint arXiv:2202.00512,

    arXiv

  7. [20]

    Lyra 2.0: Explorable generative 3d worlds.arXiv preprint arXiv:2604.13036,

    Tianchang Shen, Sherwin Bahmani, Kai He, Sangeetha Grama Srinivasan, Tianshi Cao, Jiawei Ren, Ruilong Li, Zian Wang, Nicholas Sharp, Zan Gojcic, et al. Lyra 2.0: Explorable generative 3d worlds.arXiv preprint arXiv:2604.13036,

    arXiv

  8. [21]

    Consistency models.arXiv preprint arXiv:2303.01469,

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models.arXiv preprint arXiv:2303.01469,

    arXiv

  9. [22]

    Worldplay: Towards long-term geometric consistency for real-time interactive world modeling.arXiv preprint arXiv:2512.14614,

    Wenqiang Sun, Haiyu Zhang, Haoyuan Wang, Junta Wu, Zehan Wang, Zhenwei Wang, Yunhong Wang, Jun Zhang, Tengfei Wang, and Chunchao Guo. Worldplay: Towards long-term geometric consistency for real-time interactive world modeling.arXiv preprint arXiv:2512.14614,

    arXiv

  10. [23]

    Kimi linear: An expressive, efficient attention architecture.arXiv preprint arXiv:2510.26692,

    Kimi Team, Yu Zhang, Zongyu Lin, Xingcheng Yao, Jiaxi Hu, Fanqing Meng, Chengyin Liu, Xin Men, Songlin Yang, Zhiyuan Li, et al. Kimi linear: An expressive, efficient attention architecture.arXiv preprint arXiv:2510.26692,

    arXiv

  11. [24]

    Diffusion models are real-time game engines.arXiv preprint arXiv:2408.14837,

    Dani Valevski, Yaniv Leviathan, Moab Arar, and Shlomi Fruchter. Diffusion models are real-time game engines.arXiv preprint arXiv:2408.14837,

    arXiv

  12. [25]

    Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025a

    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, 2025a. Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-...

    arXiv

  13. [26]

    Warp-as-history: Generalizable camera-controlled video generation from one training video.arXiv preprint arXiv:2605.15182,

    Yifan Wang and Tong He. Warp-as-history: Generalizable camera-controlled video generation from one training video.arXiv preprint arXiv:2605.15182,

    arXiv

  14. [27]

    Matrix-game 3.0: Real-time and streaming interactive world model with long-horizon memory.arXiv preprint arXiv:2604.08995,

    Zile Wang, Zexiang Liu, Jiaxing Li, Kaichen Huang, Baixin Xu, Fei Kang, Mengyin An, Peiyu Wang, Biao Jiang, Yichen Wei, et al. Matrix-game 3.0: Real-time and streaming interactive world model with long-horizon memory.arXiv preprint arXiv:2604.08995,

    arXiv

  15. [28]

    Longlive: Real-time interactive long video generation.arXiv preprint arXiv:2509.22622,

    Shuai Yang, Wei Huang, Ruihang Chu, Yicheng Xiao, Yuyang Zhao, Xianbang Wang, Muyang Li, Enze Xie, Yingcong Chen, Yao Lu, et al. Longlive: Real-time interactive long video generation.arXiv preprint arXiv:2509.22622,

    arXiv

  16. [29]

    Gated linear attention trans- formers with hardware-efficient training.arXiv preprint arXiv:2312.06635,

    Songlin Yang, Bailin Wang, Yikang Shen, Rameswar Panda, and Yoon Kim. Gated linear attention trans- formers with hardware-efficient training.arXiv preprint arXiv:2312.06635,

    arXiv

  17. [30]

    Improved distribution matching distillation for fast image synthesis

    Tianwei Yin, Michaël Gharbi, Taesung Park, Richard Zhang, Eli Shechtman, Fredo Durand, and Bill Freeman. Improved distribution matching distillation for fast image synthesis. InNeurIPS, 2024a. Tianwei Yin, Michaël Gharbi, Richard Zhang, Eli Shechtman, Fredo Durand, William T F...

    arXiv

  18. [31]

    Helios: Real real-time long video generation model.arXiv preprint arXiv:2603.04379,

    Shenghai Yuan, Yuanyang Yin, Zongjian Li, Xinwei Huang, Xiao Yang, and Li Yuan. Helios: Real real-time long video generation model.arXiv preprint arXiv:2603.04379,

  19. [32]

    Packing input frame context in next-frame prediction models for video generation.arXiv preprint arXiv:2504.12626,

    Lvmin Zhang and Maneesh Agrawala. Packing input frame context in next-frame prediction models for video generation.arXiv preprint arXiv:2504.12626,

  20. [33]

    Matrix-game: Interactive world foundation model.arXiv preprint arXiv:2506.18701,

    Yifan Zhang, Chunli Peng, Boyang Wang, Puyi Wang, Qingcheng Zhu, Fei Kang, Biao Jiang, Zedong Gao, Eric Li, Yang Liu, et al. Matrix-game: Interactive world foundation model.arXiv preprint arXiv:2506.18701,

    arXiv

  21. [34]

    Vbench-2.0: Advancing video generation benchmark suite for intrinsic faithfulness

    Dian Zheng, Ziqi Huang, Hongbo Liu, Kai Zou, Yinan He, Fan Zhang, Lulu Gu, Yuanhan Zhang, Jingwen He, Wei-Shi Zheng, et al. Vbench-2.0: Advancing video generation benchmark suite for intrinsic faithfulness. arXiv preprint arXiv:2503.21755,

    arXiv

  22. [35]

    Causal forcing: Autoregres- sive diffusion distillation done right for high-quality real-time interactive video generation.arXiv preprint arXiv:2602.02214,

    Hongzhou Zhu, Min Zhao, Guande He, Hang Su, Chongxuan Li, and Jun Zhu. Causal forcing: Autoregres- sive diffusion distillation done right for high-quality real-time interactive video generation.arXiv preprint arXiv:2602.02214,

    arXiv

  23. [36]

    18 Appendix A Training curves Figure 9 is the optimization record of the two distillation stages that produce the released Evoke student: long-distillagainst the Evoke Teacher on 6 × 8 GPUs, and the shortpost-distillcontinuation that yields the released student. 0 1k 2k 0 0.15...

    1981

  24. [2021]

    Self forcing: Bridging the train-test gap in autoregressive video diffusion.arXiv preprint arXiv:2506.08009, 2025a

    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, 2025a. Ziqi Huang, Yinan He, Jiashuo Yu, Fan Zhang, Chenyang Si, Yuming Jiang, Yuanhan Zhang, Tianxi...

    arXiv

  25. [2022]

    Relic: Interactive video world model with long-horizon memory.arXiv preprint arXiv:2512.04040,

    15 Yicong Hong, Yiqun Mei, Chongjian Ge, Yiran Xu, Yang Zhou, Sai Bi, Yannick Hold-Geoffroy, Mike Roberts, Matthew Fisher, Eli Shechtman, Kalyan Sunkavalli, Feng Liu, Zhengqi Li, and Hao Tan. Relic: Interactive video world model with long-horizon memory.arXiv preprint arXiv:25...

  26. [2023]

    Moga: Mixture-of-groups attention for end-to-end long video generation.arXiv preprint arXiv:2510.18692,

    Weinan Jia, Yuning Lu, Mengqi Huang, Hualiang Wang, Binyuan Huang, Nan Chen, Mu Liu, Jidong Jiang, and Zhendong Mao. Moga: Mixture-of-groups attention for end-to-end long video generation.arXiv preprint arXiv:2510.18692,

  27. [2024]

    Mixture of contexts for long video generation.arXiv preprint arXiv:2508.21058,

    Shengqu Cai, Ceyuan Yang, Lvmin Zhang, Yuwei Guo, Junfei Xiao, Ziyan Yang, Yinghao Xu, Zhenheng Yang, Alan Yuille, Leonidas Guibas, et al. Mixture of contexts for long video generation.arXiv preprint arXiv:2508.21058,

  28. [2025]

    Infinite worlds with versatile interactions.arXiv preprint arXiv:2607.07534,

    Zelin Gao, Qiuyu Wang, Jiapeng Zhu, Jingye Chen, Zichen Liu, Qingyan Bai, Jiahao Wang, Yufeng Yuan, Hanlin Wang, Yichong Lu, et al. Infinite worlds with versatile interactions.arXiv preprint arXiv:2607.07534,

    arXiv

  29. [2026]

    Tianrui Feng, Zhi Li, Shuo Yang, Haocheng Xi, Muyang Li, Xiuyu Li, Lvmin Zhang, Keting Yang, Kelly Peng, Song Han, et al

    Blog post; no accompanying paper. Tianrui Feng, Zhi Li, Shuo Yang, Haocheng Xi, Muyang Li, Xiuyu Li, Lvmin Zhang, Keting Yang, Kelly Peng, Song Han, et al. Streamdiffusionv2: A streaming system for dynamic and interactive video generation. arXiv preprint arXiv:2511.07399,

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.