Pith. sign in

REVIEW 3 major objections 6 minor 37 references

Game world models need explicit internal states, not just pixels, to obey the rules.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-30 22:03 UTC pith:BEITFEDO

load-bearing objection Solid subfield paper: explicit state heads actually help rule-consistent game rollouts on SF3, but the 18.6% headline leans too hard on an LLM judge that also labeled the data. the 3 major comments →

arxiv 2607.26754 v1 pith:BEITFEDO submitted 2026-07-29 cs.CV

StatePlay: State-Aware Game World Models for Mechanics-Consistent Generation

Zijun Lin , Zeqing Wang , Cheston Tan , Bihan Wen , Yeying Jin This is my paper
classification cs.CV
keywords game world modelsstate-aware generationmechanics consistencymixture of transformersaction-conditioned videoStreet Fighter 3flow matchinginteractive simulation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Visually realistic game generators can still break the game: characters attack at zero health or fire super moves without a full meter. StatePlay argues that playable simulation requires jointly predicting the hidden rule variables—health, skill meters, timers—together with the frames. It builds a dual-branch mixture-of-transformers model on a new Street Fighter 3 dataset of synchronized frames, actions, and states, so predicted states can steer what appears on screen. The result is accurate state traces and substantially higher mechanics fidelity than strong baselines that only learn from pixels, without sacrificing visual quality or action control. The point for a sympathetic reader is simple: if you want generated games that are actually playable, you must model the state machine, not only the pixels.

Core claim

Explicitly coupling internal game-state prediction with visual generation produces rollouts that follow state-dependent rules far better than pixel-only game world models. On Street Fighter 3, StatePlay keeps average normalized state error below 0.06 and raises mechanics fidelity by about 18.6% over the best comparable baseline while holding visual quality and action controllability.

What carries the argument

A mixture-of-transformers (MoT) dual branch: a lightweight state expert trained with Smooth L1 regression and a visual expert trained with flow matching, linked by bidirectional joint attention so states guide frames and frames inform states, with player actions injected into both.

Load-bearing premise

That scoring “mechanics fidelity” with the same class of vision-language judges used to label the training set is a fair measure of true rule compliance rather than agreement with that labeling stack.

What would settle it

Run the same action sequences in the real SF3 emulator and in StatePlay; if win/lose screens, super-art gates, and HP/meter traces diverge often—or if a non-LLM judge (engine memory dump or human raters) erases the reported fidelity gap versus a fine-tuned pixel-only baseline—the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Pixel-only game world models will keep producing rule-breaking rollouts even after in-domain fine-tuning.
  • Datasets that pair frames and actions with explicit states become necessary infrastructure for playable generators.
  • Modality-specific losses beat forcing compact rule variables through the same flow-matching objective used for video.
  • Separating state and visual experts while sharing joint attention is a practical recipe other game world models can adopt.
  • Mechanics-critical events (KO, super art, meter reset) should be oversampled when training interactive generators.

Where Pith is reading between the lines

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

  • The same state-branch pattern should transfer to other genres where rules hide in counters (ammo, nitro, cooldowns), not only fighting games.
  • Closing the remaining HUD-vs-predicted-state mismatch may need tighter pixel supervision on bars and meters, not only latent state regression.
  • If LLM judges and LLM labelers share biases, future benchmarks will need engine-grounded oracles before claiming “playable” status.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. StatePlay argues that game world models must jointly predict internal game states (timer, health, skill meters) and pixels to produce mechanics-consistent rollouts, rather than learning rules only implicitly from frames and actions. The authors build a synchronized SF3 state–frame–action dataset (10k clips, mechanics-balanced), and train a MoT-style dual-branch model (0.76B state expert + 5B visual expert on Wan2.2-TI2V) with joint attention (Eq. 3), flow-matching for video (Eq. 4), and Smooth-L1 regression for states (Eq. 5). On a 100-clip test set they report normalized state L1 below 0.06 (alignment 0.947) and mechanics fidelity of 82.3%/78.3% (Gemini/GPT) versus 63.7%/59.7% for the best fine-tuned visual-only baseline (ReactiveGWM), i.e. the advertised ~18.6% gain, with ablations favoring MoT over shared backbones and regression over flow matching for states (Tab. 2).

Significance. The paper correctly identifies a real gap: recent game world models optimize visual and action controllability but routinely violate state-dependent rules (super arts without meter, play after KO). Coupling an explicit, engine-readable state head with frame generation is a natural and transferable idea, and the MoT-plus-modality-specific-loss design is a concrete, reusable recipe. Strengths include a carefully balanced state-aware dataset, clear architecture (Fig. 3), grounded state-alignment numbers, informative ablations (Tab. 2), and qualitative cases where win/lose screens and meter resets track rules better than a strong in-domain baseline (Fig. 4/7). If the mechanics-fidelity claim holds under engine-grounded checks and broader games, this would push the field from pixel demos toward playable simulators. Credit is due for releasing a full training recipe and for showing that explicit state modeling need not destroy SSIM/LPIPS or action accuracy.

major comments (3)
  1. [§4.1, §4.3, Tab. 1, Appendix D] The headline 18.6% mechanics-fidelity gain (abstract; §4.3; Tab. 1: ReactiveGWM 63.7% → StatePlay 82.3% under Gemini-3.1-Pro) is measured almost entirely by LLM visual judges (Gemini-3.1-Pro and GPT-5.5) with the Appendix D prompt and reference images (§4.1). The same Gemini stack is used to label training clips for result_win/lose and macro_success/fail and to write NPC strategy text (§3.1, Appendix B). This creates a load-bearing circularity risk: the primary claim metric can reward agreement with the labeler–judge stack rather than true SF3 rule compliance. Please add an engine-grounded mechanics evaluation on generated rollouts—e.g., whether predicted or implied HP/meter/timer trajectories and termination/super-art events match emulator memory or deterministic rule checks conditioned on the commanded actions—and report agreement between LLM judges and those checks. Until then the 18.
  2. [§4.1 State Alignment / Mechanics Fidelity; Appendix C] State alignment (normalized L1 < 0.06, score 0.947) is engine-supervised and strong, but the paper does not close the causal loop from accurate state prediction to mechanics-consistent pixels on the generated sequence. Mechanics fidelity is defined as category match to target labels (win/lose/macro/normal), not as consistency between the model’s predicted state trace, the generated HUD/events, and emulator transitions. Appendix C already notes HUD/state visual mismatches. A quantitative bridge—e.g., correlation of predicted HP/meter with rendered bars, or rule-violation rate under forced zero-HP / full-meter action sequences—would make the central “mechanics-consistent generation” claim much more secure.
  3. [§4.1–4.3, §5, Tab. 1–2] All quantitative claims rest on a single game (SF3) and a 100-clip test set evenly covering mechanics categories (§4.1). Zero-shot baselines are largely out-of-domain and weak on visuals; the decisive comparison is fine-tuned ReactiveGWM on the same data (Tab. 1). LingBot-family fine-tunes are omitted for compute reasons. The directional MoT/regression ablations (Tab. 2) are informative but inherit the same judge and domain. The manuscript should either (i) add at least one additional title or a held-out character/rule split with engine checks, or (ii) clearly scope the contribution as a single-domain proof-of-concept and temper abstract/conclusion language about “complete and mechanically faithful game generation.”
minor comments (6)
  1. [Abstract; Tab. 1] Abstract and intro state “average normalized L1 distance below 0.06” while Tab. 1 reports only the aggregate alignment score 0.947. Please report per-variable normalized L1 (timer, P1/P2 HP, both meters) in the main table or appendix.
  2. [§3.2; Appendix E] Eq. (1)–(2) use flow-style noise on states even though the state objective is regression (Eq. 5). Appendix E compares broadcast vs noise inputs, but the main text should briefly justify why noisy state inputs are retained under a non-flow state loss.
  3. [§4.1 Action Control] ClipAttackNet (ResNet-18 + dilated TCN, conf. 0.7) and Move-Acc thresholds are custom; report validation accuracy of the attack classifier and the exact displacement thresholds so action-control numbers are reproducible.
  4. [Fig. 1, Fig. 4, Fig. 7] Figure 1’s bottom “table” is hard to read in text form; ensure the camera-ready figure has legible axis labels. Fig. 4/7 would benefit from overlaid predicted vs GT state traces next to frames.
  5. [Throughout] Typos/spacing artifacts from PDF extraction appear throughout (e.g., “Recentgameworldmodels”, “StatePlayjointly”). Clean compound-word spacing in the camera-ready version.
  6. [§2.2] Related work cites WildWorld and From Pixels to States as closest prior datasets; a short explicit comparison of state schemas and whether those works train joint generators would sharpen novelty.

Circularity Check

0 steps flagged

No derivation circularity: state targets are emulator memory; fidelity gains are empirical comparisons, not by-construction identities.

full rationale

StatePlay’s load-bearing technical claims do not reduce to their inputs by definition. Internal states (timer, HP, skill meters) are read from emulator memory and supervised with Smooth L1 regression (Eq. 5); visual latents use standard flow matching (Eq. 4); the MoT joint-attention coupling (Eq. 3) and modality-specific losses are architectural choices validated by ablation (Tab. 2), not tautologies. State alignment (normalized L1 < 0.06, score 0.947) is checked against held-out emulator traces. The 18.6% mechanics-fidelity lift is an empirical delta versus fine-tuned visual-only baselines on the same data (Tab. 1), not a fitted parameter renamed as a prediction. Gemini is used for dataset balancing labels and as a post-hoc visual judge; that is a shared-stack validity risk for the metric, not a self-definitional or fitted-input circularity in the model’s derivation. No uniqueness theorem, self-citation chain, or ansatz-smuggling step forces the central result. Per the analyzer’s scope, evaluation-judge overlap is outside circularity; the paper’s chain is self-contained against engine-grounded state targets and comparative baselines.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 4 invented entities

The claim rests on standard generative-model practice (VAE latents, flow matching, transformer experts), the domain premise that SF3 mechanics are captured by five memory-read scalars plus actions, and paper-specific choices: MoT joint attention, Smooth-L1 state loss, loss weights, dataset balancing, and LLM judges as the fidelity oracle. No new physical entities; invented pieces are architectural and metric constructs.

free parameters (5)
  • lambda_state, lambda_video = 1, 1
    Joint training weights in Eq. (6); set to 1 and 1 without a reported sweep in the main text.
  • Training schedule and capacity knobs = 40k steps; bs=4; lr=5e-5; N=30
    40,000 steps, batch size 4, LR 5e-5, N=30 blocks, C=3072, state branch 0.76B—chosen design hyperparameters that affect the reported margins.
  • Dataset category mix = 10k clips; 4×1k critical + 6k normal
    40% state-critical (10% each win/lose/macro success/fail) vs 60% normal is a hand-set balancing policy (§3.1, Tab. 4) that shapes how much mechanics signal the model sees.
  • Macro skill threshold tau = tau (game-defined, not numerically listed in main text)
    Skill-meter threshold for macro success/fail labeling (Tab. 4); game-rule constant used in supervision construction.
  • Attack classifier confidence threshold = 0.7
    ClipAttackNet Att-Acc uses confidence 0.7 (§4.1); affects action-control scores though not the headline mechanics delta.
axioms (5)
  • domain assumption Internal SF3 mechanics relevant to playability are adequately represented by timer, both HPs, and both skill meters extracted from emulator memory, synchronized to frames and an 11-D action space.
    Dataset and state branch dimension S are built on this (§3.1); if important rules live outside these variables, state alignment would not imply mechanics fidelity.
  • domain assumption Flow-matching video objectives on Wan-style latents plus Smooth L1 on states are appropriate modality-specific risks (Eqs. 4–5).
    Standard for diffusion/flow video; regression choice is justified empirically in §4.4 against FM for states.
  • ad hoc to paper Bidirectional joint attention between state and visual tokens (Eq. 3) is sufficient cross-modal coupling without full weight sharing.
    Core MoT design claim; supported by ablation vs shared backbone (Tab. 2) but not derived from theory.
  • ad hoc to paper Gemini-3.1-Pro / GPT-5.5 judgments under the Appendix D prompt equal ground-truth mechanics classes for generated video.
    Load-bearing for the 18.6% fidelity claim (§4.1); not an engine oracle.
  • standard math Transformer, attention, and VAE latent algebra behave as in the cited Wan2.2 / flow-matching literature.
    Background ML machinery assumed throughout §3.2.
invented entities (4)
  • StatePlay MoT dual-branch world model (0.76B state expert + 5B visual expert) no independent evidence
    purpose: Jointly denoise/predict states and frames with cross-attention coupling.
    Architectural construct; evaluated only inside this paper’s SF3 setup.
  • SF3 synchronized state–frame–action dataset (10k train clips) no independent evidence
    purpose: Enable supervised state dynamics and balanced mechanics events.
    New resource claimed; not evidenced as publicly released in the manuscript.
  • Mechanics fidelity metric via LLM judges no independent evidence
    purpose: Score win/lose/macro correctness of generated rollouts.
    Evaluation construct defined in §4.1/App. D; depends on judge models.
  • ClipAttackNet (ResNet-18 + dilated TCN) no independent evidence
    purpose: Measure attack-command consistency for Att-Acc.
    Custom classifier trained on ~5k clips; no external validation reported.

pith-pipeline@v1.2.0-daily-grok45 · 19590 in / 4278 out tokens · 84343 ms · 2026-07-30T22:03:31.092944+00:00 · methodology

0 comments
read the original abstract

Recent game world models can generate visually realistic and interactive environments conditioned on player actions. However, games are not defined by pixels alone; they are governed by explicit mechanics, namely state-dependent rules that control health reduction, skill activation, and game termination. These mechanics depend on precise internal states, such as health points, skill meters, and timers, which are tightly coupled with visual observations and determine how gameplay evolves. Without modeling these state dynamics, existing game world models may generate visually plausible rollouts but violate the underlying game rules. In this paper, we propose StatePlay, a novel state-aware game world model that jointly predicts visual content and game states to promote mechanics-consistent generation. StatePlay adopts a mixture-of-transformers (MoT)-style architecture that preserves specialized visual and state representations while enabling cross-modal interaction, allowing predicted states to guide frame generation. Each branch is further optimized with a distinct objective suited to its modality. Experiments show that StatePlay achieves an average normalized L1 distance below 0.06 for state prediction. Furthermore, compared with models without explicit state modeling, our method improves mechanics fidelity in generated game rollouts by 18.6%. Overall, our work highlights the importance of state-aware game world modeling and advances beyond pixel-level realism toward complete and mechanically faithful game generation.

Figures

Figures reproduced from arXiv: 2607.26754 by Bihan Wen, Cheston Tan, Yeying Jin, Zeqing Wang, Zijun Lin.

Figure 1
Figure 1. Figure 1: Existing game world models produce mechanically [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of StatePlay. By explicitly modeling game states, including timers, health points, and skill meters, StatePlay guides frame generation to remain consistent with the underlying game mechanics (highlighted in green bounding box). To address this gap, we introduce StatePlay, the first state￾aware game world model that jointly predicts game states and visual content for mechanics-consistent gameplay g… view at source ↗
Figure 3
Figure 3. Figure 3: StatePlay block architecture during training. StatePlay adopts a Mixture-of-Transformers (MoT)-style architecture with separate state and visual branches. A shared joint-attention module enables information exchange between the two branches. The state branch is optimized with a regression loss, while the visual branch is trained using a flow-matching objective. White-filled boxes denote inputs and supervis… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of three example rollouts (a)–(c) generated by the best-performing baseline, ReactiveGWM, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: NPC strategy annotation prompt for classifying [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Evaluation prompt for mechanics fidelity (top), with three ground-truth reference images (bottom). [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison of three example rollouts (a)–(c) generated by the best-performing baseline, ReactiveGWM, [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗

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

37 extracted references · 18 linked inside Pith

  1. [1]

    Z.; and Koltun, V

    Bai, S.; Kolter, J. Z.; and Koltun, V. 2018. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271

    Pith/arXiv arXiv 2018

  2. [2]

    R.; Finn, C.; Fusai, N.; Galliker, M

    Black, K.; Brown, N.; Darpinian, J.; Dhabalia, K.; Driess, D.; Esmail, A.; Equi, M. R.; Finn, C.; Fusai, N.; Galliker, M. Y.; et al. 2025. _ 0.5 : a Vision-Language-Action Model with Open-World Generalization. In 9th Annual Conference on Robot Learning

    2025

  3. [3]

    D.; Edwards, A.; Parker-Holder, J.; Shi, Y.; Hughes, E.; Lai, M.; Mavalankar, A.; Steigerwald, R.; Apps, C.; et al

    Bruce, J.; Dennis, M. D.; Edwards, A.; Parker-Holder, J.; Shi, Y.; Hughes, E.; Lai, M.; Mavalankar, A.; Steigerwald, R.; Apps, C.; et al. 2024. Genie: Generative interactive environments. In Forty-first International Conference on Machine Learning

    2024

  4. [4]

    Capcom . 1998. Street Fighter Alpha 3 . Arcade

    1998

  5. [5]

    W.; Constant, N.; Garcia, X.; Roberts, A.; Tay, Y.; Narang, S.; and Firat, O

    Chung, H. W.; Constant, N.; Garcia, X.; Roberts, A.; Tay, Y.; Narang, S.; and Firat, O. 2023. Unimax: Fairer and more effective language sampling for large-scale multilingual pretraining. arXiv preprint arXiv:2304.09151

    Pith/arXiv arXiv 2023

  6. [6]

    L.; Zhang, H.; Gao, J.; Feng, T.; Liu, Y.; Yao, Y.; Xu, Y.; Zhu, X.; Shen, Y.; and Ouyang, H

    Gao, Z.; Wang, Q.; Zhu, J.; Chen, J.; Liu, Z.; Bai, Q.; Wang, J.; Yuan, Y.; Wang, H.; Lu, Y.; Cheng, K. L.; Zhang, H.; Gao, J.; Feng, T.; Liu, Y.; Yao, Y.; Xu, Y.; Zhu, X.; Shen, Y.; and Ouyang, H. 2026. Infinite Worlds with Versatile Interactions. arXiv:2607.07534

    Pith/arXiv arXiv 2026

  7. [7]

    Google DeepMind . 2026. Gemini 3.1 Pro Model Card. https://deepmind.google/models/model-cards/gemini-3-1-pro/. Accessed: 2026-07-28

    2026

  8. [8]

    He, K.; Zhang, X.; Ren, S.; and Sun, J. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770--778

    2016

  9. [9]

    Hou, B.; Li, G.; Jia, J.; An, T.; Guo, X.; Leng, S.; Geng, H.; Ze, Y.; Harada, T.; Torr, P.; et al. 2026. World model for robot learning: A comprehensive survey. arXiv preprint arXiv:2605.00080

    Pith/arXiv arXiv 2026

  10. [10]

    Hu, A.; Corrado, G.; Griffiths, N.; Murez, Z.; Gurau, C.; Yeo, H.; Kendall, A.; Cipolla, R.; and Shotton, J. 2022. Model-based imitation learning for urban driving. Advances in Neural Information Processing Systems, 35: 20703--20716

    2022

  11. [11]

    Hu, Y.; Zhang, J.; Luo, Y.; Guo, Y.; Chen, X.; Sun, X.; Feng, K.; Lu, Q.; Chen, S.; Zhang, Y.; et al. 2026. Bagelvla: Enhancing long-horizon manipulation via interleaved vision-language-action generation. arXiv preprint arXiv:2602.09849

    arXiv 2026

  12. [12]

    J.; Gao, Y.; Lin, T.-Y.; Lin, Y.-C.; Ge, Y.; Lam, G.; Liang, P.; Song, S.; Liu, M.-Y.; Finn, C.; et al

    Kim, M. J.; Gao, Y.; Lin, T.-Y.; Lin, Y.-C.; Ge, Y.; Lam, G.; Liang, P.; Song, S.; Liu, M.-Y.; Finn, C.; et al. 2026. Cosmos policy: Fine-tuning video models for visuomotor control and planning. arXiv preprint arXiv:2601.16163

    Pith/arXiv arXiv 2026

  13. [13]

    J.; Pertsch, K.; Karamcheti, S.; Xiao, T.; Balakrishna, A.; Nair, S.; Rafailov, R.; Foster, E

    Kim, M. J.; Pertsch, K.; Karamcheti, S.; Xiao, T.; Balakrishna, A.; Nair, S.; Rafailov, R.; Foster, E. P.; Sanketi, P. R.; Vuong, Q.; et al. 2024. OpenVLA: An Open-Source Vision-Language-Action Model. In 8th Annual Conference on Robot Learning

    2024

  14. [14]

    Li, Z.; Meng, Z.; Shi, S.; Peng, W.; Wu, Y.; Zheng, B.; Li, C.; and Zhang, K. 2026 a . Wildworld: A large-scale dataset for dynamic world modeling with actions and explicit state toward generative arpg. arXiv preprint arXiv:2603.23497

    arXiv 2026

  15. [15]

    Li, Z.; Meng, Z.; Shi, S.; Zhai, M.; Tan, J.; Li, C.; and Zhang, K. 2026 b . From Pixels to States: Rethinking Interactive World Models as Game Engines. arXiv:2607.14076

    Pith/arXiv arXiv 2026

  16. [16]

    Liu, S.; Zeng, Z.; Ren, T.; Li, F.; Zhang, H.; Yang, J.; Jiang, Q.; Li, C.; Yang, J.; Su, H.; et al. 2024. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. In European conference on computer vision, 38--55. Springer

    2024

  17. [17]

    Mao, X.; Li, Z.; Li, C.; Xu, X.; Ying, K.; and Zhang, K. 2026. Yume1. 5: A text-controlled interactive world generation model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 7752--7761

    2026

  18. [18]

    OpenAI . 2026. GPT-5.5 System Card. https://openai.com/index/gpt-5-5-system-card/. Accessed: 2026-07-28

    2026

  19. [19]

    Poliquin, M. 2026. Stable Retro : A Maintained Fork of OpenAI 's Gym Retro . Accessed: 2026-07-07

    2026

  20. [20]

    Ravi, N.; Gabeur, V.; Hu, Y.-T.; Hu, R.; Ryali, C.; Ma, T.; Khedr, H.; R \"a dle, R.; Rolland, C.; Gustafson, L.; et al. 2025. Sam 2: Segment anything in images and videos. In International Conference on Learning Representations, volume 2025, 28085--28128

    2025

  21. [21]

    Seedance, T.; Chen, D.; Chen, L.; Chen, X.; Chen, Y.; Chen, Z.; Chen, Z.; Cheng, F.; Cheng, T.; Cheng, Y.; et al. 2026. Seedance 2.0: Advancing video generation for world complexity. arXiv preprint arXiv:2604.14148

    Pith/arXiv arXiv 2026

  22. [22]

    G.; Cao, T.; Ren, J.; Li, R.; Wang, Z.; Sharp, N.; Gojcic, Z.; et al

    Shen, T.; Bahmani, S.; He, K.; Srinivasan, S. G.; Cao, T.; Ren, J.; Li, R.; Wang, Z.; Sharp, N.; Gojcic, Z.; et al. 2026. Lyra 2.0: Explorable generative 3d worlds. arXiv preprint arXiv:2604.13036

    Pith/arXiv arXiv 2026

  23. [23]

    Sun, W.; Zhang, H.; Wang, H.; Wu, J.; Wang, Z.; Wang, Z.; Wang, Y.; Zhang, J.; Wang, T.; and Guo, C. 2025. Worldplay: Towards long-term geometric consistency for real-time interactive world modeling. arXiv preprint arXiv:2512.14614

    Pith/arXiv arXiv 2025

  24. [24]

    Team, K.; Chen, J.; Ci, Y.; Du, X.; Feng, Z.; Gai, K.; Guo, S.; Han, F.; He, J.; He, K.; et al. 2025. Kling-Omni Technical Report. arXiv preprint arXiv:2512.16776

    Pith/arXiv arXiv 2025

  25. [25]

    L.; Li, Y.; Wang, H.; Xu, Y.; Ma, S.; et al

    Team, R.; Gao, Z.; Wang, Q.; Zeng, Y.; Zhu, J.; Cheng, K. L.; Li, Y.; Wang, H.; Xu, Y.; Ma, S.; et al. 2026. Advancing Open-source World Models. arXiv preprint arXiv:2601.20540

    Pith/arXiv arXiv 2026

  26. [26]

    Tong, Z.; Lai, H.; Wang, Z.; Xing, Z.; Cheng, K.; Xu, H.; Pu, Z.; Zhu, S.; Feng, R.; Zhao, J.; et al. 2026. SCOPE: Simulating Cross-game Operations in Playable Environments for FPS World Models. arXiv preprint arXiv:2605.23345

    Pith/arXiv arXiv 2026

  27. [27]

    Wan, T.; Wang, A.; Ai, B.; Wen, B.; Mao, C.; Xie, C.-W.; Chen, D.; Yu, F.; Zhao, H.; Yang, J.; et al. 2025. Wan: Open and advanced large-scale video generative models. arXiv preprint arXiv:2503.20314

    Pith/arXiv arXiv 2025

  28. [28]

    Wang, X.; Zhu, Z.; Huang, G.; Chen, X.; Zhu, J.; and Lu, J. 2024. Drivedreamer: Towards real-world-drive world models for autonomous driving. In European conference on computer vision, 55--72. Springer

    2024

  29. [29]

    C.; Sheikh, H

    Wang, Z.; Bovik, A. C.; Sheikh, H. R.; and Simoncelli, E. P. 2004. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4): 600--612

    2004

  30. [30]

    Wang, Z.; Chen, D.; Xing, Z.; Tong, Z.; Zhang, Y.; Yang, X.; and Jin, Y. 2026 a . ReactiveGWM: Steering NPC in Reactive Game World Models. arXiv preprint arXiv:2605.15256

    Pith/arXiv arXiv 2026

  31. [31]

    Wang, Z.; Liu, Z.; Li, J.; Huang, K.; Xu, B.; Kang, F.; An, M.; Wang, P.; Jiang, B.; Wei, Y.; et al. 2026 b . Matrix-game 3.0: Real-time and streaming interactive world model with long-horizon memory. arXiv preprint arXiv:2604.08995

    Pith/arXiv arXiv 2026

  32. [32]

    Wu, H.; Yu, J.; Zou, Y.; and Liu, X. 2026. Multiworld: Scalable multi-agent multi-view video world models. arXiv preprint arXiv:2604.18564

    Pith/arXiv arXiv 2026

  33. [33]

    L.; Zhu, C.; Xiang, J.; et al

    Ye, S.; Ge, Y.; Zheng, K.; Gao, S.; Yu, S.; Kurian, G.; Indupuru, S.; Tan, Y. L.; Zhu, C.; Xiang, J.; et al. 2026. World action models are zero-shot policies. arXiv preprint arXiv:2602.15922

    Pith/arXiv arXiv 2026

  34. [34]

    Yu, J.; Qin, Y.; Wang, X.; Wan, P.; Zhang, D.; and Liu, X. 2025. Gamefactory: Creating new games with generative interactive videos. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 11590--11599

    2025

  35. [35]

    A.; Shechtman, E.; and Wang, O

    Zhang, R.; Isola, P.; Efros, A. A.; Shechtman, E.; and Wang, O. 2018. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, 586--595

    2018

  36. [36]

    Zheng, W.; Chen, W.; Huang, Y.; Zhang, B.; Duan, Y.; and Lu, J. 2024. Occworld: Learning a 3d occupancy world model for autonomous driving. In European conference on computer vision, 55--72. Springer

    2024

  37. [37]

    Zhu, S.; Peng, Q.; Pu, Z.; Shu, Z.; Ke, X.; Xing, Z.; Tong, Z.; Wang, Z.; Cui, X.; Wang, H.; et al. 2026. Incantation: Natural Language as the Action Interface for Multi-Entity Video World Models. arXiv preprint arXiv:2605.18601

    Pith/arXiv arXiv 2026