MiniWorld: Democratizing the Training of Video World Models from Scratch

Hongcan Guo; Jian Zhang; Jie Chen; Ruochong Zheng; Yian Zhao; Yu Yan

REVIEW 4 major objections 4 minor 39 references

MiniWorld: Democratizing the Training of Video World Models from Scratch

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read MiniWorld demonstrates that a streaming video world model can be trained from scratch on a single 8-GPU server in days, using a block-causal diffusion transformer with a rolling KV cache, and produces stable long-horizon action-conditioned

desk verdict A solid, genuinely useful systems paper: the first fully open from-scratch streaming world model recipe that trains in days on 8 GPUs, with coherent method and released code, but the empirical claims rest on a single internal baseline and the latent-space bottleneck is never inspected. read the letter →

arxiv 2608.01127 v2 pith:YXXAJAEF submitted 2026-08-02 cs.CV

Yian Zhao , Ruochong Zheng , Hongcan Guo , Yu Yan , Jian Zhang , Jie Chen This is my paper · ORCID

classification cs.CV

keywords videoworldmodelsstreaminggenerationdiffusiontransformersautoregressiveflowmatchingKVcacheaction-conditionedfrom-scratchtraining

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

This paper aims to show that a streaming video world model—one that predicts future frames from past frames and agent actions—can be trained from scratch, without adapting a pretrained video generator, using only a single 8-GPU server for a few days. The proposed recipe combines a block-causal Video Diffusion Transformer, a chunk-wise non-decreasing noise schedule built on Diffusion Forcing, two-stage continued training, and a rolling KV cache with pipelined asynchronous denoising. The authors report stable long-horizon, action-conditioned rollouts on robot manipulation (DROID) and camera-control (RealEstate10K) benchmarks, outperforming a bidirectional sliding-window baseline. If these results hold, the work provides a lightweight, transparent baseline that makes video world modeling research accessible, and it isolates train–test alignment as the key to stable streaming generation.

What carries the argument

The load-bearing mechanism is train–test alignment: the block-causal attention pattern and the chunk-wise non-decreasing noise schedule make training mimic autoregressive streaming inference. The named ingredients are: (1) a block-causal Video DiT, a Video Diffusion Transformer with bidirectional attention inside each temporal chunk and causal attention across chunks; (2) a Chunk-oriented Probability Propagation (CoPP) scheduler, which samples monotonic chunk-wise timestep sequences anchored at a random chunk, preserving balanced timestep coverage; (3) two-stage continued training, first on short 21/46-frame clips and then on long 125/253-frame sequences with timestep shifting; and (4) a str

What would settle it

Train the identical MiniWorld recipe with a VAE that has 8x spatial compression instead of 16x (or with a VAE that preserves small action deltas) and compare DROID trajectory accuracy and depth error. If the metrics clearly improve, the original frozen latent space was discarding control-relevant detail, undermining the claim that the latent is a faithful substrate for world modeling. Alternatively, measure whether small gripper displacements produce distinguishable latent codes in the frozen VAE; if they do not, the representation cannot support fine-grained action conditioning.

Watch

Extended reading notes

Core claim

MiniWorld claims that a streaming video world model can be trained from scratch as a block-causal next-state predictor. The architecture is a Video Diffusion Transformer whose attention is bidirectional inside each temporal chunk and causal across chunks, trained with Flow Matching on chunk-wise independent diffusion timesteps that are non-decreasing over the sequence, so earlier chunks are always cleaner than later ones. During streaming inference, completed chunks are committed to a rolling KV cache while a pipelined asynchronous denoiser keeps the active window bounded; the authors report stable long-horizon, action-conditioned rollouts on DROID and RealEstate10K that beat a bidirectional

Load-bearing premise

The approach assumes the frozen pretrained video autoencoder's latent space keeps all the action- and camera-relevant information, so that the compressed video representation used for training contains enough detail to learn how the world responds to controls.

Editorial extensions

If this is right

If correct, streaming world models become trainable on modest hardware, so researchers can study novel memory and conditioning mechanisms rather than only post-training large generators.
The KV cache yields a 2.2x throughput gain and 15.2x lower first-chunk latency, making real-time interactive world-model applications feasible.
The same architecture transfers across robot-action and camera-pose conditioning, suggesting a single recipe for embodied and scene-generation benchmarks.
Scaling experiments indicate that extra capacity mainly strengthens dynamics and 3D geometry rather than appearance, directing future scaling efforts.
Quality stays roughly flat when the retained history shrinks from 24 to 6 chunks at 64-frame horizons, implying the active window and sink anchor carry most of the memory load.

Reading between the lines

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

The frozen VAE's 16x spatial and 4x temporal compression is the representation bottleneck: if it discards fine-grained contact or sub-pixel camera motion, no training can recover them. Re-running the recipe with a higher-fidelity VAE would test whether trajectory accuracy improves.
The non-decreasing timestep constraint is presented as fixed; its exact contribution is not isolated in the ablations, so a future study could ablate it independently to see how much of the stability comes from the monotonicity itself.
The 'stable long-horizon' claim rests on rollouts up to 381 frames, where geometry and fidelity already degrade; extending to much longer horizons would likely require memory compression beyond the fixed-capacity KV cache.
The benchmarks contain only 50 held-out videos each; running the released pipeline on a new dataset or action distribution would establish whether the recipe generalizes independent of the chosen benchmarks.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

This is a useful, well-executed systems contribution. What's actually new is the complete package: block-causal Video DiT, chunk-wise non-decreasing noise schedule (CoPP), two-stage continued training, and rolling KV cache with pipelined denoising, all integrated and released as code and checkpoints. The method sections are coherent, the flow-matching objective is standard, and the throughput numbers (2.2x speedup, 15x latency reduction) are concrete and reproducible. The authors also deserve credit for being honest about limitations; they frame MiniWorld as a baseline, not a SOTA system, and their ablations on cache size and sink frames are thoughtful.

The soft spots are real but not fatal. The evaluation compares only against their own bidirectional sliding-window baseline, normalized to 1, with no error bars and no external baselines. The rollouts are 253–381 frames on 50 videos, which is modest for a 'long-horizon' claim. Those large relative gains should therefore be read as evidence that the streaming formulation helps within this fixed setup, not as a general superiority claim.

The bigger concern, which the reader and stress-test both flag, is the frozen Wan2.2 VAE with 4x temporal compression. The paper never reports reconstruction fidelity on DROID or RE10K, so it doesn't check whether fine-grained gripper or camera dynamics survive the latent bottleneck. That's a fair criticism, though it is a shared limitation of latent video world models rather than a specific flaw in MiniWorld's training recipe. For the 'action-conditioned world modeling' claim, a simple probe or reconstruction metric would close the gap. I wouldn't call it load-bearing for the paper's main purpose—a reproducible baseline—but it should be addressed in revision.

The math and the released artifacts look solid. The authors adapt published components in a straightforward way, and the citation pattern is appropriate. This is the kind of paper that helps the community: it gives researchers with modest compute a real starting point. It deserves a serious referee, and the main revision requests should be external baselines, uncertainty reporting, and a VAE fidelity check. I would cite it in my own work and bring it to a reading group focused on practical world model training.

Referee Report

4 major / 4 minor

Summary. MiniWorld proposes a from-scratch training recipe for streaming video world models using a block-causal Video Diffusion Transformer trained with flow matching in a frozen Wan2.2 VAE latent space. The method introduces a chunk-wise non-decreasing noise schedule (CoPP), two-stage continued training with timestep shifting, and a streaming inference pipeline with rolling KV cache, RoPE re-shifting, and pipelined asynchronous denoising. The authors claim the full model can be trained on a single 8-GPU server in several days and demonstrate stable long-horizon action-conditioned generation on DROID and RE10K, reporting large relative gains over a self-implemented bidirectional short-video baseline on a range of appearance, dynamics, geometry, VLM, and fidelity metrics.

Significance. If the empirical claims hold, MiniWorld would provide a valuable, accessible, and fully reproducible baseline for streaming world models, addressing a real gap in the community. The approach is methodologically coherent: flow matching on prediction chunks with non-decreasing noise schedules is a reasonable instantiation of Diffusion Forcing, and the streaming inference design with rolling KV cache is practical. The release of code and checkpoints is a notable strength that will facilitate future work. However, the current empirical validation is not yet sufficient to fully establish the central claims: the reliance on an unvalidated frozen VAE representation, the relative-only comparison against a self-implemented baseline, and the lack of absolute metrics leave the actual world-modeling capability and the democratization claim under-supported.

major comments (4)

[§2, §4.1] All experiments operate in the latent space of the frozen Wan2.2 VAE (4× temporal, 16× spatial, 48 channels). The paper never validates that this latent representation preserves the action-sensitive information (e.g., gripper-object contacts, small end-effector displacements, camera translation) that the world model is supposed to predict. A reconstruction-fidelity check on DROID/RE10K (PSNR/SSIM/LPIPS over decoded vs. original frames), or a probe showing actions are decodable from latent chunks, is required to substantiate the claim of action-conditioned world modeling. Without it, the central result is contingent on an unverified representation bottleneck.
[§4.1, Figure 2] All headline gains are reported as relative scores with the authors' own bidirectional short-video baseline normalized to 1, and error metrics are inverted. This makes the magnitudes uninterpretable (e.g., Trajectory Accuracy +249% could be a change from 0.2 to 0.7 or 0.01 to 0.03) and precludes comparison with any existing baseline. Report absolute values with confidence intervals, and include at least one external baseline or an ablation with identical inference settings (CFG, sampler, window).
[§4.1, Figure 2] It is unclear whether the bidirectional baseline uses classifier-free guidance and, if so, at what scale. MiniWorld uses CFG scale 2 in the main results. The CFG ablation shows a ~5.5% average improvement, but the headline gains are far larger; to rule out inference-settings confounds, the baseline must be evaluated under the same CFG and sampling configuration.
[§1, §4.1] The central 'several days on a single 8-GPU server' claim is not quantified. The paper should report GPU type (e.g., A100/H100), total training wall-clock time per model size, and the number of training steps/samples seen. Without this, the democratization claim is not reproducible or verifiable.

minor comments (4)

[§4.3] The 253 vs 381 frame comparison shows degradation in Geometry, VLM, and Fidelity metrics; the abstract's 'stable long-horizon' should be qualified with the exact rollout horizon used (253 frames) and the observed degradation at longer horizons.
[§3.5] The notation Δτ_m^s ≤ 0 is confusing; state explicitly that integrating from τ=1 to τ=0 yields negative timestep increments, so the update corresponds to a standard backward Euler step.
[§4.1] For action normalization in Eq. (8), clarify whether the percentile statistics q_0.01 and q_0.99 are computed over the entire dataset or per episode; this affects reproducibility.
[§3.3] The CoPP scheduler description would benefit from a precise algorithmic definition (e.g., pseudocode) of how the anchor chunk and outward propagation are implemented, including boundary handling.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all load-bearing components are credited to external prior work, and the central claims rest on empirical comparison, not on a fitted parameter or self-citation chain.

full rationale

I walked the paper's derivation chain looking for predictions that reduce by construction to their inputs, fitted parameters renamed as predictions, or load-bearing self-citations. None are present. The core technical components—block-causal attention, Diffusion Forcing-style chunk-wise noise schedules, non-decreasing timestep constraints, timestep shifting, the Muon optimizer, and the Wan2.2 VAE—are all explicitly attributed to external prior work (e.g., Diffusion Forcing [2], AR-Diffusion [27], Esser et al. [5], Wan [29]), with no overlapping authors in this paper's reference list. The CoPP scheduler is presented as a chunk-wise generalization of AR-Diffusion's FoPP, clearly cited rather than repackaged as a new first-principles result. The evaluation compares MiniWorld against an internally implemented bidirectional sliding-window baseline, but that is a comparison choice, not a fitted input: no parameter is tuned to the evaluation set, and the datasets (DROID, RE10K) are external benchmarks. The claimed improvements are relative to that baseline and are empirical measurements, not quantities that are equal to the training objective by construction. The paper also explicitly acknowledges its limitations and frames itself as a reproducible baseline rather than a performance ceiling, which further mitigates any concern of hiding a circular argument. Therefore, the central claims are self-contained and any weaknesses are matters of external validity or reproducibility, not circularity.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central recipe rests mainly on standard diffusion and flow-matching machinery plus the frozen VAE and the control-conditioning interface. No physical entities are introduced. The listed hyperparameters are design choices rather than fitted constants, but they materially determine the reported results; the listed assumptions are the places where the method could silently fail.

free parameters (7)

Latent chunk size = 2 latent frames
Chosen by hand; controls granularity of causal attention and diffusion-forcing schedule; all training and inference numbers depend on it.
Noise schedule discreteness (bins) = 50
Training-time diffusion timestep bins with monotonic CoPP compositions; changing this shifts the denoising trajectory distribution.
Logit-normal timestep sampling (Pmean, Pstd) = 0, 1
Default from SD3 and AR-Diffusion; changes which noise levels are seen most often during training.
CFG scale = 2.0
Used for quality evaluation; chosen without a reported sweep; inflates or deflates perceptual versus fidelity scores.
Condition dropout probability = 0.10
Enables classifier-free guidance; standard but tuned.
Two-stage training lengths = 21/46 then 125/253 frames
Central to the long-horizon claim; if longer clips were unnecessary or insufficient, the recipe changes.
KV cache capacity and in-flight window = 8 in-flight chunks, up to 24 retained chunks
Throughput and quality trade-off; experiments show 6-24 chunks mostly flat, but the default is a design choice.

assumptions (5)

domain assumption Wan2.2 VAE latent space preserves the information needed for action-conditioned world dynamics at 4x temporal and 16x spatial compression
Modeling is done entirely in the latent space (Section 2); if the VAE drops action-sensitive detail, the downstream model cannot recover it.
standard math Rectified flow linear interpolation and its velocity objective provide a valid generative training signal for video latents
Borrowed from Flow Matching and Rectified Flow; no new proof; accepted background.
domain assumption DROID 7-D actions and RE10K camera poses are sufficient controls, and four-frame action concatenation correctly aligns with the 4x temporal VAE compression
Section 4.1; if action semantics are misaligned or under-specified, the conditioning signal is wrong.
domain assumption The WorldArena-style metrics used for evaluation are valid proxies for world-model quality
Section 4.1; no independent validation of these metrics is provided.
domain assumption Chunk-wise non-decreasing noise schedules learned under CoPP generalize to asynchronous streaming inference schedules
The training sampler covers non-decreasing compositions via CoPP, but the specific inference schedule with pipelined denoising is a deployment-time approximation rather than a proven match.

how reviews work

0 comments

Cite this review

Pith. "Pith review of MiniWorld: Democratizing the Training of Video World Models from Scratch." pith.science (2026). https://pith.science/paper/YXXAJAEF

@misc{pith2026260801127,
  author       = {Pith},
  title        = {Pith review of: MiniWorld: Democratizing the Training of Video World Models from Scratch},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YXXAJAEF}},
  note         = {Machine review of arXiv:2608.01127}
}

read the original abstract

Video world models predict future observations conditioned on historical observations and control signals, enabling long-horizon generation through autoregressive state transitions. Unlike conventional video generation models that primarily capture visual appearance and motion, video world models learn the underlying dynamics governing environment evolution under agent actions, providing a foundation for embodied AI and interactive simulation. Recent progress has largely relied on adapting pretrained video generation models through post-training or distillation. Although effective, these approaches often require complex training pipelines, substantial computational resources, and suffer from the mismatch between bidirectional pretraining and causal streaming inference. Recent studies have shown that training autoregressive video world models from scratch is feasible and scalable. However, the community still lacks a lightweight, transparent, and fully reproducible baseline trainable end-to-end with modest computational resources. We present MiniWorld, a reproducible framework for training streaming video world models from scratch. MiniWorld employs a block-causal Video Diffusion Transformer trained with Flow Matching in the latent space of a pretrained Video VAE. Building on Diffusion Forcing, it adopts a chunk-wise non-decreasing noise schedule and two-stage continued training to improve temporal modeling and stability. During inference, MiniWorld combines a rolling KV cache with pipelined asynchronous denoising for efficient streaming generation under bounded computation. The entire model can be trained within several days on a single 8-GPU server. By releasing the training and inference codebase and pretrained checkpoints, we hope MiniWorld will facilitate future research on video world modeling.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 12 canonical work pages

[1]

Happyoyster: Real-time interactive world model, 2026

Alibaba Token Hub. Happyoyster: Real-time interactive world model, 2026

work page 2026
[2]

Diffusion forcing: Next-token prediction meets full-sequence diffusion

Boyuan Chen et al. Diffusion forcing: Next-token prediction meets full-sequence diffusion. In Advances in Neural Information Processing Systems, 2024. URL https://arxiv.org/abs/2407. 01392

work page 2024
[3]

Dreamx-world 1.0: A general-purpose interactive world model, 2026

DreamX Team. Dreamx-world 1.0: A general-purpose interactive world model, 2026. URL https://arxiv.org/abs/2606.16993

arXiv 2026
[4]

Mirage 2: Ai-native ugc game engine powered by real-time world models, 2025

Dynamics Lab. Mirage 2: Ai-native ugc game engine powered by real-time world models, 2025. URLhttps://blog.dynamicslab.ai/

work page 2025
[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. InInternational Conference on Machine Learning, 2024. URL https://arxiv.org/...

arXiv 2024
[6]

Ca2-vdm: Efficient autoregressive video diffusion model with causal generation and cache sharing

Peng Gao et al. Ca2-vdm: Efficient autoregressive video diffusion model with causal generation and cache sharing. InInternational Conference on Machine Learning, 2025. URL https: //arxiv.org/abs/2411.16375

arXiv 2025
[7]

Dreamdojo: A generalist robot world model from large-scale human videos,

Sheng Gao et al. Dreamdojo: A generalist robot world model from large-scale human videos,

work page
[8]

Lingbot-world 2.0: Infinite worlds with versatile interactions, 2026

Zhen Gao et al. Lingbot-world 2.0: Infinite worlds with versatile interactions, 2026. URL https://arxiv.org/abs/2607.07534

arXiv 2026

Show all 39 references

[9]

Genie 3: A new frontier for world models, 2025

Google DeepMind. Genie 3: A new frontier for world models, 2025. URLhttps://deepmind. google/blog/genie-3-a-new-frontier-for-world-models/

2025
[10]

Recurrent world models facilitate policy evolution

David Ha and Jürgen Schmidhuber. Recurrent world models facilitate policy evolution. In Advances in Neural Information Processing Systems, 2018. URL https://arxiv.org/abs/1809. 01999

2018
[11]

Mastering diverse domains through world models, 2023

Danijar Hafner, Jurgis Pasukonis, Jimmy Ba, and Timothy Lillicrap. Mastering diverse domains through world models, 2023. URLhttps://arxiv.org/abs/2301.04104

2023 arXiv
[12]

Matrix-game 2.0: An open-source, real-time, and streaming interactive world model, 2025

Xianglong He et al. Matrix-game 2.0: An open-source, real-time, and streaming interactive world model, 2025. URLhttps://arxiv.org/abs/2508.13009

2025 arXiv
[13]

Self forcing: Bridging the train-test gap in autoregressive video diffusion

Xiaoyu Huang, Zhen Li, Guande He, Mingyuan Zhou, and Eli Shechtman. Self forcing: Bridging the train-test gap in autoregressive video diffusion. InAdvances in Neural Information Processing Systems, 2025. URLhttps://arxiv.org/abs/2506.08009

2025 arXiv
[14]

Muon: An optimizer for the hidden layers of neural networks, 2024

Keller Jordan et al. Muon: An optimizer for the hidden layers of neural networks, 2024. URL https://kellerjordan.github.io/posts/muon/

2024
[15]

Droid: A large-scale in-the-wild robot manipulation dataset

Alexander Khazatsky et al. Droid: A large-scale in-the-wild robot manipulation dataset. In Robotics: Science and Systems, 2024. URLhttps://arxiv.org/abs/2403.12945. 17

2024 arXiv
[16]

Advancing open-source world models, 2026

LingBot Team. Advancing open-source world models, 2026. URLhttps://arxiv.org/abs/ 2601.20540

2026 arXiv
[17]

Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow matching for generative modeling. InInternational Conference on Learning Representations,
[18]

Rolling forcing: Autoregressive long video diffusion in real time, 2025

Hao Liu et al. Rolling forcing: Autoregressive long video diffusion in real time, 2025. URL https://arxiv.org/abs/2509.25161

2025 arXiv
[19]

Flow straight and fast: Learning to generate and transfer data with rectified flow

Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InInternational Conference on Learning Representations,
[20]

Yume-1.5: A text-controlled interactive world generation model

Xiaofeng Mao, Zhengkai Li, Chuanhao Li, Xiaoyan Xu, Kaipeng Ying, Tong He, Jiangmiao Pang, Yu Qiao, and Kaipeng Zhang. Yume-1.5: A text-controlled interactive world generation model. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,
[21]

Cosmos world foundation model platform for physical ai, 2025

NVIDIA et al. Cosmos world foundation model platform for physical ai, 2025. URLhttps: //arxiv.org/abs/2501.03575

2025 arXiv
[22]

URLhttps://arxiv.org/abs/2209.03003

arXiv
[23]

Worldarena: A unified benchmark for evaluating perception and functional utility of embodied world models, 2026

Jingkang Shang et al. Worldarena: A unified benchmark for evaluating perception and functional utility of embodied world models, 2026

2026
[24]

URLhttps://arxiv.org/abs/2512.22096
[25]

Skyreels-v2: Infinite-length film generative model, 2025

SkyworkAI Team. Skyreels-v2: Infinite-length film generative model, 2025. URL https: //arxiv.org/abs/2504.13074

2025 arXiv
[26]

Magi-1: Autoregressive video generation at scale, 2025

SandAI Team. Magi-1: Autoregressive video generation at scale, 2025. URLhttps://arxiv. org/abs/2505.13211

2025 arXiv
[27]

Ar-diffusion: Asynchronous video generation with auto-regressive diffusion

Peize Sun et al. Ar-diffusion: Asynchronous video generation with auto-regressive diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025. URLhttps://arxiv.org/abs/2503.07418

2025 arXiv
[28]

Matrix-game 3.0: Real-time and streaming interactive world model with long-horizon memory, 2026

Skywork AI Matrix-Game Team. Matrix-game 3.0: Real-time and streaming interactive world model with long-horizon memory, 2026. URLhttps://arxiv.org/abs/2604.08995

2026 arXiv
[29]

Wan: Open and advanced large-scale video generative models, 2025

Wan Team. Wan: Open and advanced large-scale video generative models, 2025. URL https://arxiv.org/abs/2503.20314

2025 arXiv
[30]

History-guided video diffusion

Yuda Song et al. History-guided video diffusion. InInternational Conference on Machine Learning, 2025. URLhttps://arxiv.org/abs/2502.06764

2025 arXiv
[31]

Geometry forcing: Marrying video diffusion and 3d representation for consistent world modeling

Haofei Wu, Dong Wu, Tong He, Jing Guo, Yufei Ye, Yilun Duan, and Jiang Bian. Geometry forcing: Marrying video diffusion and 3d representation for consistent world modeling. In 18 International Conference on Learning Representations, 2026. URL https://arxiv.org/abs/ 2507.07982

2026 arXiv
[32]

Hy-world 1.5: A systematic framework for interactive world modeling with real-time latency and geometric consistency, 2025

Tencent Hunyuan World Team. Hy-world 1.5: A systematic framework for interactive world modeling with real-time latency and geometric consistency, 2025. URLhttps://3d-models. hunyuan.tencent.com/world/world1_5/HYWorld_1.5_Tech_Report.pdf

2025
[33]

Efros, Eli Shechtman, and Oliver Wang

Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018. URL https://arxiv.org/abs/1801.03924

2018 arXiv
[34]

Bovik, Hamid R

Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli. Image quality assessment: From error visibility to structural similarity.IEEE Transactions on Image Processing, 2004

2004
[35]

Causal forcing: Autoregres- sive diffusion distillation done right for high-quality real-time interactive video generation

Hanyu Zhu, Ming Zhao, Guande He, Hao Su, Cheng Li, and Jun Zhu. Causal forcing: Autoregres- sive diffusion distillation done right for high-quality real-time interactive video generation. InIn- ternational Conference on Machine Learning, 2026. URL https://arxiv.org/abs/2602.02214. 19

2026 arXiv
[36]

From slow bidirectional to fast autoregressive video diffusion models

Tianwei Yin et al. From slow bidirectional to fast autoregressive video diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025. URLhttps://arxiv.org/abs/2412.07772

2025
[38]

Stereo magnifica- tion: Learning view synthesis using multiplane images.ACM Transactions on Graphics, 2018

Tinghui Zhou, Richard Tucker, John Flynn, Graham Fyffe, and Noah Snavely. Stereo magnifica- tion: Learning view synthesis using multiplane images.ACM Transactions on Graphics, 2018. URLhttps://arxiv.org/abs/1805.09817

2018 arXiv
[2023]

URLhttps://arxiv.org/abs/2210.02747

arXiv
[2026]

URLhttps://arxiv.org/abs/2602.06949

arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Happyoyster: Real-time interactive world model, 2026

Alibaba Token Hub. Happyoyster: Real-time interactive world model, 2026

work page 2026

[2] [2]

Diffusion forcing: Next-token prediction meets full-sequence diffusion

Boyuan Chen et al. Diffusion forcing: Next-token prediction meets full-sequence diffusion. In Advances in Neural Information Processing Systems, 2024. URL https://arxiv.org/abs/2407. 01392

work page 2024

[3] [3]

Dreamx-world 1.0: A general-purpose interactive world model, 2026

DreamX Team. Dreamx-world 1.0: A general-purpose interactive world model, 2026. URL https://arxiv.org/abs/2606.16993

arXiv 2026

[4] [4]

Mirage 2: Ai-native ugc game engine powered by real-time world models, 2025

Dynamics Lab. Mirage 2: Ai-native ugc game engine powered by real-time world models, 2025. URLhttps://blog.dynamicslab.ai/

work page 2025

[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. InInternational Conference on Machine Learning, 2024. URL https://arxiv.org/...

arXiv 2024

[6] [6]

Ca2-vdm: Efficient autoregressive video diffusion model with causal generation and cache sharing

Peng Gao et al. Ca2-vdm: Efficient autoregressive video diffusion model with causal generation and cache sharing. InInternational Conference on Machine Learning, 2025. URL https: //arxiv.org/abs/2411.16375

arXiv 2025

[7] [7]

Dreamdojo: A generalist robot world model from large-scale human videos,

Sheng Gao et al. Dreamdojo: A generalist robot world model from large-scale human videos,

work page

[8] [8]

Lingbot-world 2.0: Infinite worlds with versatile interactions, 2026

Zhen Gao et al. Lingbot-world 2.0: Infinite worlds with versatile interactions, 2026. URL https://arxiv.org/abs/2607.07534

arXiv 2026