pith. machine review for the scientific record. sign in

arxiv: 2605.06051 · v1 · submitted 2026-05-07 · 💻 cs.CV

Recognition: unknown

RealCam: Real-Time Novel-View Video Generation with Interactive Camera Control

Youcan Xu , Jiaxin Shi , Zhen Wang , Wensong Song , Feifei Shao , Chen Liang , Jun Xiao , Long Chen
Authors on Pith no claims yet

Pith reviewed 2026-05-09 15:17 UTC · model grok-4.3

classification 💻 cs.CV
keywords real-time video generationnovel view synthesiscamera controlautoregressive modelknowledge distillationvideo-to-video generationtemporal consistencyloop consistency
0
0 comments X

The pith

RealCam turns slow non-causal video generation into real-time interactive camera-controlled synthesis by distilling a cross-frame teacher into a fast causal student.

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

Existing camera-controlled video-to-video methods depend on full-sequence bidirectional attention, which creates high latency and prevents real-time streaming or variable inputs. RealCam builds a teacher model that interleaves source and target frames in contextual pairs to support length-agnostic causal processing. It then applies self-forcing and distribution matching distillation to produce a few-step causal student capable of on-the-fly generation. Loop-closed data augmentation creates consistent training sequences for closed trajectories. If these steps hold, interactive high-fidelity novel-view video synthesis becomes practical for applications that need live viewpoint changes.

Core claim

The paper establishes that an autoregressive framework grounded in cross-frame in-context learning can be distilled via self-forcing and distribution matching distillation into a few-step causal model, augmented by loop-closed data augmentation, to deliver state-of-the-art visual fidelity and temporal consistency with orders-of-magnitude faster inference than prior non-causal methods for interactive camera-controlled video generation.

What carries the argument

The Cross-frame In-context Learning paradigm, which interleaves source and target frames into synchronized contextual pairs to enable causal, length-agnostic processing, then distilled into a student model.

If this is right

  • Enables streaming synthesis for arbitrary-length inputs without quadratic scaling.
  • Supports truly interactive camera control during generation at real-time speeds.
  • Maintains higher temporal consistency than rigid prefix-concatenation approaches.
  • Allows deployment in live broadcasting or interactive filmmaking scenarios.
  • Generalizes across variable-length monocular videos without retraining.

Where Pith is reading between the lines

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

  • The same teacher-to-causal-student distillation pattern could transfer to other control signals in video generation beyond camera poses.
  • Real-time performance opens testing in augmented-reality pipelines where viewpoint must update continuously.
  • LoopAug-style augmentation might help consistency in other cyclic generation tasks such as periodic motion synthesis.
  • If the student scales to longer horizons, it could reduce the need for separate keyframe planning stages in video pipelines.

Load-bearing premise

The high-fidelity cross-frame teacher can be distilled into a few-step causal student without substantial loss of quality, and loop-closed data augmentation is enough to eliminate inconsistencies on closed trajectories.

What would settle it

Measure the student model's output against the teacher on a held-out set of closed-loop camera trajectories; if visual fidelity drops markedly or loop seams remain visible, the central claim fails.

Figures

Figures reproduced from arXiv: 2605.06051 by Chen Liang, Feifei Shao, Jiaxin Shi, Jun Xiao, Long Chen, Wensong Song, Youcan Xu, Zhen Wang.

Figure 1
Figure 1. Figure 1: Comparison of direct temporal concatenation and ours cross-frame concatenation. Our method generalizes to arbitrary video length during inference and naturally extends to causal attention. (V2V) generative model [Bai et al. 2025; Chen et al. 2025; Luo et al. 2025; Park et al. 2025b; Van Hoorick et al. 2024; Yu et al. 2025a], have emerged as a transformative post-production paradigm. By synthesizing novel v… view at source ↗
Figure 2
Figure 2. Figure 2: Model architecture and training pipeline. Left: The training pipeline of teacher camera-controlled video-to-video model. (a) A latent diffusion model is optimized to reconstruct the target video 𝑉𝑡𝑔𝑡 , conditioned on the source video 𝑉𝑠𝑟𝑐 , target camera pose 𝑐cam, and target prompt 𝑐text. (b) We propose Cross-frame Concatenation to inject the source video condition, and following ReCamMaster [Bai et al. 2… view at source ↗
Figure 3
Figure 3. Figure 3: Length-agnostic inference without retraining. Each row corresponds to a distinct video sequence synthesized at different inference lengths (e.g., 49 denotes the total frames of the generated video). Both trained on fixed 81-frame clips, our model maintains high performance. In contrast, the direct concatenation method (e.g., ReCamMaster [Bai et al. 2025]) exhibits severe degradation when the inference hori… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison with SOTA methods. Red boxes indicated low quality content across frames. Our method achieves better camera control and excellent temporal synchronization view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative ablation on long video. The camera trajectory first translates down with rotation and then back to the origin. Red boxes indicated inconsistency with the source video view at source ↗
read the original abstract

Camera-controlled video-to-video (V2V) generation enables dynamic viewpoint synthesis from monocular footage, holding immense potential for interactive filmmaking and live broadcasting. However, existing implicit synthesis methods fundamentally rely on non-causal, full-sequence processing and rigid prefix-style temporal concatenation. This architectural paradigm mandates bidirectional attention, resulting in prohibitive computational latency, quadratic complexity scaling, and inherent incompatibility with real-time streaming or variable-length inputs. To overcome these limitations, we introduce \texttt{RealCam}, a novel autoregressive framework for interactive, real-time camera-controlled V2V generation. We first design a high-fidelity teacher model grounded in a \textbf{Cross-frame In-context Learning} paradigm. By interleaving source and target frames into synchronized contextual pairs, our design inherently enables length-agnostic generalization and naturally facilitates causal adaptation, breaking the rigid prefix bottleneck. We then distill this teacher into a few-step causal student via Self-Forcing with Distribution Matching Distillation, enabling efficient, on-the-fly streaming synthesis. Furthermore, to mitigate severe loop inconsistency in closed-loop trajectories, we propose \textbf{Loop-Closed Data Augmentation (LoopAug)}, a novel paradigm that synthesizes globally consistent loop sequences from existing multiview datasets. Extensive experiments demonstrate that \texttt{RealCam} achieves state-of-the-art visual fidelity and temporal consistency while enabling truly interactive camera control with orders-of-magnitude faster inference than existing paradigms. Our project page is at https://xyc-fly.github.io/RealCam/.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The manuscript introduces RealCam, an autoregressive framework for real-time camera-controlled video-to-video generation from monocular input. It first trains a high-fidelity teacher using Cross-frame In-context Learning by interleaving source and target frames into contextual pairs, then distills this into a few-step causal student via Self-Forcing and Distribution Matching Distillation for streaming inference. To handle closed-loop trajectories, it proposes Loop-Closed Data Augmentation (LoopAug) that synthesizes consistent loop sequences from multiview datasets. The central claim is that RealCam simultaneously achieves state-of-the-art visual fidelity and temporal consistency while delivering orders-of-magnitude faster inference than prior non-causal methods.

Significance. If the distillation preserves teacher quality and LoopAug demonstrably resolves loop inconsistency, the work would be a notable contribution to real-time generative video models. Enabling interactive, causal, length-agnostic novel-view synthesis at interactive speeds could impact applications in live broadcasting and virtual production. The explicit use of distillation for causal adaptation and the LoopAug data synthesis strategy are technically interesting and could be adopted more broadly if supported by evidence.

major comments (3)
  1. [§3.2] §3.2 (Distillation procedure): The central efficiency claim rests on distilling the Cross-frame teacher into a few-step causal student 'without substantial quality loss,' yet no teacher-vs-student metric deltas (e.g., on novel-view accuracy, temporal coherence, or FID) or ablations on distillation steps are reported. This directly undermines the assertion that RealCam retains SOTA fidelity while achieving real-time performance.
  2. [§4.3] §4.3 (LoopAug evaluation): The paper states that LoopAug 'sufficiently resolves loop inconsistency for closed trajectories,' but provides no quantitative before/after consistency metrics or comparisons on closed-loop test sequences. Without these, the claim that the framework handles variable-length and closed inputs reliably cannot be assessed.
  3. [Abstract and §4] Abstract and §4 (Experiments): The abstract asserts 'state-of-the-art visual fidelity and temporal consistency' plus 'orders-of-magnitude faster inference' with no numerical results, baseline comparisons, or table references. This absence of concrete evidence makes the headline claims unverifiable from the manuscript as presented.
minor comments (2)
  1. [§3] The free parameters (few-step count, LoopAug synthesis parameters) are listed but never given concrete values or ranges used in the reported experiments; adding these would improve reproducibility.
  2. [§3.2] Notation for the causal student (e.g., how 'Self-Forcing' differs from standard teacher-forcing) could be clarified with a short equation or pseudocode block.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. The comments highlight areas where additional quantitative evidence would strengthen the manuscript. We address each major comment below and will incorporate the requested additions and clarifications in the revised version.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Distillation procedure): The central efficiency claim rests on distilling the Cross-frame teacher into a few-step causal student 'without substantial quality loss,' yet no teacher-vs-student metric deltas (e.g., on novel-view accuracy, temporal coherence, or FID) or ablations on distillation steps are reported. This directly undermines the assertion that RealCam retains SOTA fidelity while achieving real-time performance.

    Authors: We agree that explicit teacher-student comparisons are essential to validate the distillation. In the revised manuscript we will add a new ablation subsection (or expand §3.2) that reports quantitative deltas on FID, novel-view accuracy (PSNR/SSIM), and temporal coherence (FVD and frame-wise CLIP similarity) between the full teacher and the distilled student at 1-, 2-, and 4-step settings. We will also include an ablation table varying the number of distillation steps to show the quality-efficiency trade-off. revision: yes

  2. Referee: [§4.3] §4.3 (LoopAug evaluation): The paper states that LoopAug 'sufficiently resolves loop inconsistency for closed trajectories,' but provides no quantitative before/after consistency metrics or comparisons on closed-loop test sequences. Without these, the claim that the framework handles variable-length and closed inputs reliably cannot be assessed.

    Authors: We acknowledge that the current manuscript lacks direct quantitative validation of LoopAug. In the revision we will add a dedicated evaluation in §4.3 that measures loop-closure consistency (pose drift error and perceptual loop-closure score) on synthesized closed trajectories before and after LoopAug, together with comparisons against training without the augmentation. These results will be presented in a new table and accompanying figure. revision: yes

  3. Referee: [Abstract and §4] Abstract and §4 (Experiments): The abstract asserts 'state-of-the-art visual fidelity and temporal consistency' plus 'orders-of-magnitude faster inference' with no numerical results, baseline comparisons, or table references. This absence of concrete evidence makes the headline claims unverifiable from the manuscript as presented.

    Authors: While abstracts conventionally summarize high-level claims, we accept that the current version would benefit from concrete anchors. We will revise the abstract to include key numerical highlights (e.g., specific FID, temporal consistency, and speedup factors) with explicit references to the corresponding tables in §4. In addition, we will ensure that the experimental section explicitly states all baseline comparisons with numerical values and clear in-text table citations. revision: yes

Circularity Check

0 steps flagged

No significant circularity in RealCam derivation chain

full rationale

The paper's core claims rest on introducing a Cross-frame In-context Learning teacher, distilling it via Self-Forcing and Distribution Matching Distillation into a causal student, and applying LoopAug for closed trajectories. These are framed as new architectural and training choices whose fidelity, consistency, and speed advantages are asserted via external experiments and comparisons to prior paradigms. No equations or definitions reduce a claimed output to an input by construction, no fitted parameters are relabeled as predictions, and no load-bearing premises collapse to self-citations or author-imported uniqueness theorems. The framework is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 1 invented entities

The central claims rest on the effectiveness of the new teacher paradigm and distillation process plus the utility of the invented LoopAug technique; these are introduced without external benchmarks in the abstract.

free parameters (2)
  • few-step count in student model
    Number of denoising steps chosen for the causal student; value not stated but required for the efficiency claim.
  • LoopAug synthesis parameters
    Parameters controlling how multiview data is turned into globally consistent loops; tuned to mitigate inconsistency.
axioms (1)
  • domain assumption Interleaving source and target frames into contextual pairs enables length-agnostic causal generalization.
    Invoked to justify breaking the prefix bottleneck in the teacher model.
invented entities (1)
  • LoopAug no independent evidence
    purpose: Synthesize globally consistent loop sequences from multiview datasets to reduce closed-loop inconsistency.
    New data augmentation paradigm introduced to address a specific failure mode.

pith-pipeline@v0.9.0 · 5586 in / 1237 out tokens · 51135 ms · 2026-05-09T15:17:39.645158+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

57 extracted references · 26 canonical work pages · 7 internal anchors

  1. [1]

    European Conference on Computer Vision (ECCV) , year=

    Generative Camera Dolly: Extreme Monocular Dynamic Novel View Synthesis , author=. European Conference on Computer Vision (ECCV) , year=

  2. [2]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Bai, Jianhong and Xia, Menghan and Fu, Xiao and Wang, Xintao and Mu, Lianrui and Cao, Jinwen and Liu, Zuozhu and Hu, Haoji and Bai, Xiang and Wan, Pengfei and Zhang, Di , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2025 , pages =

    2025

  3. [3]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Yu, Mark and Hu, Wenbo and Xing, Jinbo and Shan, Ying , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2025 , pages =

    2025

  4. [4]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    Reconstruct, Inpaint, Test-Time Finetune: Dynamic Novel-view Synthesis from Monocular Videos , author =. Advances in Neural Information Processing Systems (NeurIPS) , year =

  5. [5]

    2025 , eprint=

    ReDirector: Creating Any-Length Video Retakes with Rotary Camera Encoding , author=. 2025 , eprint=

    2025

  6. [6]

    and Pritch, Yael and Mosseri, Inbar and Shou, Mike Zheng and Wadhwa, Neal and Ruiz, Nataniel , booktitle=

    Zhang, David Junhao and Paiss, Roni and Zada, Shiran and Karnad, Nikhil and Jacobs, David E. and Pritch, Yael and Mosseri, Inbar and Shou, Mike Zheng and Wadhwa, Neal and Ruiz, Nataniel , booktitle=. ReCapture: Generative Video Camera Controls for User-Provided Videos using Masked Video Fine-Tuning , year=

  7. [7]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Reangle-a-video: 4d video generation as video-to-video translation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  8. [8]

    arXiv preprint arXiv:2510.26796 , year=

    SEE4D: Pose-Free 4D Generation via Auto-Regressive Video Inpainting , author=. arXiv preprint arXiv:2510.26796 , year=

    work page arXiv

  9. [9]

    The Thirteenth International Conference on Learning Representations,

    Zeqi Xiao and Wenqi Ouyang and Yifan Zhou and Shuai Yang and Lei Yang and Jianlou Si and Xingang Pan , title =. The Thirteenth International Conference on Learning Representations,. 2025 , url =

    2025

  10. [10]

    Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    Camclonemaster: Enabling reference-based camera control for video generation , author=. Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    2025

  11. [11]

    arXiv preprint arXiv:2511.01266 (2025)

    Motionstream: Real-time video generation with interactive motion controls , author=. arXiv preprint arXiv:2511.01266 , year=

    work page arXiv

  12. [12]

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

    Self forcing: Bridging the train-test gap in autoregressive video diffusion , author=. arXiv preprint arXiv:2506.08009 , year=

    work page internal anchor Pith review arXiv

  13. [13]

    arXiv preprint arXiv:2510.08131 (2025) 10 MuSteerNet 19

    Real-Time Motion-Controllable Autoregressive Video Diffusion , author=. arXiv preprint arXiv:2510.08131 , year=

    work page arXiv

  14. [14]

    RoFormer: Enhanced Transformer with Rotary Position Embedding,

    Jianlin Su and Murtadha Ahmed and Yu Lu and Shengfeng Pan and Wen Bo and Yunfeng Liu , keywords =. RoFormer: Enhanced transformer with Rotary Position Embedding , journal =. 2024 , issn =. doi:https://doi.org/10.1016/j.neucom.2023.127063 , url =

    work page doi:10.1016/j.neucom.2023.127063 2024

  15. [15]

    arXiv preprint arXiv:2507.16869 (2025)

    Controllable video generation: A survey , author=. arXiv preprint arXiv:2507.16869 , year=

    work page arXiv

  16. [16]

    CameraCtrl: Enabling Camera Control for Text-to-Video Generation

    Cameractrl: Enabling camera control for text-to-video generation , author=. arXiv preprint arXiv:2404.02101 , year=

    work page internal anchor Pith review arXiv

  17. [17]

    ACM SIGGRAPH 2024 Conference Papers , pages=

    Motionctrl: A unified and flexible motion controller for video generation , author=. ACM SIGGRAPH 2024 Conference Papers , pages=

    2024

  18. [18]

    Advances in Neural Information Processing Systems , volume=

    Collaborative video diffusion: Consistent multi-video generation with camera control , author=. Advances in Neural Information Processing Systems , volume=

  19. [19]

    and Tulyakov, Sergey , title =

    Bahmani, Sherwin and Skorokhodov, Ivan and Qian, Guocheng and Siarohin, Aliaksandr and Menapace, Willi and Tagliasacchi, Andrea and Lindell, David B. and Tulyakov, Sergey , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2025 , pages =

    2025

  20. [20]

    Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers , pages=

    Cinemaster: A 3d-aware and controllable framework for cinematic text-to-video generation , author=. Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers , pages=

  21. [21]

    Training-free camera control for video generation.arXiv preprint arXiv:2406.10126,

    Training-free camera control for video generation , author=. arXiv preprint arXiv:2406.10126 , year=

    work page arXiv

  22. [22]

    Mo- tionclone: Training-free motion cloning for controllable video generation.arXiv preprint arXiv:2406.05338, 2024

    Motionclone: Training-free motion cloning for controllable video generation , author=. arXiv preprint arXiv:2406.05338 , year=

    work page arXiv

  23. [23]

    European Conference on Computer Vision , pages=

    Motiondirector: Motion customization of text-to-video diffusion models , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  24. [24]

    Camco: Camera- controllable 3d-consistent image-to-video generation.arXiv preprint arXiv:2406.02509, 2024

    Camco: Camera-controllable 3d-consistent image-to-video generation , author=. arXiv preprint arXiv:2406.02509 , year=

    work page arXiv

  25. [25]

    ViewCrafter: Taming Video Diffusion Models for High-fidelity Novel View Synthesis , year=

    Yu, Wangbo and Xing, Jinbo and Yuan, Li and Hu, Wenbo and Li, Xiaoyu and Huang, Zhipeng and Gao, Xiangjun and Wong, Tien-Tsin and Shan, Ying and Tian, Yonghong , journal=. ViewCrafter: Taming Video Diffusion Models for High-fidelity Novel View Synthesis , year=

  26. [26]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Li, Teng and Zheng, Guangcong and Jiang, Rui and Zhan, Shuigen and Wu, Tao and Lu, Yehao and Lin, Yining and Deng, Chuanyun and Xiong, Yepan and Chen, Min and Cheng, Lin and Li, Xi , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2025 , pages =

    2025

  27. [27]

    arXiv preprint arXiv:2411.06525 , year=

    I2vcontrol-camera: Precise video camera control with adjustable motion strength , author=. arXiv preprint arXiv:2411.06525 , year=

    work page arXiv

  28. [28]

    Wan: Open and Advanced Large-Scale Video Generative Models

    Wan: Open and advanced large-scale video generative models , author=. arXiv preprint arXiv:2503.20314 , volume=

    work page internal anchor Pith review Pith/arXiv arXiv

  29. [29]

    HunyuanVideo: A Systematic Framework For Large Video Generative Models

    Hunyuanvideo: A systematic framework for large video generative models , author=. arXiv preprint arXiv:2412.03603 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  30. [30]

    CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer

    Cogvideox: Text-to-video diffusion models with an expert transformer , author=. arXiv preprint arXiv:2408.06072 , year=

    work page internal anchor Pith review arXiv

  31. [31]

    Advances in Neural Information Processing Systems , volume=

    Diffusion forcing: Next-token prediction meets full-sequence diffusion , author=. Advances in Neural Information Processing Systems , volume=

  32. [32]

    Proceedings of the 42nd International Conference on Machine Learning , articleno =

    Gao, Kaifeng and Shi, Jiaxin and Zhang, Hanwang and Wang, Chunping and Xiao, Jun and Chen, Long , title =. Proceedings of the 42nd International Conference on Machine Learning , articleno =. 2026 , publisher =

    2026

  33. [33]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    From slow bidirectional to fast autoregressive video diffusion models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  34. [34]

    The Thirteenth International Conference on Learning Representations,

    Yang Jin and Zhicheng Sun and Ningyuan Li and Kun Xu and Hao Jiang and Nan Zhuang and Quzhe Huang and Yang Song and Yadong Mu and Zhouchen Lin , title =. The Thirteenth International Conference on Learning Representations,. 2025 , url =

    2025

  35. [35]

    Rolling forcing: Autoregressive long video diffusion in real time.arXiv preprint arXiv:2509.25161, 2025a

    Rolling forcing: Autoregressive long video diffusion in real time , author=. arXiv preprint arXiv:2509.25161 , year=

    work page arXiv

  36. [36]

    MAGI-1: Autoregressive Video Generation at Scale

    Magi-1: Autoregressive video generation at scale , author=. arXiv preprint arXiv:2505.13211 , year=

    work page internal anchor Pith review arXiv

  37. [37]

    LongLive: Real-time Interactive Long Video Generation

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

    work page internal anchor Pith review arXiv

  38. [38]

    arXiv preprint arXiv:2512.05081 (2025)

    Deep forcing: Training-free long video generation with deep sink and participative compression , author=. arXiv preprint arXiv:2512.05081 , year=

    work page arXiv

  39. [39]

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

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

    work page arXiv

  40. [40]

    Hunyuan-gamecraft: High-dynamic interactive game video generation with hybrid history condition,

    Hunyuan-gamecraft: High-dynamic interactive game video generation with hybrid history condition , author=. arXiv preprint arXiv:2506.17201 , volume=

    work page arXiv

  41. [41]

    Yume: An interactive world generation model

    Yume: An interactive world generation model , author=. arXiv preprint arXiv:2507.17744 , year=

    work page arXiv

  42. [42]

    arXiv preprint arXiv:2512.11253 , year=

    PersonaLive! Expressive Portrait Image Animation for Live Streaming , author=. arXiv preprint arXiv:2512.11253 , year=

    work page arXiv

  43. [43]

    Omniforcing: Unleashing real-time joint audio-visual generation.arXiv preprint arXiv:2603.11647, 2026

    OmniForcing: Unleashing Real-time Joint Audio-Visual Generation , author=. arXiv preprint arXiv:2603.11647 , year=

    work page arXiv

  44. [44]

    arXiv preprint arXiv:2512.23576 , year=

    LiveTalk: Real-Time Multimodal Interactive Video Diffusion via Improved On-Policy Distillation , author=. arXiv preprint arXiv:2512.23576 , year=

    work page arXiv

  45. [45]

    arXiv preprint arXiv:2512.06065 , year=

    EgoEdit: Dataset, Real-Time Streaming Model, and Benchmark for Egocentric Video Editing , author=. arXiv preprint arXiv:2512.06065 , year=

    work page arXiv

  46. [46]

    URL: https://deepmind

    Genie 2: A large-scale foundation world model , author=. URL: https://deepmind. google/discover/blog/genie-2-a-large-scale-foundation-world-model , volume=

  47. [47]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    One-step diffusion with distribution matching distillation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  48. [48]

    Advances in neural information processing systems , volume=

    Improved distribution matching distillation for fast image synthesis , author=. Advances in neural information processing systems , volume=

  49. [49]

    International Conference on Learning Representations , year=

    Decoupled Weight Decay Regularization , author=. International Conference on Learning Representations , year=

  50. [50]

    OpenAI Blog , volume=

    Video generation models as world simulators , author=. OpenAI Blog , volume=

  51. [51]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Vbench: Comprehensive benchmark suite for video generative models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  52. [52]

    arXiv preprint arXiv:2508.10934 (2025)

    Vipe: Video pose engine for 3d geometric perception , author=. arXiv preprint arXiv:2508.10934 , year=

    work page arXiv

  53. [53]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Steerx: Creating any camera-free 3d and 4d scenes with geometric steering , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  54. [54]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Met3r: Measuring multi-view consistency in generated images , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  55. [55]

    CVPR , year =

    Hu, Wenbo and Gao, Xiangjun and Li, Xiaoyu and Zhao, Sijie and Cun, Xiaodong and Zhang, Yong and Quan, Long and Shan, Ying , title =. CVPR , year =

  56. [56]

    2024 , howpublished=

    Black Forest Labs , title=. 2024 , howpublished=

    2024

  57. [57]

    Forty-first international conference on machine learning , year=

    Scaling rectified flow transformers for high-resolution image synthesis , author=. Forty-first international conference on machine learning , year=