pith. sign in

arxiv: 2606.20799 · v1 · pith:ELDBIY3Unew · submitted 2026-06-18 · 💻 cs.CV · cs.AI

GroundShot: Visually Consistent Multi-Shot Long Video Generation via Entity-Grounded Shot Scheduling

Yixuan Lai , Tianjia Shao , Kun Zhou , Weijia Dou , Siyu Zhu , Jingdong Wang This is my paper

Pith reviewed 2026-06-26 18:14 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords multi-shot video generationvisual consistencyentity groundingshot schedulingtraining-freevideo synthesisGroundBench
0
0 comments X

The pith

GroundShot keeps entities consistent across shots in generated videos by anchoring later appearances to verified first views stored in an online memory.

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

The paper establishes that multi-shot video generation can achieve better visual consistency without any model training by building an entity-level visual memory from reliable initial appearances and using it to guide subsequent shots. A reader would care because existing methods suffer from accumulating inconsistencies that make characters and objects change appearance over long sequences, reducing their usefulness for narrative or simulation tasks. GroundShot does this by scheduling the order of shot generation according to how useful each will be as a reference for entities, grounding entities from generated content, verifying them, and retrieving matches before generation. It also provides GroundBench to measure entity-level consistency under controlled conditions.

Core claim

GroundShot is a training-free framework that builds an online entity visual memory from accepted shots, schedules generation order by expected usefulness as entity references, verifies entity reliability before storage, and retrieves references for each new shot to prevent drift from initial appearances.

What carries the argument

Entity-grounded shot scheduling with online verified memory, which decides shot order based on reference usefulness and anchors generation to first reliable entity appearances.

If this is right

  • Existing video generation models can produce longer consistent multi-shot videos without modification.
  • Consistency can be maintained by prioritizing shots that provide clear entity views early in the process.
  • Entity-level evaluation via GroundBench isolates consistency issues from other video quality factors.
  • Verification of generated entities before memory addition reduces error propagation.

Where Pith is reading between the lines

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

  • Similar memory mechanisms might apply to other generative tasks like image sequences or 3D scene consistency.
  • Testing on videos with more than the evaluated shot counts could reveal limits of the scheduling approach.
  • The method assumes entities can be reliably grounded and verified automatically, which may vary by domain.

Load-bearing premise

That scheduling shots by their usefulness as entity references and storing only verified first appearances will stop inconsistencies from accumulating as the number of shots increases.

What would settle it

Generate a multi-shot video using GroundShot and a baseline on a sequence where entities reappear in later shots; measure if entity appearance drift is reduced compared to baseline, or if it matches baseline when shot count exceeds a threshold.

Figures

Figures reproduced from arXiv: 2606.20799 by Jingdong Wang, Kun Zhou, Siyu Zhu, Tianjia Shao, Weijia Dou, Yixuan Lai.

Figure 1
Figure 1. Figure 1: GroundShot generates visually consistent multi-shot long videos from text scripts. Each row shows a generated video with colored annotations marking recurring characters and objects whose cross-shot consistency should be compared. Despite changes in viewpoint, scale, composition, and interaction, GroundShot preserves entity identity, clothing and object details, and scene appearance. The bottom strip expan… view at source ↗
Figure 2
Figure 2. Figure 2: Motivation for quality-aware shot scheduling. (a) Narrative-order generation may build memory from an early low￾quality entity crop, leaving later shots with a weak reference. (b) Viewers judge recurring entities against a canonical reference, mo￾tivating a star-shaped consistency structure that ties all appear￾ances to the same reference. (c) GroundShot first generates shots likely to yield clear entity-l… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of GroundShot. (1) Script parsing extracts recurring characters, objects, and locations, and identifies the entity set for each shot. (2) Shot scheduling builds a dependency graph from predicted reference-source shots and topologically sorts it into a scheduled generation order. (3) Entity-level visual memory stores canonical and canonical-consistent supplementary references for characters, object… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison on a six-shot living-room script. The setting contains three recurring characters and a powder-blue [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Anchor-relative consistency over narrative distance. The [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative effect of visual memory dynamic main￾tenance. The first two rows show character cases where a sin￾gle canonical reference preserves identity but does not fully cover later expression and view changes; dynamic maintenance provides identity-safe supplementary references for these target shots. The last two rows show scene cases, where dynamic maintenance is more critical: additional scene referen… view at source ↗
read the original abstract

Generating visually consistent multi-shot videos remains an open challenge. As videos span more shots, inconsistencies can accumulate across shots, causing entities that reappear across shots -- characters, objects, and locations -- to drift away from how they first appear. We observe that viewers judge consistency by comparing each later appearance of an entity with its first clear appearance; the visual quality of this initial appearance sets the consistency ceiling for all that follows. Motivated by this, we present \textbf{GroundShot}, a training-free, model-agnostic agentic framework for entity-grounded multi-shot generation. GroundShot builds an entity-level visual memory online from accepted generated shots: it schedules shots' generation order by their expected usefulness as entity references, grounds entities from generated videos, verifies their reliability before adding them to memory, and retrieves suitable entity references from memory before each shot is generated. To evaluate this entity-centered view of consistency, we further introduce \textbf{GroundBench}, a diagnostic benchmark that measures consistency at the entity level while isolating controlled challenge dimensions. Experiments show that GroundShot improves multi-shot consistency over existing methods while requiring no additional training or model modification.

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

1 major / 2 minor

Summary. The manuscript introduces GroundShot, a training-free, model-agnostic agentic framework for multi-shot long video generation. It constructs an online entity-level visual memory from verified generated shots, orders shot generation by expected reference usefulness, performs entity grounding and reliability verification before memory writes, and retrieves entity references prior to each shot. The work also presents GroundBench, a diagnostic benchmark for entity-level consistency evaluation under controlled challenge dimensions, and reports that GroundShot improves consistency over baselines without model modification or retraining.

Significance. If the empirical results on GroundBench hold, the approach provides a practical, immediately deployable solution to entity drift in multi-shot video synthesis by grounding consistency to first-appearance references and gating memory via verification. The training-free and model-agnostic design, combined with the entity-centered diagnostic benchmark, represents a useful contribution to the video generation literature.

major comments (1)
  1. [Experiments / Results] The central claim that the entity-memory + usefulness scheduling loop measurably reduces entity drift rests on the experimental comparison in the results section; without the specific quantitative metrics, baselines, and statistical analysis from that section, the improvement cannot be assessed for load-bearing support.
minor comments (2)
  1. [Abstract] The abstract states that experiments show improvement but supplies no numbers, error bars, or dataset sizes; moving at least one key quantitative result into the abstract would strengthen the claim.
  2. [Method] Notation for the usefulness scoring function and the verification threshold should be defined explicitly with a short equation or pseudocode block to avoid ambiguity when describing the scheduling and memory-write logic.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive evaluation and recommendation of minor revision. We address the single major comment below.

read point-by-point responses

  1. Referee: [Experiments / Results] The central claim that the entity-memory + usefulness scheduling loop measurably reduces entity drift rests on the experimental comparison in the results section; without the specific quantitative metrics, baselines, and statistical analysis from that section, the improvement cannot be assessed for load-bearing support.

    Authors: We agree that the quantitative results are essential to substantiate the central claim. The results section reports entity-level consistency metrics on GroundBench across controlled challenge dimensions, direct comparisons against multiple baselines (including both training-free and fine-tuned methods), and statistical analysis with means and standard deviations computed over repeated generations. To make this support more explicit and self-contained, we will expand the results presentation with additional tabulated breakdowns and clearer linkage between the scheduling/memory components and the observed gains in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper describes a training-free, model-agnostic framework that builds an online entity memory directly from verified generated shots, schedules by usefulness, and retrieves references before generation. No equations, fitted parameters, or predictions are present. No self-citations are invoked as load-bearing premises, and the central claims rest on empirical evaluation against baselines on the introduced GroundBench. The method is externally falsifiable via the benchmark metrics and does not reduce any result to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Review limited to abstract; no explicit free parameters, axioms, or invented entities detailed beyond the high-level framework description.

axioms (1)
  • domain assumption Viewers judge consistency primarily by comparing later entity appearances to the first clear appearance.
    Explicitly stated as the motivating observation in the abstract.
invented entities (1)
  • entity-level visual memory no independent evidence
    purpose: Stores verified entity references extracted from generated shots for use in future generations.
    Core component of the proposed framework introduced in the abstract.

pith-pipeline@v0.9.1-grok · 5748 in / 1139 out tokens · 29283 ms · 2026-06-26T18:14:06.724157+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

73 extracted references · 17 linked inside Pith

  1. [1]

    Onestory: Coherent multi-shot video generation with adaptive memory.arXiv preprint arXiv:2512.07802, 2025

    Zhaochong An, Menglin Jia, Haonan Qiu, Zijian Zhou, Xi- aoke Huang, Zhiheng Liu, Yiqi Yuan, Zuwei Li, Huazhu Chang, and Ziwei Liu. Onestory: Coherent multi-shot video generation with adaptive memory.arXiv preprint arXiv:2512.07802, 2025

    arXiv 2025

  2. [2]

    Vidu: a highly consistent, dynamic and skilled text-to-video generator with diffusion models.arXiv preprint arXiv:2405.04233, 2024

    Fan Bao, Chendong Xiang, Gang Yue, Guande He, Hongzhou Zhu, Kaiwen Zheng, Min Zhao, Shilong Liu, Yaole Wang, and Jun Zhu. Vidu: a highly consistent, dynamic and skilled text-to-video generator with diffusion models.arXiv preprint arXiv:2405.04233, 2024

    arXiv 2024

  3. [3]

    Align your latents: High-resolution video synthesis with la- tent diffusion models

    Andreas Blattmann, Robin Rombach, Huan Ling, Tim Dock- horn, Seung Wook Kim, Sanja Fidler, and Karsten Kreis. Align your latents: High-resolution video synthesis with la- tent diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    2023

  4. [4]

    Agentic video generation: From text to executable event graphs via tool-constrained llm planning.arXiv preprint arXiv:2604.10383, 2026

    Nicolae Cudlenco, Mihai Masala, and Marius Leordeanu. Agentic video generation: From text to executable event graphs via tool-constrained llm planning.arXiv preprint arXiv:2604.10383, 2026

    Pith/arXiv arXiv 2026

  5. [5]

    Arcface: Additive angular margin loss for deep face recognition

    Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou. Arcface: Additive angular margin loss for deep face recognition. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), pages 4690–4699, 2019

    2019

  6. [6]

    Infinitystory: Unlimited video generation with world consistency and character-aware shot transitions.arXiv preprint arXiv:2603.03646, 2025

    Mohamed Elmoghany, Liangbing Zhao, Xiaoqian Shen, Subhojyoti Mukherjee, Yang Zhou, Gang Wu, Viet Dac Lai, Seunghyun Yoon, Ryan Rossi, Abdullah Rashwan, Puneet Mathur, Varun Manjunatha, Daksh Dangi, Chien Nguyen, Nedim Lipka, Trung Bui, Krishna Kumar Singh, Ruiyi Zhang, Xiaolei Huang, Jaemin Cho, Yu Wang, Namyong Park, Zhengzhong Tu, Hongjie Chen, Hoda Eld...

    arXiv 2025

  7. [7]

    Seedance 1.0: Exploring the boundaries of video generation models.arXiv preprint arXiv:2506.09113, 2025

    Yu Gao, Haoyuan Guo, Tuyen Hoang, Weilin Huang, Lu Jiang, Fangyuan Kong, Huixia Li, Jiashi Li, Liang Li, Xi- aojie Li, Xunsong Li, Yifu Li, Shanchuan Lin, Zhijie Lin, Jiawei Liu, Shu Liu, Xiaonan Nie, Zhiwu Qing, Yuxi Ren, Li Sun, Zhi Tian, Rui Wang, Sen Wang, Guoqiang Wei, Guohong Wu, Jie Wu, Ruiqi Xia, Fei Xiao, Xuefeng Xiao, Jiangqiao Yan, Ceyuan Yang,...

    Pith/arXiv arXiv 2025

  8. [8]

    Animatediff: Animate your personalized text- to-image diffusion models without specific tuning

    Yuwei Guo, Ceyuan Yang, Anyi Rao, Zhengyang Liang, Yaohui Wang, Yu Qiao, Maneesh Agrawala, Dahua Lin, and Bo Dai. Animatediff: Animate your personalized text- to-image diffusion models without specific tuning. InThe Twelfth International Conference on Learning Representa- tions (ICLR), 2024

    2024

  9. [9]

    Entitybench: Towards entity-consistent long-range multi- shot video generation.arXiv preprint arXiv:2605.15199, 2026

    Ruozhen He, Meng Wei, Ziyan Yang, and Vicente Ordonez. Entitybench: Towards entity-consistent long-range multi- shot video generation.arXiv preprint arXiv:2605.15199, 2026

    Pith/arXiv arXiv 2026

  10. [10]

    Camera artist: A multi-agent framework for cinematic language sto- rytelling video generation.arXiv preprint arXiv:2604.09195, 2025

    Haobo Hu, Qi Mao, Yuanhang Li, and Libiao Jin. Camera artist: A multi-agent framework for cinematic language sto- rytelling video generation.arXiv preprint arXiv:2604.09195, 2025

    Pith/arXiv arXiv 2025

  11. [11]

    Storyagent: Cus- tomized storytelling video generation via multi-agent collab- oration.arXiv preprint arXiv:2411.04925, 2024

    Panwen Hu, Jin Jiang, Jianqi Chen, Mingfei Han, Shengcai Liao, Xiaojun Chang, and Xiaodan Liang. Storyagent: Cus- tomized storytelling video generation via multi-agent collab- oration.arXiv preprint arXiv:2411.04925, 2024

    arXiv 2024

  12. [12]

    VBench: Com- prehensive benchmark suite for video generative models

    Ziqi Huang, Yinan He, Jiashuo Yu, Fan Zhang, Chenyang Si, Yuming Jiang, Yuanhan Zhang, Tianxing Wu, Qingyang Jin, Nattapol Chanpaisit, Yaohui Wang, Xinyuan Chen, Limin Wang, Dahua Lin, Yu Qiao, and Ziwei Liu. VBench: Com- prehensive benchmark suite for video generative models. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Reco...

    2024

  13. [13]

    Ozgur Kara, Krishna Kumar Singh, Feng Liu, Duygu Cey- lan, and James M. Rehg. Shotadapter: Text-to-multi-shot 13 video generation with diffusion models.arXiv preprint arXiv:2505.07652, 2025

    arXiv 2025

  14. [14]

    Text2video-zero: Text- to-image diffusion models are zero-shot video generators

    Levon Khachatryan, Andranik Movsisyan, Vahram Tade- vosyan, Roberto Henschel, Zhangyang Wang, Shant Navasardyan, and Humphrey Shi. Text2video-zero: Text- to-image diffusion models are zero-shot video generators. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023

    2023

  15. [15]

    Kling-omni technical report.arXiv preprint arXiv:2512.16776, 2025

    Kling Team. Kling-omni technical report.arXiv preprint arXiv:2512.16776, 2025

    Pith/arXiv arXiv 2025

  16. [16]

    Gligen: Open-set grounded text-to-image generation

    Yuheng Li, Haotian Liu, Qingyang Wu, Fangzhou Mu, Jian- wei Yang, Jianfeng Gao, Chunyuan Li, and Yong Jae Lee. Gligen: Open-set grounded text-to-image generation. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    2023

  17. [17]

    Photomaker: Customizing re- alistic human photos via stacked id embedding

    Zhen Li, Mingdeng Cao, Xintao Wang, Zhongang Qi, Ming- Ming Cheng, and Ying Shan. Photomaker: Customizing re- alistic human photos via stacked id embedding. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  18. [18]

    Llm- grounded diffusion: Enhancing prompt understanding of text-to-image diffusion models with large language models

    Long Lian, Boyi Li, Adam Yala, and Trevor Darrell. Llm- grounded diffusion: Enhancing prompt understanding of text-to-image diffusion models with large language models. Transactions on Machine Learning Research (TMLR), 2024

    2024

  19. [19]

    Llm-grounded video diffusion models

    Long Lian, Baifeng Shi, Adam Yala, Trevor Darrell, and Boyi Li. Llm-grounded video diffusion models. InInter- national Conference on Learning Representations (ICLR), 2024

    2024

  20. [20]

    Referdino: Referring video object segmentation with visual grounding founda- tions

    Tianming Liang, Kun-Yu Lin, Chaolei Tan, Jianguo Zhang, Wei-Shi Zheng, and Jian-Fang Hu. Referdino: Referring video object segmentation with visual grounding founda- tions. InProceedings of the IEEE/CVF International Con- ference on Computer Vision (ICCV), 2025

    2025

  21. [21]

    Videodirectorgpt: Consistent multi-scene video generation via llm-guided planning

    Han Lin, Abhay Zala, Jaemin Cho, and Mohit Bansal. Videodirectorgpt: Consistent multi-scene video generation via llm-guided planning. InConference on Language Mod- eling (COLM), 2024

    2024

  22. [22]

    Ad- vancing narrative long video generation via training-free identity-aware memory.arXiv preprint arXiv:2605.18733, 2026

    Jinzhuo Liu, Jiangning Zhang, Wencan Jiang, Yabiao Wang, Dingkang Liang, Zhucun Xue, Ran Yi, and Yong Liu. Ad- vancing narrative long video generation via training-free identity-aware memory.arXiv preprint arXiv:2605.18733, 2026

    Pith/arXiv arXiv 2026

  23. [23]

    Phantom: Subject-consistent video generation via cross- modal alignment.arXiv preprint arXiv:2502.11079, 2025

    Lijie Liu, Tianxiang Ma, Bingchuan Li, Zhuowei Chen, Ji- awei Liu, Gen Li, Siyu Zhou, Qian He, and Xinglong Wu. Phantom: Subject-consistent video generation via cross- modal alignment.arXiv preprint arXiv:2502.11079, 2025

    arXiv 2025

  24. [24]

    Grounding dino: Mar- rying dino with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, and Lei Zhang. Grounding dino: Mar- rying dino with grounded pre-training for open-set object detection. InEuropean Conference on Computer Vision (ECCV), 2024

    2024

  25. [25]

    Sora: A review on background, technology, limitations, and opportunities of large vision models.arXiv preprint arXiv:2402.17177, 2024

    Yixin Liu, Kai Zhang, Yuan Li, Zhiling Yan, Chujie Gao, Ruoxi Chen, Zhengqing Yuan, Yue Huang, Hanchi Sun, Jian- feng Gao, Lifang He, and Lichao Sun. Sora: A review on background, technology, limitations, and opportunities of large vision models.arXiv preprint arXiv:2402.17177, 2024

    Pith/arXiv arXiv 2024

  26. [26]

    Do Xuan Long, Xingchen Wan, Hootan Nakhost, Chen-Yu Lee, Tomas Pfister, and Sercan ¨O. Arık. Vista: A test- time self-improving video generation agent.arXiv preprint arXiv:2510.15831, 2025

    arXiv 2025

  27. [27]

    Do Xuan Long, Yale Song, Min-Yen Kan, Tomas Pfister, and Long T. Le. A2rd: Agentic autoregressive diffusion for long video consistency.arXiv preprint arXiv:2605.06924, 2026

    Pith/arXiv arXiv 2026

  28. [28]

    Videostudio: Generating consistent-content and multi-scene videos.arXiv preprint arXiv:2401.01256, 2024

    Fuchen Long, Zhaofan Qiu, Ting Yao, and Tao Mei. Videostudio: Generating consistent-content and multi-scene videos.arXiv preprint arXiv:2401.01256, 2024

    arXiv 2024

  29. [29]

    Luma dream machine.https : / / lumalabs

    Luma AI, Inc. Luma dream machine.https : / / lumalabs . ai / dream - machine, 2024. Accessed: 2025

    2024

  30. [30]

    Filmweaver: Weaving consistent multi-shot videos with cache-guided autoregressive diffu- sion.arXiv preprint arXiv:2512.11274, 2025

    Xiangyang Luo, Qingyu Li, Xiaokun Liu, Wenyu Qin, Miao Yang, and Meng Wang. Filmweaver: Weaving consistent multi-shot videos with cache-guided autoregressive diffu- sion.arXiv preprint arXiv:2512.11274, 2025

    arXiv 2025

  31. [31]

    Shotstream: Streaming multi-shot video generation for inter- active storytelling.arXiv preprint arXiv:2603.25746, 2025

    Yawen Luo, Xiaoyu Shi, Junhao Zhuang, Yutian Chen, Quande Liu, Xintao Wang, Pengfei Wan, and Tianfan Xue. Shotstream: Streaming multi-shot video generation for inter- active storytelling.arXiv preprint arXiv:2603.25746, 2025

    arXiv 2025

  32. [32]

    Holocine: Holis- tic generation of cinematic multi-shot long video narratives

    Yihao Meng, Hao Ouyang, Yue Yu, Qiuyu Wang, Wen Wang, Kaijun Chen, Ying He, and Lu Fang. Holocine: Holis- tic generation of cinematic multi-shot long video narratives. arXiv preprint arXiv:2510.20822, 2025

    arXiv 2025

  33. [33]

    Causalcine: Real-time autoregressive generation for multi-shot video narratives.arXiv preprint arXiv:2605.12496, 2026

    Yihao Meng, Zichen Liu, Hao Ouyang, Qiuyu Wang, Ka Leong Cheng, Yue Yu, Hanlin Wang, Haobo Li, Jia- peng Zhu, Yanhong Zeng, Xing Zhu, Yujun Shen, Qifeng Chen, and Huamin Qu. Causalcine: Real-time autoregressive generation for multi-shot video narratives.arXiv preprint arXiv:2605.12496, 2026

    Pith/arXiv arXiv 2026

  34. [34]

    Minimax hailuo ai video generator.https://hailuoai.video/, 2024

    MiniMax Technology Co., Ltd. Minimax hailuo ai video generator.https://hailuoai.video/, 2024. Ac- cessed: 2025

    2024

  35. [35]

    GPT-4o system card.arXiv preprint arXiv:2410.21276, 2024

    OpenAI. GPT-4o system card.arXiv preprint arXiv:2410.21276, 2024

    Pith/arXiv arXiv 2024

  36. [36]

    OpenAI. GPT-4.1. 2025

    2025

  37. [37]

    Maxime Oquab, Timoth ´ee Darcet, Th´eo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mah- moud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Herv ´e J´egou, Julien Mairal...

    Pith/arXiv arXiv 2024

  38. [38]

    Sampson, Shikai Li, Simone Parmeggiani, Steve Fine, Tara Fowler, Vladan Petro- vic, and Yuming Du

    Adam Polyak, Amit Zohar, Andrew Brown, Andros Tjandra, Animesh Sinha, Ann Lee, Apoorv Vyas, Bowen Shi, Chih- Yao Ma, Ching-Yao Chuang, David Yan, Dhruv Choudhary, Dingkang Wang, Geet Sethi, Guan Pang, Haoyu Ma, Ishan Misra, Ji Hou, Jialiang Wang, Kiran Jagadeesh, Kunpeng Li, Luxin Zhang, Mannat Singh, Mary Williamson, Matt Le, Matthew Yu, Mitesh Kumar Sin...

    Pith/arXiv arXiv 2024

  39. [39]

    Mask 2dit: Dual mask-based diffusion transformer for multi-scene long video generation.arXiv preprint arXiv:2503.19881, 2025

    Tianhao Qi, Jianlong Yuan, Wanquan Feng, Shancheng Fang, Jiawei Liu, SiYu Zhou, Qian He, Hongtao Xie, and Yongdong Zhang. Mask 2dit: Dual mask-based diffusion transformer for multi-scene long video generation.arXiv preprint arXiv:2503.19881, 2025

    arXiv 2025

  40. [40]

    Runway Gen-3 alpha.https : / / runwayml

    Runway AI, Inc. Runway Gen-3 alpha.https : / / runwayml . com / research / gen - 3 - alpha, 2024. Accessed: 2025

    2024

  41. [41]

    Muse: A multi-agent frame- work for unconstrained story envisioning via closed-loop cognitive orchestration.arXiv preprint arXiv:2602.03028, 2026

    Wenzhang Sun, Zhenyu Wang, Zhangchi Hu, Chunfeng Wang, Hao Li, and Wei Chen. Muse: A multi-agent frame- work for unconstrained story envisioning via closed-loop cognitive orchestration.arXiv preprint arXiv:2602.03028, 2026

    arXiv 2026

  42. [42]

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

    Wan Team, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianx- iao Yang, Jianyuan Zeng, Jiayu Wang, Jingfeng Zhang, Jin- gren Zhou, Jinkai Wang, Jixuan Chen, Kai Zhu, Kang Zhao, Keyu Yan, Lianghua Huang, Mengyang Feng, Ningyi Zhang, Pandeng Li, Pingyu Wu, Ruihang Chu, Ruili Feng, Shiwei Zhang, Siyang Sun, Tao Fan...

    Pith/arXiv arXiv 2025

  43. [43]

    Anyid: Ultra-fidelity universal identity-preserving video generation from any visual references.arXiv preprint arXiv:2603.25188, 2025

    Jiahao Wang, Hualian Sheng, Sijia Cai, Yuxiao Yang, Weizhan Zhang, Caixia Yan, Bing Deng, and Jieping Ye. Anyid: Ultra-fidelity universal identity-preserving video generation from any visual references.arXiv preprint arXiv:2603.25188, 2025

    arXiv 2025

  44. [44]

    Echoshot: Multi-shot portrait video generation.arXiv preprint arXiv:2506.15838, 2025

    Jiahao Wang, Hualian Sheng, Sijia Cai, Weizhan Zhang, Caixia Yan, Yachuang Feng, Bing Deng, and Jieping Ye. Echoshot: Multi-shot portrait video generation.arXiv preprint arXiv:2506.15838, 2025

    arXiv 2025

  45. [45]

    Multishotmaster: A controllable multi-shot video generation framework.arXiv preprint arXiv:2512.03041, 2025

    Qinghe Wang, Xiaoyu Shi, Baolu Li, Weikang Bian, Quande Liu, Huchuan Lu, Xintao Wang, Pengfei Wan, Kun Gai, and Xu Jia. Multishotmaster: A controllable multi-shot video generation framework.arXiv preprint arXiv:2512.03041, 2025

    arXiv 2025

  46. [46]

    Internvid: A large-scale video-text dataset for multimodal understanding and generation.arXiv preprint arXiv:2307.06942, 2023

    Yi Wang, Yinan He, Yizhuo Li, Kunchang Li, Jiashuo Yu, Xin Ma, Xinhao Li, Guo Chen, Xinyuan Chen, Yaohui Wang, Conghui He, Ping Luo, Ziwei Liu, Yali Wang, Limin Wang, and Yu Qiao. Internvid: A large-scale video-text dataset for multimodal understanding and generation.arXiv preprint arXiv:2307.06942, 2023

    Pith/arXiv arXiv 2023

  47. [47]

    Dreamvideo: Composing your dream videos with customized subject and motion

    Yujie Wei, Shiwei Zhang, Zhiwu Qing, Hangjie Yuan, Zhi- heng Liu, Yu Liu, Yingya Zhang, Jingren Zhou, and Hong- ming Shan. Dreamvideo: Composing your dream videos with customized subject and motion. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  48. [48]

    Echo-forcing: A scene memory framework for interactive long video generation

    Mingqiang Wu, Weilun Feng, Zhefeng Zhang, Haotong Qin, Yuqi Li, Guoxin Fan, Xiaokun Liu, Zhulin An, Libo Huang, Yongjun Xu, and Chuanguang Yang. Echo-forcing: A scene memory framework for interactive long video generation. arXiv preprint arXiv:2605.16003, 2026

    Pith/arXiv arXiv 2026

  49. [49]

    Auto- mated movie generation via multi-agent cot planning.arXiv preprint arXiv:2503.07314, 2025

    Weijia Wu, Zeyu Zhu, and Mike Zheng Shou. Auto- mated movie generation via multi-agent cot planning.arXiv preprint arXiv:2503.07314, 2025

    arXiv 2025

  50. [50]

    Shotdirector: Directorially controllable multi-shot video generation with cinematographic transitions.arXiv preprint arXiv:2512.10286, 2025

    Xiaoxue Wu, Xinyuan Chen, Yaohui Wang, and Yu Qiao. Shotdirector: Directorially controllable multi-shot video generation with cinematographic transitions.arXiv preprint arXiv:2512.10286, 2025

    arXiv 2025

  51. [51]

    Captain cinema: Towards short movie generation.arXiv preprint arXiv:2507.18634, 2025

    Junfei Xiao, Ceyuan Yang, Lvmin Zhang, Shengqu Cai, Yang Zhao, Yuwei Guo, and Gordon Wetzstein. Captain cinema: Towards short movie generation.arXiv preprint arXiv:2507.18634, 2025

    arXiv 2025

  52. [52]

    Bissyand´e

    Zhifei Xie, Daniel Tang, Dingwei Tan, Jacques Klein, and Tegawend´e F. Bissyand´e. Dreamfactory: Pioneering multi- scene long video generation with a multi-agent framework. arXiv preprint arXiv:2408.11788, 2024

    arXiv 2024

  53. [53]

    Mm- storyagent: Immersive narrated storybook video generation with a multi-agent paradigm across text, image and audio

    Xuenan Xu, Jiahao Mei, Chenliang Li, Yuning Wu, Ming Yan, Shaopeng Lai, Ji Zhang, and Mengyue Wu. Mm- storyagent: Immersive narrated storybook video generation with a multi-agent paradigm across text, image and audio. arXiv preprint arXiv:2503.05242, 2025

    arXiv 2025

  54. [54]

    Filmagent: A multi- agent framework for end-to-end film automation in virtual 3d spaces.arXiv preprint arXiv:2501.12909, 2025

    Zhenran Xu, Longyue Wang, Jifang Wang, Zhouyi Li, Sen- bao Liao, Baotian Tan, and Min Zhang. Filmagent: A multi- agent framework for end-to-end film automation in virtual 3d spaces.arXiv preprint arXiv:2501.12909, 2025

    arXiv 2025

  55. [55]

    Gloria: Con- sistent character video generation via content anchors.arXiv preprint arXiv:2603.29931, 2025

    Yuhang Yang, Fan Zhang, Huaijin Pi, Shuai Guo, Guowei Xu, Wei Zhai, Yang Cao, and Zheng-Jun Zha. Gloria: Con- sistent character video generation via content anchors.arXiv preprint arXiv:2603.29931, 2025

    arXiv 2025

  56. [56]

    Cogvideox: Text-to-video diffusion models with an expert transformer.arXiv preprint arXiv:2408.06072, 2024

    Zhuoyi Yang, Jiayan Teng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Yang, Wenyi Hong, Xiao- han Zhang, Guanyu Feng, Da Yin, Yuxuan Zhang, Weihan Wang, Yean Cheng, Bin Xu, Xiaotao Gu, Yuxiao Dong, and Jie Tang. Cogvideox: Text-to-video diffusion models with an expert transformer.arXiv preprint arXiv:2408.06072, 2024

    Pith/arXiv arXiv 2024

  57. [57]

    Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models.arXiv preprint arXiv:2308.06721, 2023

    Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models.arXiv preprint arXiv:2308.06721, 2023

    Pith/arXiv arXiv 2023

  58. [58]

    Storymem: Multi-shot long video storytelling with memory.arXiv preprint arXiv:2512.19539, 2024

    Kaiwen Zhang, Liming Jiang, Angtian Wang, Jacob Zhiyuan Fang, Tiancheng Zhi, Qing Yan, Hao Kang, Xin Lu, and 15 Xingang Pan. Storymem: Multi-shot long video storytelling with memory.arXiv preprint arXiv:2512.19539, 2024

    arXiv 2024

  59. [59]

    Anime: Adap- tive multi-agent planning for long animation generation

    Lisai Zhang, Baohan Xu, Siqian Yang, Mingyu Yin, Jing Liu, Chao Xu, Siqi Wang, Yidi Wu, Yuxin Hong, Zihao Zhang, Yanzhang Liang, and Yudong Jiang. Anime: Adap- tive multi-agent planning for long animation generation. arXiv preprint arXiv:2508.18781, 2025

    arXiv 2025

  60. [60]

    Stage: Storyboard-anchored gener- ation for cinematic multi-shot narrative.arXiv preprint arXiv:2512.12372, 2025

    Peixuan Zhang, Zijian Jia, Kaiqi Liu, Shuchen Weng, Si Li, and Boxin Shi. Stage: Storyboard-anchored gener- ation for cinematic multi-shot narrative.arXiv preprint arXiv:2512.12372, 2025

    arXiv 2025

  61. [61]

    Moviedreamer: Hierarchical genera- tion for coherent long visual sequence.arXiv preprint arXiv:2407.16655, 2024

    Canyu Zhao, Mingyu Liu, Wen Wang, Weihua Chen, Hao Chen, Zhe Yuan, Jianming Fan, Jian Wang, Ying Shan, and Chunhua Li. Moviedreamer: Hierarchical genera- tion for coherent long visual sequence.arXiv preprint arXiv:2407.16655, 2024

    arXiv 2024

  62. [62]

    Videogen-of-thought: Step-by-step generating multi- shot video with minimal manual intervention.arXiv preprint arXiv:2503.15138, 2025

    Mingzhe Zheng, Yongqi Xu, Haojian Huang, Xuran Ma, Yexin Liu, Wenjie Li, Jiahao Wang, Hao Tang, Wenqi Zhou, Yingqing Wang, Ao Wang, Pengfei Zhang, Bo Chen, and Yu Qiao. Videogen-of-thought: Step-by-step generating multi- shot video with minimal manual intervention.arXiv preprint arXiv:2503.15138, 2025

    arXiv 2025

  63. [63]

    Videomemory: Toward consis- tent video generation via memory integration.arXiv preprint arXiv:2601.03655, 2025

    Jinsong Zhou, Yihua Du, Xinli Xu, Luozhou Wang, Zijie Zhuang, Yehang Zhang, Shuaibo Li, Xiaojun Hu, Bolan Su, and Ying-cong Chen. Videomemory: Toward consis- tent video generation via memory integration.arXiv preprint arXiv:2601.03655, 2025

    arXiv 2025

  64. [64]

    Storydiffusion: Consistent self- attention for long-range image and video generation

    Yupeng Zhou, Daquan Zhou, Ming-Ming Cheng, Jiashi Feng, and Qibin Hou. Storydiffusion: Consistent self- attention for long-range image and video generation. InAd- vances in Neural Information Processing Systems (NeurIPS), 2024

    2024

  65. [65]

    Libra- gen: Playing a balance game in subject-driven video genera- tion.arXiv preprint arXiv:2603.13506, 2025

    Jiahao Zhu, Shanshan Lao, Lijie Liu, Gen Li, Tianhao Qi, Wei Han, Bingchuan Li, Fangfang Liu, Zhuowei Chen, Tianxiang Ma, Qian He, Yi Zhou, and Xiaohua Xie. Libra- gen: Playing a balance game in subject-driven video genera- tion.arXiv preprint arXiv:2603.13506, 2025

    arXiv 2025

  66. [66]

    Vlogger: Make your dream a vlog.arXiv preprint arXiv:2401.09414, 2024

    Shaobin Zhuang, Kunchang Li, Xinyuan Chen, Yaohui Wang, Ziwei Liu, Yu Qiao, and Yali Wang. Vlogger: Make your dream a vlog.arXiv preprint arXiv:2401.09414, 2024. 16 GroundShot: Visually Consistent Multi-Shot Long Video Generation via Entity-Grounded Shot Scheduling Supplementary Material

    arXiv 2024

  67. [67]

    Alex Chen, a male detective in a dark trench coat, enters the police station lobby

    Supplementary Method Details 6.1. Entity Graph Construction GroundShot relies on a lightweight entity graph to connect script-level semantics with visual grounding and reference reuse. For each shots i, the LLM parser outputs a set of entity recordsE i. Each record contains: • a stable entity ID (e.g.,char alex,obj briefcase, orloc station); • an entity t...

  68. [68]

    Chars.” and “Objs

    GroundBench: Detailed Specification 7.1. Benchmark Taxonomy and Distribution GroundBench contains 54 Y AML scripts and 309 shots. It is organized into 18 sub-modules under four diag- nostic modules; each sub-module has three scripts with challenge level∈ {1,2,3}. The level split is balanced at the script level (18 scripts per level), with 99/103/107 shots...

  69. [69]

    Characters and objects use foreground crops; locations use reconstructed scene references

    Entity-Level Visual Memory The entity-level visual memoryR={R e}e stores a com- pact active reference set for each entity. Characters and objects use foreground crops; locations use reconstructed scene references. Each entity has one protected canonical referencer ∗ e and, after canonical initialization, a small aux- iliary pool for canonical-consistent v...

  70. [70]

    High quality:q(c)≥0.85for foreground references, or qloc(cloc, e)≥0.85for location scene references

  71. [71]

    Canonical-ready visibility: face confidence≥0.7and near-frontal visibility for characters; recognizable, com- plete appearance for objects; artifact-free, foreground- free reconstruction for locations

  72. [72]

    The first candidate satisfying these conditions initializesr∗ e, regardless of narrative or execution index

    Empty canonical slot: no canonical reference already ex- ists for this entity. The first candidate satisfying these conditions initializesr∗ e, regardless of narrative or execution index. Beforer ∗ e exists, non-canonical candidates are rejected rather than stored as auxiliaries. Afterr ∗ e exists, later candidates may enter only as auxiliary references a...

  73. [73]

    use_canonical

    Agentic Reference Selection When multiple references are available, GroundShot selects a target-aware subset rather than blindly using the highest- quality image. The canonical reference remains the default consistency anchor, but it is not mandatory in every gen- eration call: characters usually keep it for identity, while objects and locations may use a...