Custom Storyboard Generation (CSG) aims to produce high-quality, multi-character consistent storytelling. Current approaches based on static diffusion models, whether used in a one-shot manner or within multi-agent frameworks, face three key limitations: (1) Static models lack dynamic expressiveness and often resort to "copy-paste" pattern. (2) One-shot inference cannot iteratively correct missing attributes or poor prompt adherence. (3) Multi-agents rely on non-robust evaluators, ill-suited for assessing stylized, non-realistic animation. To address these, we propose AnimeAgent, the first Image-to-Video (I2V)-based multi-agent framework for CSG. Inspired by Disney's "Combination of Straight Ahead and Pose to Pose" workflow, AnimeAgent leverages I2V's implicit motion prior to enhance consistency and expressiveness, while a mixed subjective-objective reviewer enables reliable iterative refinement. We also collect a human-annotated CSG benchmark with ground-truth. Experiments show AnimeAgent achieves SOTA performance in consistency, prompt fidelity, and stylization.
Figures & tables
Figure 1: Comparison of paradigms for CSG. Our CSG paradigm is inspired by the ( d ).
Figure 2: Qualitative comparison of CSG methods across five criteria: High Quality, Consistency, Logical, Multi-Subjects, and Generalizability. Our method excels in visual fidelity, identity preservation, narrative coherence, and generalization — outperforming baselines that exhibit drift, incoherence, or stylistic inconsistency.
Figure 3: Overview of AnimeAgent. The Director Agent builds a hierarchical textual dope sheet from user inputs. The Artist Agent generates a continuous motion trajectory using an I2V model (“Straight Ahead”). A Consistency Reviewer validates keyframes for identity and layout, enabling refinement, while a Subjective Critic selects the most expressive “Extremes” via “Pose-to-Pose” evaluation, producing a coherent and visually compelling storytelling.
Figure 4: “Pose-to-Pose vs. Straight Ahead in Disney”
Figure 5: Text-Guided Semantic Anchor First Frame.
Figure 6: Layout-Robust Multi-Character I2V.
Methods
CSD ↑
CIDS ↑
PA ↑
CM ↑
Aes ↑
Cross
Self
Cross
Self
Scene
Shot
CI
IA
Avg.
StoryGen Liu et al. (2024)
0.371
0.531
0.417
0.568
1.26
2.29
1.42
1.71
1.67
50.8
4.02
TheaterGen Cheng et al. (2024)
0.184
0.392
0.348
0.578
2.76
1.81
0.93
1.06
1.64
55.4
4.90
StoryDiffusion Zhou et al. (2024)
0.340
0.547
0.436
0.565
2.00
3.45
2.64
2.68
2.69
57.4
5.13
Anim-Director Li et al. (2024b)
0.288
0.510
0.401
0.578
3.64
3.07
3.32
2.69
3.18
67.4
5.59
Story-Adapter Mao et al. (2024)
0.325
0.737
0.401
0.626
1.93
3.28
2.55
2.46
2.56
59.3
4.89
Table 1: Quantitative Results of Various CSG Methods on ViStoryBench Zhuang et al. (2025) . CI: Character Interaction, IA: Individual Action; The first , second , and third values are highlighted.
Figure 7: Identifying EXTREMENS via Motion Score.
Methods
CSD ↑
CIDS ↑
PA ↑
CM ↑
Aes ↑
Cross
Self
Cross
Self
Scene
Shot
CI
IA
Avg.
MOKI Meitu Inc. (2024)
0.214
0.694
0.372
0.621
1.88
0.94
2.88
1.10
1.70
46.0
5.79
MorphicStudio Morphic, Inc. (2024)
0.577
0.628
0.603
0.677
3.07
2.66
3.36
2.11
2.80
60.8
4.96
Aibrm MagicLight AI (2025)
0.412
0.730
0.557
0.740
3.05
2.53
3.56
2.05
2.80
75.5
5.72
ShenBi Maoyan Entertainment (2025)
0.275
0.575
0.418
0.585
2.81
3.45
3.79
2.72
3.19
61.3
5.07
Typemovie TypeMovie Team (2024)
0.325
0.646
0.464
0.621
3.00
2.61
2.86
2.30
2.69
74.1
5.32
Table 2: Quantitative Results of Various Commercial Platforms on ViStoryBench Zhuang et al. (2025) .
Methods
CSD ↑
CIDS ↑
PA ↑
CM ↑
Aes ↑
CLIP-I ↑
Cross
Self
Cross
Self
Scene
Shot
CI
IA
Avg.
StoryDiffusion Zhou et al. (2024)
0.422
0.485
0.443
0.461
2.26
2.77
2.59
2.45
2.52
52.6
5.14
0.615
Anim-Director Li et al. (2024b)
0.396
0.431
0.487
0.443
3.16
3.27
2.75
2.85
3.01
54.6
5.48
0.585
OmniGen2 Wu et al. (2025b)
0.515
0.628
0.627
0.591
3.45
2.97
2.83
2.72
2.99
58.9
5.23
0.704
FreeCus Zhang et al. (2025a)
0.474
0.538
0.531
0.576
2.95
2.86
2.44
2.52
2.69
55.3
5.08
0.662
Qwen-ImageEdit Wu et al. (2025a)
0.532
0.645
0.611
0.628
3.36
3.32
3.45
2.91
3.26
67.5
5.41
0.751
Table 3: Quantitative Results of Various CSG Methods on AnimeBoard-GT.
Figure 8: Qualitative Comparison Across Methods on Alice Story.
Figure 9: Qualitative Comparison Across Methods on Aladdin Story.
Figure 10: Qualitative Comparison Across Methods on Magic Instruction.
Adapting existing manga into coherent animation that stays faithful to the source is a valuable yet challenging task. Existing methods either access the manga mainly through text, losing its characters and art style, or feed manga pages directly into video generation models, which misinterpret the manga's structure and produce incorrect storylines. In this paper, we propose Manga2Animate, a multi-agent framework that grounds every stage of generation in the source manga. We decompose the task into two intermediate representations, an animation script and a visual storyboard. Each representation is produced by a creator agent and reviewed against the source panels by a paired reviewer agent. We tailor representation-specific selection and dual-level feedback to progressively refine both before synthesizing the final animation. Extensive experiments and a user study demonstrate that Manga2Animate produces animations more faithful to the source manga and more consistent across shots than state-of-the-art approaches. Project page: https://annchenn.github.io/Manga2Animate/
Sheng-Wei Zhang, Hsi-An Chen, Yan-Ting Liou +3
Department of Computer Science, National Yang Ming Chiao Tung University
We propose Camera Artist, a multi-agent framework that models a real-world filmmaking workflow to generate narrative videos with explicit cinematic language. While recent multi-agent systems have made substantial progress in automating filmmaking workflows from scripts to videos, they often lack explicit mechanisms to structure narrative progression across adjacent shots and deliberate use of cinematic language, resulting in fragmented storytelling and limited filmic quality. To address this, Camera Artist builds upon established agentic pipelines and introduces a dedicated Cinematography Shot Agent, which integrates recursive storyboard generation to strengthen shot-to-shot narrative continuity and cinematic language injection to produce more expressive, film-oriented shot designs. Extensive quantitative and qualitative results demonstrate that our approach consistently outperforms existing baselines in narrative consistency, dynamic expressiveness, and perceived film quality.
Haobo Hu, Qi Mao, Yuanhang Li +1
School of Information and Communication Engineering, Communication University of China, Beijing, China · State Key Laboratory for Multimedia Information Processing, School of Computer Science, Peking University, Beijing, China
Long-form visual storytelling requires maintaining continuity across shots, including consistent characters, stable environments, and smooth scene transitions. While existing generative models can produce strong individual frames, they fail to preserve such continuity, leading to appearance changes, inconsistent backgrounds, and abrupt scene shifts. We introduce CANVAS (Continuity-Aware Narratives via Visual Agentic Storyboarding), a multi-agent framework that explicitly plans visual continuity in multi-shot narratives. CANVAS enforces coherence through character continuity, persistent background anchors, and location-aware scene planning for smooth transitions within the same setting We evaluate CANVAS on two storyboard generation benchmarks ST-BENCH and ViStoryBench and introduce a new challenging benchmark HardContinuityBench for long-range narrative consistency. CANVAS consistently outperforms the best-performing baseline, improving background continuity by 21.6%, character consistency by 9.6% and props consistency by 7.6%.