StoryWeaver: A Unified World Model for Knowledge-Enhanced Story Character Customization

arXiv 24.12 2024 61.3 method

TLDR

StoryWeaver uses a Character Graph knowledge base for consistent story visualization with character identity preservation and text alignment.

Reasoning

The paper introduces a novel knowledge graph and image generator for story visualization, with strong empirical results on a new benchmark. However, it lacks discussion of world dynamics or interactive elements, limiting its scope to static image generation.

Read-first score

Read-first score 61.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 19.

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=ablation,benchmark,dataset,experiment,metric

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=code,dataset,github

Recency 8%

75.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2024

Topical relevance 42%

27.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Field roles

Methodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 759.

Keyword Scores

world model

generative world model

video world model

world simulator

interactive world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Character Graph (CG) for comprehensive representation of story-related knowledge including characters, attributes, and relationships
StoryWeaver with Customization via Character Graph (C-CG) for consistent story visualization with rich text semantics
Knowledge-Enhanced Spatial Guidance (KE-SG) to improve multi-character generation by precisely injecting character semantics
New benchmark TBC-Bench for evaluating story visualization methods

Methodology

StoryWeaver introduces a knowledge graph called Character Graph (CG) that encodes characters, their attributes, and relationships. The image generator uses Customization via Character Graph (C-CG) to achieve consistent story visualization, and incorporates Knowledge-Enhanced Spatial Guidance (KE-SG) to improve multi-character generation. The method is evaluated on a new benchmark TBC-Bench using metrics DINO-I and CLIP-T.

Key Results

StoryWeaver achieves an average increase of +9.03% in DINO-I and +13.44% in CLIP-T, demonstrating superior character identity preservation and text-semantics alignment across various scenarios.

Resources

Paper Code