Reproduce “StoryState: Agent-Based State Control for Consistent and Editable Storybooks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ayushman Sarkar, Zhenyu Yu, Wei Tang, Chu Chen, Kangning Cui, Mohd Yamani Idna Idris
StoryState introduces an agent-based framework that enables precise, page-level editing of multi-page illustrated storybooks while maintaining cross-page visual consistency. It achieves this by externalizing character attributes, world settings, and per-page content into an explicit, editable story state, outperforming baseline methods in editing efficiency and perceived control while maintaining high consistency.
Key insights
- Externalizing the implicit "story state" into a structured, explicit representation is crucial for enabling fine-grained control and targeted edits in multi-page generative tasks.
- An agent-based orchestration layer can effectively manage this state, decoupling global identity invariants from page-specific content to preserve consistency during localized modifications.
- Employing selective regeneration based on state changes, rather than full story regeneration, significantly improves editing efficiency and reduces user effort.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.