Reproduce “AniMaker: Multi-Agent Animated Storytelling with MCTS-Driven Clip Generation” 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
Haoyuan Shi, Yunxin Li, Xinyu Chen, Longyue Wang, Baotian Hu, Min Zhang
AniMaker presents a multi-agent framework that generates coherent, multi-character, multi-scene animated stories from text, utilizing MCTS-driven clip generation and a novel multi-shot animation evaluation framework (AniEval). The system demonstrates enhanced visual and narrative consistency and superior action representation, achieving a 14.6% improvement in AniEval scores over baselines and significantly higher human evaluation ratings.
Key insights
- A modular, multi-agent architecture can effectively manage the complexity of generating long-form animated narratives from text.
- Adapting Monte Carlo Tree Search (MCTS-Gen) allows for efficient exploration and selection of high-quality video clips, addressing computational challenges in multi-candidate generation.
- Comprehensive, context-aware evaluation (AniEval), which assesses clips in relation to their narrative sequence, is essential for achieving visual and narrative coherence in multi-shot animation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.