A Persona-Augmented Multi-Agent System for Varied Narrative Generation

Large Language Models growing abilities in writing task has allowed them to also tap in the world of creative writing. Instead of following explicit user instructions to operate on a given text, they can be asked to generate complete stories from a single prompt. Although these models yield impressive capabilities in producing well-written narratives, they often lack diversity, relying on established style and semantics seen at training time. Considering this challenge, we present an on-demand persona-augmented agents for narrative generation. By dynamically crafting agents and personas at run-time without relying on hard-coded agent architectures with fixed roles, the system can adapt to match the specific demands of writing prompts. Leveraging LLM persona as proxy for semantic, tone and lexical the system can automatically define tasks demands and workflow, thus crafting more varied and heterogeneous outputs. Our approach demonstrates that agentic platforms can consistently surpass the single-LLM baseline. With gains of +0.28 in semantic and +0.18 in style embedding distances, the generated outputs by our best proposal, exhibit higher variety, validating the effectiveness of multi-agent persona augmentation for open writing tasks.