IBSEN: Director-Actor Agent Collaboration for Controllable and Interactive Drama Script Generation

TL;DR

IBSEN addresses the challenge of constraining multi-agent drama generation by introducing a director agent that drives storyline objectives $G_i$, generates outlines $S_G$, and instructs actor agents while optionally incorporating human players. The framework uses memory-augmented actor agents with monologue-based retrieval and a narrative dialogue paradigm to maintain character consistency and plot direction, all without fine-tuning. Through a Hedda Gabler-inspired experiment, IBSEN demonstrates controllable script generation, player adaptability, and measurable storyline quality, albeit with limitations in complex acts and potential repetition. The work contributes a centralized, prompt-based architecture for interactive and controllable drama, with implications for AI-driven storytelling and game-like simulations across domains.}

Abstract

Large language models have demonstrated their capabilities in storyline creation and human-like character role-playing. Current language model agents mainly focus on reasonable behaviors from the level of individuals, and their behaviors might be hard to constraint on the level of the whole storyline. In this paper we introduce IBSEN, a director-actor coordinate agent framework that generates drama scripts and makes the plot played by agents more controllable. The director agent writes plot outlines that the user desires to see, instructs the actor agents to role-play their characters, and reschedules the plot when human players participate in the scenario to ensure the plot is progressing towards the objective. To evaluate the framework, we create a novel drama plot that involves several actor agents and check the interactions between them under the instruction of the director agent. Evaluation results show that our framework could generate complete, diverse drama scripts from only a rough outline of plot objectives, meanwhile maintaining the characteristics of characters in the drama. Our codes and prompts are available at https://github.com/OpenDFM/ibsen.

Figures (7)

• Figure 1: IBSEN is an agent framework that uses director and actor agents to generate controlled drama scripts.
• Figure 2: Overall framework of IBSEN and an illustration of director agent's controlling process in a certain scene. Director agent ① writes an outline to continue the previous plot and ② translate it into the dialogue script, then ③ instructs the actor agent to generate a proper response. Actor agent in the next dialogue turn ④ generates the response, and the director agent ⑤ checks whether the dialogue has reached the current plot objective.
• Figure 3: An example of maintaining the dialogue log and updating the memory database during the play. Actor agent summarizes previous contexts in points, and adds them to the memory database. Each memory content is embedded as the retrieval index of the document, and has a monologue that is interpreted by the actor.
• Figure 4: An interactive roadmap showing the collaboration between director agent and characters in the scene. Actor agents' actions are instructed by the director agent to follow the storyline generated by the director. When human players involve in the storyline, director agent will actively generate new storylines to adapt to player actions.
• Figure 5: Effects of IBSEN unique components on the storyline performance. We use the same metric to score the storyline from 1 to 4.