This paper introduces LLM-based interactive drama, a new form of storytelling that allows users to immerse themselves in a narrative and interact with characters and scenes. The authors define six essential elements of interactive drama—plot, character, thought, diction, spectacle, and interaction—and propose a prototype drama script to guide LLMs in performing this interactive narrative. They address challenges such as limited drama resources, uncontrollable narrative development, and complex instruction following by introducing several techniques: Narrative Chain for finer control over narrative progression, Auto-Drama for generating drama scripts from arbitrary stories, and Sparse Instruction Tuning to enhance instruction following. The paper evaluates the performance of the trained drama LLMs using a comprehensive 5-dimensional evaluation framework. The results show that the LLMs excel in engaging in dialogue, generating rich narratives, and handling plot progression, with notable improvements in guidance and scene presentation. The paper also discusses limitations and future directions, including the need for more immersive modalities and a robust evaluation method.This paper introduces LLM-based interactive drama, a new form of storytelling that allows users to immerse themselves in a narrative and interact with characters and scenes. The authors define six essential elements of interactive drama—plot, character, thought, diction, spectacle, and interaction—and propose a prototype drama script to guide LLMs in performing this interactive narrative. They address challenges such as limited drama resources, uncontrollable narrative development, and complex instruction following by introducing several techniques: Narrative Chain for finer control over narrative progression, Auto-Drama for generating drama scripts from arbitrary stories, and Sparse Instruction Tuning to enhance instruction following. The paper evaluates the performance of the trained drama LLMs using a comprehensive 5-dimensional evaluation framework. The results show that the LLMs excel in engaging in dialogue, generating rich narratives, and handling plot progression, with notable improvements in guidance and scene presentation. The paper also discusses limitations and future directions, including the need for more immersive modalities and a robust evaluation method.