DuoDrama: Supporting Screenplay Refinement Through LLM-Assisted Human Reflection

TL;DR

DuoDrama tackles the challenge of guiding screenplay refinement by coordinating internal character experience with external narrative evaluation through a two-perspective, performance-theory–inspired workflow called ExReflect. The system uses a multi-agent architecture where each character is embodied in an experience role and then evaluated from an external perspective, producing experience-grounded, contextually aligned feedback. A formative study identified four design goals, and a two-session user study with 14 professional screenwriters showed that DuoDrama improves feedback quality, alignment, and the depth and richness of user reflection, while balancing immersion and critical distance. The work advances AI-assisted reflection in creative writing by bridging inner psychology and outer narrative structure, with implications for broader domains requiring grounded experiential feedback and adaptive, two-perspective evaluation. It also suggests future extensions to multi-modal grounding, adaptive orchestration, and alternative reasoning paradigms to further empower human–AI collaborative reflection.

Abstract

AI has been increasingly integrated into screenwriting practice. In refinement, screenwriters expect AI to provide feedback that supports reflection across the internal perspective of characters and the external perspective of the overall story. However, existing AI tools cannot sufficiently coordinate the two perspectives to meet screenwriters' needs. To address this gap, we present DuoDrama, an AI system that generates feedback to assist screenwriters' reflection in refinement. To enable DuoDrama, based on performance theories and a formative study with nine professional screenwriters, we design the Experience-Grounded Feedback Generation Workflow for Human Reflection (ExReflect). In ExReflect, an AI agent adopts an experience role to generate experience and then shifts to an evaluation role to generate feedback based on the experience. A study with fourteen professional screenwriters shows that DuoDrama improves feedback quality and alignment and enhances the effectiveness, depth, and richness of reflection. We conclude by discussing broader implications and future directions.

Figures (5)

• Figure 1: The Experience-Grounded Feedback Generation Workflow for Human Reflection (ExReflect). A single agent sequentially enacts two stakeholder roles from the same scenario. In the experience role, the agent adopts a stakeholder persona and interaction context to generate personal experiences. It then shifts to the evaluation role, adopting a different stakeholder perspective to generate feedback based on that personal experience. This design grounds feedback in personal experience while introducing evaluation distance, thereby balancing internal immersion with external critique.
• Figure 2: The detailed requirements for ExReflect in the DuoDrama system. A single agent enacts two complementary stakeholder roles within the screenplay refinement process. First, it acts as a character in the screenplay to generate the character’s current inner thoughts and its current scene text, where the current inner thoughts refer to this character produced in the present turn, and the current scene text aggregates all inner thoughts generated by this character in previous turns. Together, these form the agent’s short-term memory as personal experience (experience role). Second, the agent shifts to the actor portraying that character to generate questions, using the personal experience as an additional input that supplements the interaction context (evaluation role). The agent then applies the evaluation metrics to assess these questions, and those that satisfy the criteria become the final feedback presented to the user. For detailed evaluation metrics, please refer to Sec. \ref{['sec:dual_memory']} and the supplementary material.
• Figure 3: The panels in the frontend interface of the DuoDrama, mapped to design goals (DGs). Control Panel manages screenplay materials and roles to set the stage for high-quality simulation (DG1) and ensure screenplay alignment (DG2). Enactment Panel (DG1, DG2, DG3) presents line-by-line dialogues and inner thoughts to support high-quality simulation (DG1) and contextual alignment (DG2), while triggering creative insight and refinement willingness (DG3) via instant feedback. Critique Panel (DG1, DG2, DG3) lists post-hoc feedback to facilitate critical understanding (DG1) and checks contextual alignment (DG2) to trigger creative insight and motivate refinement willingness (DG3). Value Marking Function (DG3, DG4) enables users to identify and retain useful AI outputs for refinement (DG3), while being designed to balance intervention timing with creative flow (DG4).
• Figure 4: The Process and Features of the DuoDrama. Step 1: Uploading and customizing material. Step 2: Assigning role identities. Step 3: Experiencing characters' inner thoughts. Step 4: Reviewing actor' s feedback feedback. This includes: (a) instant feedback that appear directly below relevant inner thoughts, and (b) a series of post-hoc feedback provided after the scene simulation. Step 5: Marking valuable content.
• Figure 5: Results of the user experience survey in Session 1, showing participant ratings on a 7-point Likert scale. The questions are grouped by four DGs: (a) DG1 - High-Quality Embodied Simulation and Understandable Critical Feedback (Q1, Q2, Q5, Q8); (b) DG2 - Alignment Between Simulation, Reflection, and Screenplay (Q3, Q6, Q7); (c) DG3 - Stimulating Creative Insight and Refinement Willingness (Q4, Q9, Q10); and (d) DG4 - Balancing Intervention Frequency and Creative Flow in Feedback Timing (Q11, Q12, Q13). In this figure, "questions" in the survey items refer to the feedback questions generated by DuoDrama.