StageDesigner: Artistic Stage Generation for Scenography via Theater Scripts

TL;DR

StageDesigner introduces a novel, training-free pipeline that converts theater scripts into immersive stage designs by coupling LLM-based script interpretation with layout-controlled diffusion for backgrounds. It explicitly models audience sightlines via a Foreground Projection and enforces spatial coherence with a Multi-level Collision Map, enabling believable foreground layouts and unobstructed backgrounds. The StagePro-V1 dataset provides 276 professionally curated stage scenes with scripts and 3D layouts, enabling rigorous assessment and benchmarking against a LayoutGPT baseline. Across quantitative metrics, qualitative analyses, and user studies, StageDesigner demonstrates superior layout coherence, background fitness, and overall design preference, indicating strong potential for rapid, artist-guided scenography workflows in both research and professional contexts.

Abstract

In this work, we introduce StageDesigner, the first comprehensive framework for artistic stage generation using large language models combined with layout-controlled diffusion models. Given the professional requirements of stage scenography, StageDesigner simulates the workflows of seasoned artists to generate immersive 3D stage scenes. Specifically, our approach is divided into three primary modules: Script Analysis, which extracts thematic and spatial cues from input scripts; Foreground Generation, which constructs and arranges essential 3D objects; and Background Generation, which produces a harmonious background aligned with the narrative atmosphere and maintains spatial coherence by managing occlusions between foreground and background elements. Furthermore, we introduce the StagePro-V1 dataset, a dedicated dataset with 276 unique stage scenes spanning different historical styles and annotated with scripts, images, and detailed 3D layouts, specifically tailored for this task. Finally, evaluations using both standard and newly proposed metrics, along with extensive user studies, demonstrate the effectiveness of StageDesigner. Project can be found at: https://deadsmither5.github.io/2025/01/03/StageDesigner/

Figures (16)

• Figure 1: Human Stage Design: Traditional stage design requires manual script analysis, collaborative planning, and model creation, making it time-consuming and expertise-dependent. StageDesigner: StageDesigner automates stage design by transforming scripts into 3D foreground elements and then placing background elements in unobstructed areas within the audience’s line of sight. Application: General users can refine stage designs through conversational edits by feeding results back into the LLM, while professionals can import designs into Blender for detailed adjustments, real model creation, and performance preparation.
• Figure 2: Overview of the StageDesigner pipeline. StageDesigner transforms an input theater script into a 3D stage layout through three main modules: (1) Script Analysis extracts key scene and imagery details; (2) Foreground Generation creates and places stage entities, retrieves corresponding 3D assets, and ensures spatial coherence using a multi-level collision map; (3) Background Generation produces a background image that complements the scene, guided by thematic elements and avoiding occlusions with foreground objects.
• Figure 3: The example of collision map. (a) Collision map of stage floor. (b) Collision map of wall’s front surface.
• Figure 4: The stage samples in our dataset. (a) The Cruel Game, (b) Beyond the Horizon, (c) Lend Me a Tenor, (d) Shooting Star.
• Figure 5: Qualitative comparison between LayoutGPT and StageDesigner. The first row shows stages generated by LayoutGPT, and the second row shows stages generated by StageDesigner. StageDesigner results show better entity placement and atmospheric expression. e.g." Family Portrait " denotes the theater script.