SceneWeave: Audience-Guided Scene Exploration for Dramatic Adaptation
Authors: Aolin Li, Xiao Xu, Haihua Xie
Organizations: Department of Science Theatre, The Central Academy of Drama, Beijing, China · Hithink RoyalFlush Information Network Co., Ltd., Hangzhou, China · China Telecom Artificial Intelligence Technology (Beijing) Co., Ltd., Beijing, China · Beijing Institute of Mathematical Sciences and Applications (BIMSA), Beijing, China
Adapting prose into a script requires transforming narration and inner experience into scenes that convey relationships and psychological contradictions. Planning specifies intended effects but leaves their realization in dialogue and action unresolved. Drawing on screenwriters' accounts of revising scenes and their connections to the surrounding work, we present SceneWeave, which uses alternative scene drafts to revise the adaptation as a whole. It writes contrasting treatments of a focal scene, compares the interactions they create, and revises characterization, preceding preparation, subsequent development, and scene order before generating the complete script. The central contribution is to use actual scene writing to inform adaptation decisions and propagate their consequences across the work. We also develop a task-specific writing model and a human-preference model for comparing script drafts. The writing model learns from adaptation examples and human preferences; the preference model combines human and synthetic supervision. Blind reader comparisons favor SceneWeave over a baseline workflow and description-based controls, while content assessment indicates reduced narrative redundancy and strong psychological treatment.
Figures & tables
Figure 1: SceneWeave. The scene-first example develops two treatments of a mother–daughter encounter and uses interpretations of those scenes to revise the adaptation (4–5). All four approaches provide material for complete writing (6); anonymous reader preferences evaluate the resulting scenes (7). INT. denotes an interior location in the screenplay headings. π0 is the initial plan; y0,y1,y2 denote the initial, developed, and revised drafts.
Figure 2: Four adaptation strategies. Solid arrows indicate generation order; dashed arrows supply interpretations or written context to later scenes. Each strategy contributes material for plan revision and complete-script generation.
Comparison
Wins / Ties / Losses
Preference (%)
Works
SceneWeave vs. SuperWriter prompts
11 / 1 / 4
71.9
4
Scene exploration vs. detailed planning
5 / 0 / 3
62.5
2
Written vs. described input for plan revision
4 / 3 / 1
68.8
2
Table 1: Blind scene preferences from four theatre graduate students. Wins, ties, and losses count individual judgments; preference is the mean of work-level scores, with ties assigned 0.5. The Works column gives the number of distinct source works, each judged by all four readers.
Narrative
Character
Emotion
Logic
Conflict
Method
Redundancy (%) ↓
OOC (%) ↓
Voice (%) ↑
Grade ↑
Errors (%) ↓
Grade ↑
SceneWeave
0.05
0.00
69.64
2.90
1.33
2.60
SuperWriter prompts
0.09
0.00
71.10
2.87
1.33
2.67
Dramaturge workflow
2.77
0.00
65.28
2.73
1.33
2.57
Strong prompt
0.06
0.00
66.63
2.90
2.44
2.67
Table 2: Complete-script content assessment on fifteen matched works. Redundancy, out-of-character dialogue (OOC), and world-continuity errors are lower-is-better rates; character-specific voice is higher-is-better. Rates are percentages averaged across scripts. Emotion and Conflict use higher-is-better anchored grades (0–3), combining three criteria and two readings.
Baseline
Wins / Ties / Losses
Works
Preference (%)
95% CI
SuperWriter prompts
10 / 3 / 2
15
76.7
[57.1,93.3]
Dramaturge workflow
11 / 3 / 1
15
83.3
[66.7,96.4]
Strong prompt
8 / 4 / 3
15
66.7
[46.4,85.7]
Table 3: Complete-script predictions of the frozen human-preference model. Wins, ties, and losses count predicted outcomes for SceneWeave; preference counts ties as half a win. Author-cluster intervals use 10,000 bootstrap samples. The same model selects candidates and predicts comparison outcomes.
Movie screenplays are rich long-form narratives that interleave complex character relationships, temporally ordered events, and dialogue-driven interactions. While prior benchmarks target individual subtasks such as question answering or dialogue generation, they rarely evaluate whether models can construct a coherent story world and use it consistently across multiple forms of reasoning and generation. We introduce STAGE (Screenplay Text, Agents, Graphs and Evaluation), a unified benchmark for narrative understanding over full-length movie screenplays. STAGE defines four tasks: knowledge graph construction, scene-level event summarization, long-context screenplay question answering, and in-script character role-playing, all grounded in a shared narrative world representation. The benchmark provides cleaned scripts, curated knowledge graphs, and event- and character-centric annotations for 150 films across English and Chinese, enabling holistic evaluation of models' abilities to build world representations, abstract and verify narrative events, reason over long narratives, and generate character-consistent responses.
Qiuyu Tian, Zequn Liu, Yiding Li +9
Southeast University, Nanjing, China · Beijing Zhongguancun Academy, Beijing, China · ZhuiWen Technology Co., Ltd., Beijing, China +1
Between a screenplay and a film sits a planning problem that is spatial first: who stands where, and what a camera sees from where it stands. An image diffusion model asked for a shot in free text settles that plan by its own defaults. We present PACE (Precise AI Cinematic Expression), a typed representation for the plan: the screenplay evidence, the characters, props and locations it needs, where each subject stands, and what the camera does. A value is written once at the level it belongs to (script, scene, shot or panel) and inherited below it. A compiler turns the result into both the prompt sent to the diffusion model and a 3D scene built in metres, and a camera solver places the camera so that the declared framing is the framing built. Where a declared value becomes geometry, PACE measures, field by field, how far the compiled camera and the staged render sit from the declaration, rather than asking a model to judge. On the 11-scene Automatic Drive screenplay, every staged single-subject panel places its subject within 1.2% of frame width of its declared position; with two or three subjects one camera pose cannot satisfy every position, and the residual is reported rather than absorbed. On 204 external director-storyboard shots, delivered head height is 1.906 times the staged target from the director's words, 1.733 from the compiled prompt, and 0.955 with the greybox control; the condition that holds framing best draws the described action least. Declaring the pose on 30 shots raises the action drawn from 58.9% to 74.4% without moving the framing. Transitions, fitted motion and human review of the generated panels remain open. Code: https://github.com/StudioPiLabs/pace-core
Long-form narrative-to-film generation requires shot-level controllability and cross-clip consistency in both visual identity and character behavior-requirements that remain difficult to satisfy with current prompt-based workflows. A core reason existing workflows remain brittle is the lack of a structured intermediate layer between scripts and video models, especially when screenplays are underspecified at key cinematic decision points. We introduce a structured orchestration layer for film-oriented script-to-video generation, implemented as a multi-agent framework that operates between scripts and off-the-shelf video generators. The layer is centered on FilmDSL, a film-oriented domain-specific language that makes cinematic constraints explicit, including shot and camera directives, asset and continuity requirements, and persona cues, so that agents coordinate through a shared structured specification for planning, generation, critique, and repair. Specifically, a generation agent constructs asset packs and storyboard keyframes that anchor composition before clip-by-clip synthesis, while a critic agent produces structured QA signals and triggers targeted refinement without retraining the base model. Experiments on TV-style segments show improved controllability and consistency over text-only and reference-only baselines.
Jiaben Chen, Sixun Dong, Qinhong Zhou +4
University of Massachusetts Amherst, Amherst, MA 01003, USA · Massachusetts Institute of Technology, Cambridge, MA 02139, USA