Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

FilmAgent: A Multi-Agent Framework for End-to-End Film Automation in Virtual 3D Spaces

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A coordinated team of specialized LLM agents can automate end-to-end filmmaking in virtual 3D spaces and, according to the paper's human evaluation, scores 3.98/5 while beating a single-agent o1 baseline.

desk verdict A real, well-scoped systems paper on multi-agent LLM filmmaking whose central comparison is plausible but under-reported statistically; worth reviewing, but the evaluation needs to be redone. read the letter →

arxiv 2501.12909 v1 pith:JFPUKPOL submitted 2025-01-22 cs.CL cs.GRcs.MA

Zhenran Xu , Longyue Wang , Jifang Wang , Zhouyi Li , Senbao Shi , Xue Yang , Yiyu Wang , Baotian Hu
show 2 more authors
Jun Yu Min Zhang
This is my paper · ORCID
classification cs.CLcs.GRcs.MA
keywords multi-agentLLMfilmautomationvirtualproductionscriptwritingcinematographyhumanevaluationhallucinationreduction3Dspaces
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that filmmaking, treated as a communication-driven collaborative task, can be automated end-to-end by a team of large-language-model agents assigned to film-crew roles. It introduces FilmAgent, where a director, screenwriter, actors, and cinematographers move from a one-line story idea to a fully annotated script and then to rendered video inside pre-built 3D virtual spaces. Human evaluation across 15 ideas gives the full system an average score of 3.98 out of 5, above all single-agent baselines on the four judged aspects, and the paper reports that this multi-agent setup built on GPT-4o outperforms a single-agent o1 setup. If true, this would show that role specialization and iterative critique can substitute for raw model capability in a creative production task.

What carries the argument

The operational unit is the annotated script: a structured representation that binds every dialogue line to the speaking character's position, an action from a fixed vocabulary of 21 gestures, and one of 9 static or dynamic camera shots, so the whole film can be simulated automatically inside the 3D space. The two agent-interaction protocols are the load-bearing mechanisms: Critique-Correct-Verify, which forces a critic to check actions, theme consistency, and plot flow before the writer revises and the critic re-verifies, and Debate-Judge, which makes two cinematographers propose shots, critique each other, and submit to a director's decision. These loops use the constrained action and shot vocabularies as hard constraints, and the paper credits the iterative feedback with reducing hallucinations, improving plot coherence, and diversifying camera choices.

What would settle it

Run a preregistered human evaluation with multiple blind annotators per video, report inter-annotator agreement and confidence intervals, keep the theme-fidelity aspect in the scored set, and resample the action-accuracy check repeatedly from all generated actions; if the average falls to or below the o1 baseline, or the action-accuracy gap between the group and solo configurations disappears, the central claim would fail.

Watch

Extended reading notes

Core claim

Filmmaking is modeled as a language-driven pipeline with three stages: idea development, scriptwriting, and cinematography. A director agent generates character profiles and scene outlines, a screenwriter drafts dialogue and annotates each line with positions and one of 21 allowed actions, actor agents review their own lines against their character profiles, and cinematographer agents choose among 9 predefined static and dynamic shot types. Two collaboration algorithms carry the improvement: Critique-Correct-Verify, in which a critic reviews, the action agent revises, and the critic verifies, and Debate-Judge, in which two cinematographers propose and critique shots before a director issues the final camera plan. On 15 ideas, human annotators award the collaborative group configuration scores of 0.88 for action accuracy, 3.53 for plot coherence, 4.44 for profile alignment, and 3.53 for camera appropriateness, averaging 3.98 out of 5 and exceeding every single-agent baseline. The paper's central claim is that a well-coordinated multi-agent system of weaker models can outperform a single, more advanced reasoning model on this multi-step creative task.

Load-bearing premise

The headline result rests on the human ratings: the paper does not report how many annotators evaluated the videos, whether they were blind to which system produced each film, or how much their scores agreed, and it drops the theme-fidelity aspect after a preliminary check; if those ratings are unreliable, the claimed 3.98 average and the advantage over o1 are not established.

Editorial extensions

If this is right

  • From a single brainstormed idea, FilmAgent can output a rendered short film without human intervention beyond rating, within the limits of its pre-built 3D spaces.
  • The fixed 3D stage, action vocabulary, and shot vocabulary make the pipeline verifiable: invalid actions and scene-to-scene contradictions can be caught by the director agent before filming.
  • The comparison with a single-agent o1 suggests a practical route to improving creative LLM systems through orchestration and role specialization rather than only scaling the underlying model.
  • The Sora comparison indicates that for coherent, physics-compliant storytelling, a constrained 3D pipeline can offer consistency that open-ended text-to-video generation does not yet reliably provide.
  • The win-rate analysis of revised versus original scripts and camera plans gives direct evidence that the critique and debate steps, not just prompt design, drive the reported quality gains.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The evaluation covers only 15 ideas and 50 sampled actions, so an immediate extension would test how the framework degrades when the 3D environment, action set, or shot set is expanded or when longer films are requested.
  • The collaboration protocols are not film-specific; a controlled comparison on non-film planning or content-creation tasks could isolate whether Critique-Correct-Verify and Debate-Judge generalize beyond the film domain.
  • The paper itself notes that all verification is text-based, so adding a multimodal critic that actually watches rendered frames would be a natural next step toward catching visual errors that text reviews miss.
  • A cost and token-usage comparison between the multi-agent GPT-4o pipeline and the single-agent o1 pipeline would clarify whether the quality advantage remains practical outside the research setting.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes FILMAGENT, an LLM-based multi-agent framework for end-to-end film production in pre-built Unity 3D environments. The framework assigns distinct crew roles (director, screenwriter, actors, cinematographers) and organizes work into three stages: idea development, scriptwriting (with a Critique-Correct-Verify collaboration loop), and cinematography (with a Debate-Judge collaboration loop). The authors evaluate videos generated from 15 hand-authored story ideas by human raters on four aspects (action accuracy, plot coherence, actor-profile alignment, camera appropriateness), reporting an average score of 3.98/5 for the full multi-agent system, which they claim outperforms all baselines, including a single-agent o1 model. They also include a qualitative comparison with OpenAI's Sora.

Significance. If the empirical claims are established, the work is a useful demonstration that role-specialized multi-agent LLM collaboration can outperform a stronger single-agent model on a creative, end-to-end production task. The framework is well-specified, with detailed prompts, environment design, and accessible code/videos, and the paper addresses an underexplored application area. However, the significance of the comparative claims depends entirely on the human evaluation, and the current manuscript does not provide the statistical support needed to validate them. The qualitative Sora comparison is illustrative rather than evidence.

major comments (4)
  1. [Section 4.1 / Table 2] Table 2 and the associated text report only mean scores. The manuscript does not state the number of annotators, inter-annotator agreement, confidence intervals, or significance tests, yet Section 1 says FILMAGENT 'significantly outperforms' baselines and the Abstract says it 'outperforms all baselines across all aspects.' At the reported scale, the Action accuracy differences (0.68 vs 0.80 vs 0.88) are within binomial sampling error if 50 actions per condition are evaluated. Please report the full evaluation protocol (annotator count, blinding, agreement, per-condition sample sizes) and provide confidence intervals or significance tests, or weaken the comparative claims to be descriptive rather than inferential.
  2. [Section 4.1, fifth evaluation aspect] The evaluation scheme initially lists five aspects, but 'script fidelity to the intended theme' is dropped after a preliminary study reported only as 'we found that all scripts faithfully adhered to the intended story ideas.' No data, criteria, or analysis supporting this finding are shown. Excluding an aspect post hoc without evidence can bias the aggregate score and make the reported 3.98 not comparable to a hypothetical full evaluation. Please either retain the aspect in the reported results or provide the preliminary data and explicit justification for the exclusion.
  3. [Section 4.1, o1 baseline] The o1-preview baseline was accessed through the ChatGPT webpage instead of the API, so decoding parameters (e.g., temperature, top-p, seed) and system-prompt controls are not matched across conditions. Because the paper's headline comparison is FILMAGENT (GPT-4o) versus single-agent o1, the unmatched inference protocol is a confounding factor. Please re-run the o1 baseline using the API with sampling settings matched to the other conditions, and report the exact model version and access date.
  4. [Section 4.1, action accuracy sampling] The description 'we randomly select 50 actions from the generated scripts' is ambiguous: it does not specify whether the 50 actions are sampled per condition or shared across conditions, nor does it describe the annotation unit or reliability for these binary judgments. Under either reading, the sample size is small relative to the reported differences; for instance, an accuracy of 0.88 versus 0.80 on 50 items is a four-item difference with overlapping binomial confidence intervals. Please specify the sampling and annotation procedure and provide per-condition counts and intervals.
minor comments (5)
  1. [Header] The manuscript is marked 'Work in Progress' on every page; for a journal or conference submission this label should be removed, as it indicates the manuscript is not in final form.
  2. [Section 4.2] There is a typo 'adavantage' in the sentence discussing the o1 comparison; it should be 'advantage'.
  3. [Figure 14 prompt] The action list contains 'Standing Claping'; this should be 'Standing Clapping' to match the list in Appendix A.
  4. [Appendix C] The script excerpt uses 'Track Shot' and 'Tracking Shot' inconsistently; standardize the terminology to match the shot list in Section 3.1 and Table 4.
  5. [Section 5, Sora comparison] The Sora comparison is presented as a case study but reads as a systematic evaluation; label it explicitly as a qualitative anecdote and note that the observations are from a small number of videos, not a controlled comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim rests on external human evaluations, not on fitted parameters, definitional identities, or load-bearing self-citations.

full rationale

I walked the paper's claimed derivation chain: FilmAgent proposes a multi-agent LLM workflow (idea development, scriptwriting, cinematography) and evaluates generated videos against CoT, Solo, and o1 baselines using human ratings on 15 story ideas. The 3.98/5 average and per-aspect comparisons come from Table 2, which reports human annotations of independently generated videos; no parameter is fitted from the evaluation scores, and no predicted quantity is defined in terms of the framework's own outputs. The multi-agent collaboration algorithms (Critique-Correct-Verify and Debate-Judge) are design choices, not derived results, and their claimed benefit is supported by human preference annotations rather than by construction. The paper cites several works by overlapping authors (e.g., Anim-Director, MultiSkill, multi-agent peer review collaboration), but these citations appear in related work and discussion sections and are not used as evidence for the headline evaluation claim; they are not load-bearing. The main caveats in the paper are methodological rather than circular: the human evaluation omits annotator counts, inter-annotator agreement, confidence intervals, and blinding, and the fifth evaluation aspect is dropped after a preliminary finding that all scripts adhered to the theme. These are validity and reporting concerns, not instances of the derivation reducing to its inputs. No self-definitional step, no fitted input called a prediction, no imported uniqueness theorem, and no ansatz smuggled in via citation were found. Accordingly, the circularity score is 0.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

No fitted model parameters are introduced; the only free parameter is the collaboration iteration cap M=3. The analysis relies on domain assumptions about prompt-following reliability, environment sufficiency, and human evaluation validity, none of which are empirically verified within the paper. No new physical or conceptual entities are postulated.

free parameters (1)
  • Maximum collaboration iterations M = 3
    Set to 3 in Section 4.1 (Implementation Details); a hand-chosen hyperparameter that controls how many critique/correct or debate rounds occur and could affect output quality.
assumptions (3)
  • domain assumption LLM agents (GPT-4o) can reliably follow the detailed JSON-format prompts and produce scripts, action annotations, and camera plans that are executable in the 3D environment.
    Invoked throughout Section 3.4 and the prompt list in Appendix D; if the model frequently violated the JSON or action constraints, the pipeline would fail.
  • domain assumption The pre-built Unity environments (15 locations, 65 positions, 21 Mixamo actions, 9 shot types, ChatTTS audio) are sufficiently expressive and neutral to represent the stories generated from the 15 ideas without unfairly constraining any condition.
    Stated in Section 3.1; the entire comparison across baselines assumes the environment does not bottleneck one method more than another.
  • domain assumption Human Likert ratings of plot coherence, profile alignment, and camera appropriateness, plus binary action-accuracy annotations, are valid and reliable measures of film quality without requiring reported inter-annotator agreement.
    Assumed in Section 4.1 (Evaluation Scheme); the headline scores rest on this assumption, yet no annotator count, agreement metric, or blinding procedure is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FilmAgent: A Multi-Agent Framework for End-to-End Film Automation in Virtual 3D Spaces." pith.science (2026). https://pith.science/paper/JFPUKPOL

@misc{pith2026250112909,
  author       = {Pith},
  title        = {Pith review of: FilmAgent: A Multi-Agent Framework for End-to-End Film Automation in Virtual 3D Spaces},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JFPUKPOL}},
  note         = {Machine review of arXiv:2501.12909}
}
read the original abstract

Virtual film production requires intricate decision-making processes, including scriptwriting, virtual cinematography, and precise actor positioning and actions. Motivated by recent advances in automated decision-making with language agent-based societies, this paper introduces FilmAgent, a novel LLM-based multi-agent collaborative framework for end-to-end film automation in our constructed 3D virtual spaces. FilmAgent simulates various crew roles, including directors, screenwriters, actors, and cinematographers, and covers key stages of a film production workflow: (1) idea development transforms brainstormed ideas into structured story outlines; (2) scriptwriting elaborates on dialogue and character actions for each scene; (3) cinematography determines the camera setups for each shot. A team of agents collaborates through iterative feedback and revisions, thereby verifying intermediate scripts and reducing hallucinations. We evaluate the generated videos on 15 ideas and 4 key aspects. Human evaluation shows that FilmAgent outperforms all baselines across all aspects and scores 3.98 out of 5 on average, showing the feasibility of multi-agent collaboration in filmmaking. Further analysis reveals that FilmAgent, despite using the less advanced GPT-4o model, surpasses the single-agent o1, showing the advantage of a well-coordinated multi-agent system. Lastly, we discuss the complementary strengths and weaknesses of OpenAI's text-to-video model Sora and our FilmAgent in filmmaking.

Figures

Figures reproduced from arXiv: 2501.12909 by the authors.

Figure 1
Figure 1. We introduce FILMAGENT, a multi-agent collaborative framework for end-to-end film automation powered by large language models (LLMs). A team of LLM-based agents takes on film crew roles, and simulates the human workflow in 3D virtual spaces by sequentially engaging in idea development, scriptwriting, and cinematography, finally completing the filmmaking process. ABSTRACT Virtual film production requires intricate de… view at source ↗
Figure 2
Figure 2. A vertical view of one of the 3D spaces (the living room) in FILMAGENT built with Unity. The environment is pre-configured with designated positions for actors and various camera setups for cinematography. These include static shots from multiple distances and dynamic shots that either follow or orbit around characters. Full camera setup of this space is provided in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Workflow of FILMAGENT. Given a story idea and 3D virtual spaces, the director creates character profiles and a scene outline. Actors, the screenwriter, and the director then collaborate on dialogue and movements. Cinematographers annotate camera setups for each line. Finally, the film is shot within the 3D spaces. LLM-based agents take on various film crew roles, collaborating through Critique-Correct-Verify and Deb… view at source ↗
Figures from the paper (21 more)
Figure 4
Figure 4. Figure 4: The responsibilities of a screenwriter extend beyond writing dialogues; they also involve [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Compared with the original version, the win, tie, and lose rates of the updated script and camera choices after multi-agent collaboration [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Comparison of videos showing “a quarrel and breakup scene” produced by F [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Locations in our constructed 3D spaces. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: A vertical view of the living room in FILMAGENT, with additional annotations of 9 shot types based on [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: A script excerpt ready for simulation in our 3D virtual spaces. [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Prompt for the idea development stage: The director generates character profiles. [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: Prompt for the idea development stage: The director generates the scene outline. [PITH_FULL_IMAGE:figures/full_fig_p022_11.png]
Figure 12
Figure 12. Figure 12: Prompt for the scriptwriting stage: The screenwriter drafts the dialogue in the script. [PITH_FULL_IMAGE:figures/full_fig_p023_12.png]
Figure 13
Figure 13. Figure 13: Prompt for the scriptwriting stage: The screenwriter assigns positions for characters. [PITH_FULL_IMAGE:figures/full_fig_p023_13.png]
Figure 14
Figure 14. Figure 14: Prompt for the scriptwriting stage: The screenwriter annotates actions for each line. [PITH_FULL_IMAGE:figures/full_fig_p025_14.png]
Figure 15
Figure 15. Figure 15: Prompt for the scriptwriting stage: The screenwriter adds movements between positions. [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: Prompt for the scriptwriting stage: The director provides feedback on action appropri [PITH_FULL_IMAGE:figures/full_fig_p026_16.png]
Figure 17
Figure 17. Figure 17: Prompt for the scriptwriting stage: The screenwriter updates the script based on the [PITH_FULL_IMAGE:figures/full_fig_p027_17.png]
Figure 18
Figure 18. Figure 18: Prompt for the scriptwriting stage: In Director-Screenwriter discussion, the director [PITH_FULL_IMAGE:figures/full_fig_p028_18.png]
Figure 19
Figure 19. Figure 19: Prompt for the scriptwriting stage: Actors provide feedback based on character profiles. [PITH_FULL_IMAGE:figures/full_fig_p028_19.png]
Figure 20
Figure 20. Figure 20: Prompt for the scriptwriting stage: The director filters and aggregates the feedback from [PITH_FULL_IMAGE:figures/full_fig_p029_20.png]
Figure 21
Figure 21. Figure 21: Prompt for the scriptwriting stage: In Actor-Director-Screenwriter discussion, the direc [PITH_FULL_IMAGE:figures/full_fig_p030_21.png]
Figure 22
Figure 22. Figure 22: Prompt for the cinematography stage: Cinematographer annotates camera setups for each [PITH_FULL_IMAGE:figures/full_fig_p032_22.png]
Figure 23
Figure 23. Figure 23: Prompt for the cinematography stage: Cinematographers debate over camera choices. [PITH_FULL_IMAGE:figures/full_fig_p033_23.png]
Figure 24
Figure 24. Figure 24: Prompt for the cinematography stage: The director judges the final camera setup. [PITH_FULL_IMAGE:figures/full_fig_p034_24.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. GroundShot: Visually Consistent Multi-Shot Long Video Generation via Entity-Grounded Shot Scheduling

    cs.CV 2026-06 unverdicted novelty 7.0 of 10

    GroundShot introduces entity-grounded shot scheduling with online visual memory to improve consistency in multi-shot video generation and presents GroundBench for entity-level evaluation.

Reference graph

Works this paper leans on

56 extracted references · 54 canonical work pages · cited by 1 Pith paper

  1. [1]

    - The name should only have one word

    Profile content: - The profile should include the name, gender (male or female), age, occupation, personality traits and speaking style. - The name should only have one word

    work page

  2. [2]

    name": "

    Number of characters: - Include no more than 4 main characters. Your response should only contain the following JSON content: [{"name": "...", "age": "...", "gender": "...", "occupation": "...", "personality traits": "...", "speaking style": "..." }, ...] Figure 10: Prompt for the idea development stage: The director generates character profiles. 21 Work ...

    work page

  3. [3]

    Roadside: maximum capacity: 2

    work page

  4. [4]

    Gaming room: maximum capacity: 4

    work page

  5. [5]

    Male: {male_characters}

    work page

  6. [6]

    Female: {female_characters} ### Main Locations:

    work page

  7. [7]

    Apartment living room: maximum capacity: 5

    work page

  8. [8]

    Apartment kitchen: maximum capacity: 5

    work page

Show all 56 references
  1. [9]

    Sofa corner: maximum capacity: 5

  2. [10]

    Large kitchen: maximum capacity: 5

  3. [11]

    Meeting room: maximum capacity: 7

  4. [12]

    Storehouse: maximum capacity: 3

  5. [13]

    Relaxing room: maximum capacity: 5

  6. [14]

    Reception room: maximum capacity: 5

  7. [15]

    Work room: maximum capacity: 5 ### Planning Steps:

  8. [17]

    Beverage room: maximum capacity: 3

  9. [18]

    Office: maximum capacity: 3

  10. [19]

    Dining room: maximum capacity: 4

  11. [20]

    Billiard room: maximum capacity: 4

  12. [22]

    - The number of scenes should be no more than 3

    Determine the number of scenes and assign a simple sub-topic to each scene. - The number of scenes should be no more than 3

  13. [23]

    Based on the sub-topic, select the location from the Main Locations for each scene

  14. [24]

    - The number of characters selected for each scene can not exceed the maximum capacity of the selected location

    Based on the sub-topic, select several characters from the Main Characters for each scene. - The number of characters selected for each scene can not exceed the maximum capacity of the selected location. - The number of characters selected for each scene is at least two. - All...

  15. [25]

    - The story plot for each scene must include only the characters selected for this scene in step 3

    Based on the sub-topic, write a story plot for each scene. - The story plot for each scene must include only the characters selected for this scene in step 3. - All the story plots should be story-coherent, and the more dramatic and specific the story, the better. - All the st...

  16. [26]

    sub-topic

    Based on the story plot, give a final dialogue goal so that the dialogue between the characters in this scene can end naturally. Your response should only contain the following JSON content: [{"sub-topic": "...", "selected-characters": ["...","...",...], "selected-location": "...

  17. [27]

    In a scene, each character’s position must be different

  18. [28]

    scene-id

    You need to provide a reason for your choice of position. Your response should only contain the following JSON content: [{"scene-id": "...", "scene-location": "...", "reason": "...", "scene-position": [{"character": "...", "position": "..."}, ...] }, ...] Figure 13: Prompt for...

  19. [30]

    Dialogues: {dialogue_draft}

  20. [31]

    Characters’ Initial Positions and States : {initial_position} (e.g., Alex: Position A, sittable, standing) ### [Complete List of Actions]:

  21. [32]

    - Standing Thinking: show the character is thinking

    Actions performed in standing state: - Standing Talking: simply show the character is talking. - Standing Thinking: show the character is thinking. - Standing Depressed: convey the character’s depressed emotion. - Standing Crying: show the character is crying. - Standing Angry...

  22. [33]

    - Sitting Laughing: convey the character’s happy emotion

    Actions performed in sitting state: - Sitting Talking: simply show the character is talking. - Sitting Laughing: convey the character’s happy emotion. - Sitting Claping: indicate the character agrees or is happy. - Stand Up: Only when you need to change the character’s state f...

  23. [34]

    - Each character can only add one action at a time

    Basic Requirements: - Note that all actions should be selected from Complete List of Actions. - Each character can only add one action at a time. - Unnecessary actions should not be added to avoid cluttering the scene

  24. [35]

    Sit Down

    Key Requirements: - In one scene, you cannot overuse a certain action. - When a character is in the standing state, you can only choose the action performed in standing state. When a character is in the sitting state, you can only choose the action performed in sitting state. ...

  25. [36]

    Movable Characters: {characters_in_standing_state}

  26. [37]

    Optional Destinations: {position_description} ### Script Information:

  27. [38]

    Plot: {scene_outline}

  28. [39]

    <Insertion Position 0>

    Dialogues: {dialogue_with_insertion} (e.g., ["<Insertion Position 0>",{"speaker":"Alex","content":"How are you?"},"<Insertion Position 1>",{"speaker":"Taylor","content":"I’m fine"}, ...])

  29. [40]

    The current positions of characters: {initial_position} ### Requirements:

  30. [41]

    You should not add any unnecessary character movement

  31. [42]

    If you need to add a character movement, you must provide a sufficient and necessary reason

  32. [43]

    reason":

    Output Requirements: - If no character movement is required, return: {"reason": "...", "move": "None"} - If you need to add a character movement, please specify the character who will move, the destination of the movement, the reason for adding this movement, the best insertio...

  33. [44]

    Sit Down

    Action Reasonableness: - Check whether the actions used in the script do not exist in Complete List of Actions. - Check whether the actions are appropriate. If you find it inappropriate, suggest a better action in Complete List of Actions. - Check whether any character has vio...

  34. [45]

    Theme Consistency: - Evaluate whether the theme is clearly and strongly presented in the script

  35. [46]

    action-reasonableness

    Script Coherence: - Evaluate whether the script is coherent and captivating, with pacing that neither drags nor rushes. If not, first identify the dialogues in the script that lead to this outcome, then provide detailed revision suggestions. - Assess whether the script flows s...

  36. [47]

    - In each scene, the characters participating in the dialogue can only include those specified in the script for that scene

    Dialogue Requirements: - Please keep the dialogue natural, concise, and vivid, avoiding repetition, clichés, and the use of numbers. - In each scene, the characters participating in the dialogue can only include those specified in the script for that scene. - The number of dia...

  37. [48]

    Sit Down

    Action Requirements: - Note that all actions in updated script should be selected from Complete List of Actions. - In a scene, you cannot overuse a certain action. - When a character is in the standing state, you can only choose the action performed in standing state. When a c...

  38. [49]

    When the character is talking, gradually focus on the character

  39. [50]

    • Pan Shot: Pivots horizontally from a fixed position to track character movement

    When the character is talking, highlight key comedic or dramatic mo- ments. • Pan Shot: Pivots horizontally from a fixed position to track character movement. – Usage Condition :

  40. [51]

    When the character is talking, use this shot multiple times in a row to create a tense atmosphere

  41. [52]

    • Tracking Shot: Continuously follows a moving character

    When the character is moving, use this shot to display both the character and surroundings. • Tracking Shot: Continuously follows a moving character. – Usage Condition : This shot can only be used when the character is moving, immersing the audience in the character’s perspect...

  42. [53]

    This shot can be used when the character first appears in the script

  43. [54]

    reasoning

    When the character is talking, create unease and heighten scene tension. • Curve Surround Shot: Moves from the character’s feet to head, with the character standing. – Usage Condition : This shot can only be used when the character first appears in the script. • Truck Shot: Sw...

  44. [55]

    Script Content: {final_script_with_own_annotation}

  45. [56]

    Another Cinematographer’s Shot Annotations: {peer_annotation}

  46. [57]

    need update

    Complete List of Shots: {shot_list} (same as prompt in Figure 22) ### [Annotation Requirements]: - If there are too many close shots or mid shots in one scene, suggest appropriate dynamic shots to replace some, but not all, based on the usage conditions. - If the first shot of...

  47. [301]

    URL https://doi.org/10.1145/3386569

    doi: 10.1145/3386569.3392427. URL https://doi.org/10.1145/3386569. 3392427. Hongda Jiang, Marc Christie, Xi Wang, Libin Liu, Bin Wang, and Baoquan Chen. Camera keyfram- ing with style and control. ACM Trans. Graph., 40(6), December 2021. ISSN 0730-0301. doi: 10.1145/3478513.34...

  48. [2025]

    Joyful Jump

    URL https://arxiv.org/abs/2501.04227. Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. V oyager: An open-ended embodied agent with large language models, 2023. Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, ...

    2023 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.