An ai video script generation and character automatic decomposition method and system

Abstract

The application discloses an AI video script generation and character automatic decomposition method and system, constructs an AI recognizable video script through a short sentence structure, takes the short sentence as a basic conversion unit of a script, and provides standardized support for subsequent processing of an AI model; meanwhile, based on a large language model, script content is analyzed, and all character information involved in a scene is extracted, including the number of leading roles and supporting roles, the appearance frequency, and the scene distribution situation. The application solves the problems that existing video script generation and character decomposition rely on manual operation, are low in efficiency, and are insufficient in accuracy, realizes AI adaptability of script generation and automation and intelligentization of character decomposition, significantly improves the early preparation efficiency of video production, and can be widely applied to many fields such as film and television creation, short video production, virtual scene interpretation and the like.

Description

Technical Field

[0001] This invention specifically relates to an AI video script generation and automatic character decomposition method and system, belonging to the fields of artificial intelligence, natural language processing and video production technology. Background Technology

[0002] In the fields of film and television creation, short video production, and virtual scene performance, script generation and character breakdown are core aspects of the early preparation.

[0003] In traditional methods, video scripts are often written in a natural paragraph format, which is difficult for AI models to recognize and process directly. They require manual reorganization and optimization to adapt to the AI ​​production process. At the same time, the extraction of character information, quantity statistics, relationship analysis and distribution sorting in the script also rely on manual line-by-line operation, which is not only inefficient, but also prone to problems such as missing characters, incorrect classification and misjudgment of relationships due to human negligence, which will affect subsequent shooting planning, character allocation and video generation progress.

[0004] With the development of artificial intelligence technology, large language models are increasingly widely used in the field of natural language processing, but there is currently a lack of integrated technical solutions for video script generation and character decomposition.

[0005] Existing technologies either only enable simple script format conversion, failing to meet the deep processing needs of AI models; or they can only identify single character entities, making it difficult to distinguish between protagonists and supporting characters, and unable to accurately extract the relationships between characters and scene distribution.

[0006] Therefore, there is an urgent need for a technology that can generate AI-recognizable scripts and achieve automatic and accurate character decomposition to improve the intelligence level and efficiency of the pre-production process of video production. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide an AI video script generation and automatic character decomposition method and system.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] An AI-powered video script generation and automatic character decomposition method includes the following steps:

[0010] Script generation: The original paragraph-style script is transformed into an AI-recognizable video script using short sentence structures. The short sentences serve as the basic transformation units of the script, enabling the video script to support the AI ​​model in subsequent scene analysis, character extraction, and video generation processing.

[0011] Scene transition markers and character association markers are added between short sentences;

[0012] Role decomposition: The generated video script is automatically parsed based on a large language model. All scene segments in the script are traversed, and the role information involved in each scene is extracted. The role information includes the number of main characters and supporting characters, the distribution of roles in each scene, the frequency of appearance, and the relationship between roles.

[0013] In the above script generation steps, the construction criteria for short sentence structures include: each short sentence carries only a single action, scene description, or character dialogue information, and the length of the short sentence is no more than 50 characters.

[0014] The parsing process of the large language model in the above role decomposition steps includes:

[0015] The video script is preprocessed by segmenting it into sentences and segments, removing redundant information and invalid characters, and generating standardized text corpus.

[0016] Semantic analysis of standardized text corpus is performed using a pre-trained role recognition model to identify role names, role identity attributes, and relationships between roles.

[0017] Based on the frequency of a character's appearance, the importance of the scene, and the relevance to the plot, we distinguish between main characters and supporting characters, count the number of core main characters and the number of supporting characters in different scenes, and mark the distribution and appearance duration of each character in different scenes.

[0018] Furthermore, the aforementioned redundant information includes spaces, punctuation errors, and repeated sentences. Invalid characters include non-script-related garbled text and special symbols.

[0019] Furthermore, the aforementioned pre-trained role recognition model is a large language model based on the Transformer architecture. It is fine-tuned and trained using a massive script corpus and has the ability to recognize role entities, classify identities, and perform distribution statistics.

[0020] The above method distinguishes between main characters and supporting characters by setting a threshold for the number of appearances and a threshold for participation in the core plot. Characters whose appearance frequency is greater than or equal to the preset threshold for the number of appearances and whose participation in the core plot scenes accounts for greater than or equal to the threshold for participation in the core plot are judged as main characters, and the rest of the characters are classified as supporting characters.

[0021] The scene switching identifier includes scene spatial information and time information, and the role association identifier includes role name and identity association information between roles.

[0022] An AI-powered video script generation and automatic character decomposition system includes:

[0023] Script generation module: used to construct AI-recognizable video scripts using short sentence structures, and output video scripts that use short sentences as basic conversion units. These video scripts are used to support subsequent AI model processing.

[0024] Character Analysis Module: Based on a large language model, it automatically parses video scripts to extract character information, including character name, identity attributes, relationships, frequency of appearance, and distribution; it also distinguishes between main characters and supporting characters and counts the number of characters.

[0025] Output module: Used to output the generated AI-recognizable video script and decomposed character information in a standardized format for use in subsequent video production processes;

[0026] The storage module is used to store the generated video scripts, decomposed character information, historical processing data and model parameters, and supports data backtracking and secondary retrieval.

[0027] The aforementioned role parsing module also includes an optimization unit, which is used to perform deduplication and error correction on the identified role information. The error correction includes correcting different roles with the same name, missing roles, and misjudged relationships, thereby improving the accuracy of role decomposition.

[0028] The script generation module allows users to manually modify and optimize the generated AI-recognizable video scripts; the output module outputs the AI-recognizable video scripts in XML and TXT formats, and outputs the character information in the form of tables and visual charts.

[0029] The advantages of this invention are:

[0030] This invention discloses an AI video script generation and automatic character decomposition method and system, which has the following advantages:

[0031] 1. Improve the AI ​​adaptability of scripts, eliminate the need for manual secondary editing, and build standardized video scripts through short sentence structures. Combined with standardized tags containing scene and character information, the scripts can be directly recognized and deeply processed by AI models. This solves the problem that traditional paragraph-style scripts cannot adapt to the AI ​​production process and greatly improves the workflow continuity of AI video production.

[0032] 2. To automate and refine the decomposition of roles and avoid human error, based on standardized scripts, a large language model finely tuned from a massive script corpus is used to automatically identify, classify, count, and analyze the relationships between roles. Combined with deduplication and error correction of optimized units, this effectively avoids problems such as role omissions, misjudgments, and mismatches in relationships caused by manual operation.

[0033] 3. Integrated script generation and character decomposition improve overall video production efficiency. This invention integrates script generation and character decomposition into a unified technical solution, significantly shortening the pre-production preparation time and reducing labor costs. At the same time, the output data such as character relationships and scene distribution provide accurate support for subsequent character allocation, plot logic verification, and AI scene construction, further improving the quality of video production.

[0034] 4. Strong adaptability and wide range of application scenarios: The method and system of this invention support the processing of various types of scripts such as film and television scripts, short video scripts, and virtual scene performance scripts. Moreover, the threshold for determining the main character and supporting character can be customized and adjusted. It can be widely used in multiple fields such as film and television creation, short video production, virtual live streaming, and metaverse scene performance. It has strong practicality and high promotion value.

[0035] 5. The system features a modular design with excellent practicality and scalability. It adopts a modular design for script generation, role parsing, output, and storage. Each module is independent yet works in concert, allowing users to manually optimize the generated scripts. The storage module enables data backtracking and secondary retrieval, facilitating script modification and iterative optimization. The modular structure also provides convenience for subsequent system function upgrades and expansions. Attached Figure Description

[0036] Figure 1 This is a flowchart of the AI ​​video script generation and automatic character decomposition method of the present invention;

[0037] Figure 2 This is a block diagram of the AI ​​video script generation and automatic character decomposition system of the present invention;

[0038] Figure 3 This is a visual statistical chart for Example 2. Detailed Implementation

[0039] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0040] An AI-powered video script generation and automatic character decomposition method includes the following steps:

[0041] Step (1), script generation:

[0042] A short sentence structure is used to construct an AI-recognizable video script, with each short sentence serving as the basic conversion unit of the script. The short sentence structure must follow specific specifications: each short sentence carries only a single core piece of information and can be divided into three categories: action sentences, scene description sentences, and character dialogue sentences to avoid information redundancy; the length of the short sentences is preferably controlled within the range of 20-50 characters to ensure that the AI ​​model can accurately recognize and parse them; at the same time, scene switching identifiers (such as "[Scene Switch: Indoor - Living Room]") and character association identifiers (such as "[Character: Zhang San - Li Si (Colleague)]") are added between short sentences to give the script structured features, which facilitates subsequent character extraction, relationship identification, and scene association.

[0043] The generated video scripts can be directly used as the basic data for AI models to build scenes, generate actions, and synthesize dialogues.

[0044] Step (2), Character breakdown:

[0045] The generated video script (or existing script) content is automatically parsed based on a pre-trained large language model. The specific process includes:

[0046] Preprocessing stage: The input script or playbook content is segmented into sentences and paragraphs, and redundant information and invalid characters such as spaces, punctuation errors, and repeated sentences are removed to generate standardized text corpus with uniform format and clear semantics, providing a high-quality data foundation for subsequent parsing.

[0047] Role recognition stage: The pre-trained role recognition model (based on the Transformer architecture and fine-tuned by a large amount of film and television scripts and short video scripts) is called to perform semantic analysis on the standardized text corpus to identify the names, identity attributes (such as protagonist, supporting character, passerby, etc.) of all roles and the relationships between roles (such as superiors and subordinates, relatives, colleagues, etc., based on plot description and semantic association).

[0048] Character Classification and Statistics Phase: Based on the frequency of appearance, participation in the core plot, and scene coverage, the system automatically distinguishes between protagonists and supporting characters. By setting a threshold for the number of appearances (which can be customized according to the script type), characters with an appearance frequency ≥ the preset threshold and those participating in ≥ 50% of the core plot scenes are identified as core protagonists, while the rest are classified as scene supporting characters. At the same time, the system counts the number of core protagonists and scene supporting characters, marks the appearance position, appearance duration, and interacting characters of each character in different scenes, and generates a complete report on character distribution and relationships.

[0049] An AI-powered video script generation and automatic character decomposition system includes the following modules:

[0050] (1) Script generation module: Adopting the above-mentioned short sentence structure standard, it receives the original plot idea or paragraph script input by the user and converts it into a video script (short sentence script) that can be recognized by AI. Preferably, it also supports the user to manually modify and optimize the generated script to ensure the rationality and AI compatibility of the script. The generated video script is synchronously transmitted to the character parsing module and the storage module.

[0051] (2) Role parsing module: integrates pre-trained large language model and role recognition model, receives video scripts output by script generation module (or existing scripts directly imported by users), and performs preprocessing, role recognition, classification statistics and other operations;

[0052] The built-in optimization unit can deduplicate and correct the identified role information (such as correcting different roles with the same name, missing roles, and misjudged relationships), thereby improving the accuracy of role decomposition; the processed role information is transmitted to the output module and the storage module.

[0053] (3) Output module: Receives the video script from the script generation module and the character information from the character parsing module, outputs the video script in standardized formats such as XML and TXT, and outputs the character quantity statistics, character distribution and relationship reports in the form of tables, visualization charts, etc., for subsequent video shooting planning, AI character modeling, scene setting and other processes.

[0054] (4) Storage module: Receives and stores the generated video script, decomposed role information, historical processing data and model parameters, supports data query, backtracking and secondary call, which facilitates subsequent modification and iterative optimization by users; it can provide historical data support to the script generation module and role parsing module to assist in optimizing the processing results.

[0055] Example 1

[0056] The system of this invention is used to process a segment of a film or television script, with a preset threshold for determining the core protagonist as "appearance frequency ≥ 3 times". The specific process is as follows:

[0057] Step (1), script generation:

[0058] Enter the original plot idea or paragraph script: "On a weekend morning, Zhang San and Li Si (colleagues) went for a walk in the city center park and met their neighbor Wang Wu, who was walking his dog. After chatting for a few minutes, Zhang San received a call from his company, said goodbye to Li Si and Wang Wu, and hurriedly left the park."

[0059] The script generation module converts it into an AI-recognizable video script (short sentence script):

[0060] [Scene: City Center Park - Daytime] [Characters: Zhang San, Li Si (colleagues)] Zhang San and Li Si arrive at the City Center Park;

[0061] [Scene: City Center Park - Daytime] [Characters: Zhang San, Li Si (colleagues)] The two are taking a walk in the park;

[0062] [Scene: City Center Park - Daytime] [Characters: Zhang San, Li Si (colleagues), Wang Wu (neighbor)] Encountering neighbor Wang Wu, who is walking his dog;

[0063] [Scene: City center park - daytime] [Characters: Zhang San, Li Si (colleagues), Wang Wu (neighbors)] The three are having a brief chat;

[0064] [Scene: City Center Park - Daytime] [Character: Zhang San] Zhang San receives a call from his company;

[0065] [Scene: City Center Park - Daytime] [Characters: Zhang San, Li Si (colleagues), Wang Wu (neighbors)] Zhang San bids farewell to Li Si and Wang Wu;

[0066] [Scene: City center park - daytime] [Character: Zhang San] Zhang San hurriedly leaves the park.

[0067] Step (2), Character breakdown:

[0068] The role parsing module preprocesses the video scripts, removing redundant information except for format identifiers, and generates standardized corpus.

[0069] Subsequently, the role recognition model identified three role entities: Zhang San, Li Si, and Wang Wu, clarifying that Zhang San and Li Si are colleagues and Zhang San and Wang Wu are neighbors.

[0070] Based on appearance frequency (Zhang San 7 times, Li Si 5 times, Wang Wu 3 times) and plot participation, Zhang San and Li Si are determined to be the core protagonists (appearance frequency ≥ preset threshold 3 times, participating in the core plot throughout), while Wang Wu is a supporting character (appearance frequency equal to preset threshold, participating in only some scenes).

[0071] At the same time, the distribution of characters was statistically analyzed, and it was noted that the three characters only appeared in the downtown park scene, forming a report on character distribution and relationships.

[0072] Example 2

[0073] In a larger-scale test case, the system processed an entire film and television script (containing 200 scenes and 100,000 words of text). The preset threshold for identifying the core protagonist was "appearance frequency ≥ 15 times". The system successfully identified 3 core protagonists and 47 supporting characters from the script, covering a total of 50 character entities. The accuracy rate of character recognition was over 98%, the accuracy rate of protagonist and supporting character classification was over 95%, and the accuracy rate of character relationship identification was over 96%. There were no cases of missing characters, misjudgments, or mismatches in relationships, which fully verified the effectiveness and stability of the technical solution of this invention.

[0074] The system of this invention was used to process an entire film and television script, which contained 200 scenes and 100,000 words of text. A preset threshold for determining the core protagonist was set: appearance frequency ≥ 15 times, and core plot participation ≥ 60%. The specific processing results are as follows:

[0075] The system uses a script generation module to convert 100,000 fields of text into standardized short sentence scripts, adding scene and character association tags. The script generation takes about 10 minutes, which is more than 90% more efficient than manual compilation.

[0076] The character parsing module parses the standardized short sentence scripts. After preprocessing, semantic recognition, classification statistics, and deduplication and error correction by the optimization unit, it successfully identified 50 character entities, including 3 core protagonists and 47 supporting characters.

[0077] After manual verification, the accuracy rate of character recognition in this processing reached 98.5%, the accuracy rate of classifying main characters and supporting characters reached 96%, and the accuracy rate of character relationship recognition reached 97%. There were no cases of missing characters, misjudgments, or mismatched relationships.

[0078] The output module outputs short scripts of film and television scripts in XML format, and outputs the character decomposition results in the form of visual statistical charts (including character appearance frequency, scene distribution, and relationship graphs), providing accurate data support for subsequent character allocation and scene arrangement in film and television shooting; the storage module stores all the data processed in this session, providing a data foundation for subsequent script modifications and iterations.

[0079] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A method for AI video script generation and automatic character decomposition, characterized in that, Includes the following steps: Script generation: The original paragraph-style script is transformed into an AI-recognizable video script using short sentence structures. The short sentences serve as the basic transformation units of the script, enabling the video script to support the AI ​​model in subsequent scene analysis, character extraction, and video generation processing. Scene transition markers and character association markers are added between short sentences; Role decomposition: The generated video script is automatically parsed based on a large language model. All scene segments in the script are traversed, and the role information involved in each scene is extracted. The role information includes the number of main characters and supporting characters, the distribution of roles in each scene, the frequency of appearance, and the relationship between roles.

2. The method according to claim 1, characterized in that, In the script generation step, the construction criteria for short sentence structures include: each short sentence carries only a single action, scene description, or character dialogue information, and the length of the short sentence is no more than 50 characters.

3. The method according to claim 1, characterized in that, The role decomposition step, specifically the parsing process of the large language model, includes: The video script is preprocessed by segmenting it into sentences and segments, removing redundant information and invalid characters, and generating standardized text corpus. Semantic analysis of standardized text corpus is performed using a pre-trained role recognition model to identify role names, role identity attributes, and relationships between roles. Based on the frequency of a character's appearance, the importance of the scene, and the relevance to the plot, we distinguish between main characters and supporting characters, count the number of core main characters and the number of supporting characters in different scenes, and mark the distribution and appearance duration of each character in different scenes.

4. The method according to claim 3, characterized in that, The redundant information includes spaces, punctuation errors, and repeated sentences. Invalid characters include non-script-related garbled text and special symbols.

5. The method according to claim 3, characterized in that, The pre-trained role recognition model is a large language model based on the Transformer architecture. It is fine-tuned and trained using a massive script corpus and has the ability to recognize role entities, classify identities, and perform distribution statistics.

6. The method according to claim 3, characterized in that, The main characters and supporting characters are distinguished by preset thresholds for the number of appearances and the participation in the core plot. Characters whose appearance frequency is greater than or equal to the preset threshold and whose participation in the core plot scenes accounts for more than or equal to the core plot participation threshold are judged as the main characters, and the rest are classified as supporting characters.

7. The method according to claim 3, characterized in that, The scene switching identifier includes scene spatial information and time information, and the role association identifier includes role name and identity association information between roles.

8. An AI video script generation and automatic character decomposition system, characterized in that, include: Script generation module: used to construct AI-recognizable video scripts using short sentence structures, and output video scripts that use short sentences as basic conversion units. These video scripts are used to support subsequent AI model processing. Character Analysis Module: Based on a large language model, it automatically parses video scripts to extract character information, including character name, identity attributes, relationships, frequency of appearance, and distribution; it also distinguishes between main characters and supporting characters and counts the number of characters. Output module: Used to output the generated AI-recognizable video script and decomposed character information in a standardized format for use in subsequent video production processes; The storage module is used to store the generated video scripts, decomposed character information, historical processing data, and model parameters.

9. The system according to claim 8, characterized in that, The role parsing module also includes an optimization unit, which is used to perform deduplication and error correction on the identified role information; The error correction includes correcting different roles with the same name, missing roles, and misjudging relationships.

10. The system according to claim 8, characterized in that, The script generation module allows users to manually modify and optimize the generated AI-recognizable video scripts; the output module outputs the AI-recognizable video scripts in XML and TXT formats, and outputs the character information in the form of tables and visual charts.

Images