A multi-episode comic script collaborative creation method and system

Abstract

The present application provides a multi-episode comic script collaborative creation method and system, the method comprising: splitting the original script text into multiple target episode units and generating script data; extracting visual assets of each target episode unit and generating corresponding first prompt words; inputting the first prompt words into a picture generation model to generate corresponding visual asset images and storing them in a shared asset library; identifying the visual asset images in the shared asset library through a third text large model to generate split screen information and corresponding second prompt words for each target episode unit; generating dubbing audio files and corresponding timestamp information based on the split screen information through a speech synthesis model; generating corresponding split screen video clips based on the second prompt words corresponding to each split screen information through a video generation model; splicing all split screen video clips in order, superimposing corresponding dubbing audio files according to the timestamp information, and outputting complete comic video of the target episode unit.

Description

Technical Field

[0001] This invention relates to the field of video content generation technology, and in particular to a method and system for collaborative creation of multiple TV series / comic dramas. Background Technology

[0002] Comic series, as a new form of video content combining dynamic comics with short drama narratives, encompasses various content formats, including but not limited to: humanoid comic series, 3D animated comic series, 2D animated comic series, animated narration comic series, static narration comic series, and animated comic series with emojis. Its core lies in presenting static comics or novel texts through dynamic visuals, voice-over, and background music in a video format. With the development of artificial intelligence technology, especially the emergence of multimodal large-scale models, comic series production has seen a revolutionary opportunity for efficiency.

[0003] Currently, there are tools on the market that use AI to assist in the production of comics. However, existing comic creation tools require creators to frequently switch between different systems. From script breakdown and character design to material generation and manual editing, the fragmented process leads to low creation efficiency and makes it difficult to ensure a high degree of consistency of visual assets between multiple episodes. Furthermore, creators cannot build and manage an asset library that runs through the entire series on a unified platform, which means that each episode and each storyboard needs to be generated repeatedly, resulting in a waste of computing power and manpower and a reduced development cycle. Summary of the Invention

[0004] In view of this, the present invention proposes a collaborative creation method and system for multiple series of comics, which realizes end-to-end intelligent generation from script to finished product, while ensuring high visual consistency among multiple series and supporting multi-person parallel collaborative creation, significantly improving the production efficiency and quality of long-form comics.

[0005] The technical solution of this invention is implemented as follows: Firstly, a method for collaborative creation of multi-episode comics includes the following steps: S1 sets the global configuration information for creating the comic book; S2, obtain the original script text, identify the original script text through the first text model, and combine it with global configuration information to split the original script text into multiple target drama units, and generate script data for the target drama units; S3, based on the script data of the target drama series unit, extracts the visual assets of each target drama series unit through the second text model and generates the corresponding first prompt word; S4. Input the first prompt word into the image generation model to generate the corresponding visual asset image, perform image standardization processing, and store it in the shared asset library. S5, based on the visual asset images in the shared asset library, uses a third text big model for identification to generate the storyboard information and corresponding second prompt words for each target drama unit; S6 generates dubbing audio files and corresponding timestamp information based on the storyboard information through a speech synthesis model; S7: Based on the second cue words corresponding to each storyboard information, the corresponding storyboard video clips are generated through the video generation model. S8 splices all storyboard video clips in sequence, overlays the corresponding dubbing audio files according to the timestamp information, and outputs the complete comic book video of the target episode unit.

[0006] Based on the above technical solutions, preferably, the global configuration information in step S1 includes the project name, number of episodes, style type, content description and video resolution, and selects the model type and version used for creation from the model library. The model types include text model, image generation model, video generation model, speech synthesis model and music generation model.

[0007] Based on the above technical solutions, preferably, step S2, which involves obtaining the original script text, recognizing the original script text using a first text model, splitting the original script text into multiple target drama units based on global configuration information, and generating script data for the target drama units, includes the following sub-steps: S21, Obtain the original script text and perform text data preprocessing to obtain the standard script text; S22, obtain the number of episodes in the global configuration information, and construct episode prompts to guide the first text model. The episode prompts include task instructions, plot analysis instructions, integrity constraint instructions and output format instructions. S23, after concatenating the episode prompts with the standard script text, input them into the first text model to generate a candidate episode boundary set, and adjust the boundaries of the candidate episode boundary set according to the number of episodes to obtain the final episode boundary set; S24. The standard script text is divided into multiple series units according to the final episode boundary set. The text content of the multiple series units is re-input into the first text model for text optimization processing to obtain the script data of each target series unit. The script data includes the unique identifier of the target series unit, the series number, the plot points, the emotional tone, the list of characters, the list of scenes, the lines, the style tags, the text length, and the estimated duration.

[0008] Based on the above technical solutions, preferably, step S23, which involves adjusting the boundaries of the candidate episode boundary set according to the number of episodes to obtain the final episode boundary set, includes the following sub-steps: S231, Obtain the total number of words in the standard script text, and calculate the theoretical number of words per episode based on the number of episodes; S232, take the position corresponding to the theoretical number of words in each episode as the expected center point, and set an elastic offset range, and select all candidate episode boundaries that are within the elastic offset range from the candidate episode boundary set; S233, if candidate episode boundaries exist after screening, calculate the plot integrity score corresponding to each candidate episode boundary and the word count deviation value between the position of each candidate episode boundary and the expected center point; S234. Based on the plot integrity score and word count deviation value, calculate the comprehensive score of each candidate episode boundary after screening, and select the candidate boundary with the highest comprehensive score as the end boundary of the current target episode unit. S235, if no candidate boundary is found after filtering, then the candidate boundaries adjacent to the elastic offset range before and after are obtained respectively as the front candidate boundary and the back candidate boundary. Obtain the word count deviation value and plot integrity score corresponding to the previous candidate boundary and the next candidate boundary. If the word count deviation of the previous candidate boundary or the next candidate boundary is less than the preset deviation threshold and the plot integrity score is greater than the preset integrity threshold, then select the candidate boundary with the smaller word count deviation value between the previous candidate boundary and the next candidate boundary as the end boundary of the current target episode unit. If the word count deviation values ​​of the previous candidate boundary and the subsequent candidate boundary are both greater than the preset deviation threshold, or the plot integrity scores are both less than the preset integrity threshold, then the semantic boundaries are searched before and after the expected center point, and the semantic boundaries are used as the end boundaries of the current target episode unit. S236, using the end boundary of the current target episode unit as the starting position of the next target episode unit, repeat steps S232 to S235 to obtain the final episode boundary set.

[0009] Based on the above technical solutions, preferably, step S3, which involves extracting the visual assets of each target drama unit from the script data of the target drama unit using a second text model and generating corresponding first prompt words, includes the following sub-steps: S31, acquire script data of all target drama units to generate script corpus, and construct visual asset extraction prompts to guide the second text model. Visual asset extraction prompts include task instructions, asset classification instructions, attribute extraction instructions, frequency statistics instructions, emotion extraction instructions and output format instructions. S32, concatenate the asset extraction prompts with the script corpus and input them into the second text model to identify visual entities and output an initial asset list with attribute descriptions and frequency statistics for each visual entity; S33, calculate the similarity value between any two visual entity attribute descriptions based on semantic similarity. When the similarity value is greater than the preset similarity threshold, determine to merge the two assets corresponding to the two visual entity attribute descriptions, and update the list of episodes and frequency of occurrence of the merged assets. S34. Based on the global requirement of spanning at least two target series units, the merged asset list is filtered to retain assets with an occurrence frequency of not less than 2, and the final visual asset list is generated. S35, for each asset in the final visual asset list, extract visual features from the visual entity attribute description, obtain style type from global configuration information, extract emotion type from the emotion of the target drama unit with the highest asset frequency, combine visual features, style type and emotion type to form the first prompt word, and generate multiple versions of the first prompt word according to preset multi-angle requirements, including different perspectives, different expressions or different lighting conditions. S36. Store the list of visual assets and their corresponding first prompt words into the shared asset library. Each visual asset includes an asset identifier, asset type, asset name, attribute description, list of first prompt word versions, asset status, and asset version number.

[0010] Based on the above technical solutions, preferably, step S5, which involves identifying visual asset images from the shared asset library using a third text big data model to generate storyboard information and corresponding second prompt words for each target drama unit, includes the following sub-steps: S51, acquire the script data of the target drama unit, and construct storyboard generation prompts to guide the third text model. The storyboard generation prompts include task instructions, asset reference instructions, storyboard element instructions, second prompt generation instructions, and output format instructions. S52, after concatenating the storyboard generation prompts with the script data of the target series, input the third text model. The third text model recognizes the script content, generates a storyboard sequence, and retrieves matching visual assets from the shared asset library based on the character, scene and prop names in the plot description, and obtains the asset identifier of the corresponding visual asset. S53, for each scene, embed the retrieved asset identifier into the scene information, bind the scene to the shared asset library reference, and output the scene information, which includes scene number, shot type, camera movement method, on-screen character asset identifier, prop asset identifier, scene asset identifier, plot description text and lines. S54, based on the storyboard information and the referenced visual assets, the third text model generates a corresponding second cue word for each storyboard. The second cue word includes style prefix, image description, asset reference mark, mood label, and lighting and tone.

[0011] Based on the above technical solutions, preferably, step S6, which involves generating a dubbing audio file and corresponding timestamp information using a speech synthesis model based on the storyboard information, includes the following sub-steps: S61, extract all dialogue text from the storyboard information of the target drama unit, and sort them according to the storyboard order to form a dialogue sequence. Each line of dialogue has a corresponding storyboard number and character identifier. S62, based on the character identifier of each line, matches the corresponding timbre parameters from the preset timbre library; S63: Input the dialogue text and timbre parameters into the speech synthesis model, generate corresponding dubbing audio segments one by one, and record the duration of each audio segment; S64, based on the duration and order of each audio segment, calculate the start and end times of each line in the entire audio stream, and generate a subtitle file with timestamp information; S65 stores the dubbed audio clips and their corresponding timestamps in the material library of the target drama unit and associates them with the corresponding storyboard number.

[0012] Based on the above technical solutions, preferably, step S7, which involves generating corresponding storyboard video segments using a video generation model based on the second prompt word corresponding to each storyboard information, includes the following sub-steps: S71, obtain the second prompt word corresponding to each scene and the corresponding referenced visual asset identifier, and obtain the corresponding visual asset image access path from the shared asset library based on the visual asset identifier; S72, taking the second cue word and the referenced visual asset image as input, generates corresponding video segments for each scene through a video generation model; S73 stores video clips according to their storyboard numbers into the material library of the corresponding target drama unit and associates them with the corresponding storyboard information.

[0013] Based on the above technical solutions, preferably, step S8, which involves sequentially splicing all storyboard video clips, overlaying corresponding dubbing audio files according to timestamp information, and outputting the complete animated video of the target series unit, includes the following sub-steps: S81: Obtain all storyboard video clips, sort them according to the storyboard number, and then splice the sorted storyboard video clips together to form a video stream. S82, obtain the dubbing audio file and timestamp information, and overlay the dubbing audio onto the corresponding time position of the video stream according to the timestamp; S83 generates subtitles based on the dialogue text and timestamp information, and then integrates the subtitles into the video; S84 retrieves the video resolution parameters from the global configuration information, encodes and encapsulates the output video file, and outputs a complete animated video file.

[0014] Secondly, the present invention also provides a multi-episode comic book collaborative creation system, implemented using a multi-episode comic book collaborative creation method, including: The configuration settings module is used to set the global configuration information for the target comic book creation; The series segmentation module is used to obtain the original script text, identify the original script text through the first text model, and segment the original script text into multiple target series units based on global configuration information, and generate script data for the target series units. The visual asset generation module is used to extract the visual assets of each target drama unit based on the script data of the target drama unit through the second text model, and generate the corresponding first prompt words. The visual asset image generation module is used to input the first prompt word into the image generation model, generate the corresponding visual asset image, perform image standardization processing, and store it in the shared asset library. The storyboard information generation module is used to generate storyboard information and corresponding second prompt words for each target drama unit by recognizing visual asset images from the shared asset library through a third text big data model. The audio generation module is used to generate dubbing audio files and corresponding timestamp information based on the storyboard information and a speech synthesis model. The video generation module is used to generate corresponding storyboard video clips based on the second cue words corresponding to each storyboard information through a video generation model. The compositing output module is used to splice all the storyboard video clips in sequence, overlay the corresponding dubbing audio files according to the timestamp information, and output the complete comic book video of the target episode unit.

[0015] The multi-episode comic book collaborative creation method and system of the present invention has the following advantages over the prior art: (1) By configuring unified project constraints globally, the script is intelligently parsed and divided into episodes using a text model, and global visual assets are extracted to generate standardized images and store them in a shared asset library. For any episode, relevant visual assets are retrieved from the asset library to generate structured storyboard information and prompts. Then, voice and video models are used to generate dubbing and storyboard video clips, which are then automatically synthesized into a complete episode. When multiple episodes are created in parallel, they all access the same shared asset library and reference the same asset images to ensure the visual consistency of cross-episode content and improve the creation efficiency and quality of comics. (2) With the total number of episodes as a constraint, a comprehensive scoring mechanism of flexible offset range, plot integrity score and word count deviation value is used to ensure the balance of word count while avoiding the fragmentation of key plots, thus improving the efficiency and accuracy of episode division. In addition, a multi-level fault tolerance strategy of candidate boundary screening, extended search and semantic boundary completion is combined to ensure that reasonable episode division can be obtained, thereby improving the cross-episode plot continuity of comic series production. (3) By using a synonym merging mechanism to resolve ambiguity in entity representation, and combining frequency filtering to ensure the global reuse value of assets, we introduce emotional tags and multi-angle prompts to enhance the emotional dimension and perspective diversity of image generation; the final shared asset library provides unified visual materials for the parallel creation of multiple series, ensuring the visual consistency of cross-episode content. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 The flowchart is a multi-episode comic book collaborative creation method of the present invention; Figure 2 This is a schematic diagram of the user workbench interface of the multi-episode comic collaborative creation method of the present invention. Figure 3 This is a case study diagram showing the user workbench interface containing demonstration data for the multi-episode comic collaborative creation method of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1-3 As shown, in a first aspect, the present invention provides a method for collaborative creation of multi-episode comics, comprising the following steps: S1 sets the global configuration information for creating the comic book; It should be noted that users initiate a request to create a new comic book project through the system's front-end interface, thus entering the project creation process. This step aims to establish a unified top-level configuration for the entire multi-episode comic book creation project, serving as a constraint benchmark and guiding basis for all subsequent steps.

[0020] The global configuration information mentioned in step S1 includes project name, number of episodes, style type, content description and video resolution, and selects the model type and version used for creation from the model library. The model types include text model, image generation model, video generation model, speech synthesis model and music generation model.

[0021] Specifically, when a user clicks the "Create New Project" button on the front-end interface, a project configuration form pops up. The user fills in the project name, number of episodes, style type, content description, and video resolution. The number of episodes specifies the total number of episodes planned for the project. The style type is the overall visual style of the project selected from the system's preset style tag library. Style tags include, but are not limited to: ancient Chinese fantasy, cyberpunk, urban romance, realism, anime, and ink painting style. This style type will serve as a global visual constraint, running through all subsequent episode asset generation, storyboard generation, and video generation stages. The content description is a brief description of the project content entered by the user, used to assist the AI ​​model in understanding the project background and improve the relevance of the subsequently generated content. The video resolution is the selected resolution of the output video, used to unify the output format of all episode compilations.

[0022] The system provides a pre-set model library containing various types of AI model instances and their version information. Users can specify the models to be used in each subsequent creative stage according to project needs and model performance. These models include text models, image generation models, video generation models, speech synthesis models, and music generation models. Text models are used for text processing tasks such as script analysis, asset extraction, and storyboard generation. Users can select large text models from the library, such as Tongyi Qianwen-Max, Doubao-Pro, or GPT-4, and specify the specific version number. Image generation models are used for generating global visual asset images. Users can select models such as Tongyi Wanxiang and Seed. Large image generation models such as REAM or Midjourney can be selected, with the version specified. Video generation models are used to generate storyboard video clips. Users can choose large video generation models such as Keling, RunwayGen-2, and Pika, and specify the version. Speech synthesis models are used to generate dubbing audio. Users can choose speech synthesis models such as Microsoft AzureTTS and Volcano Engine, and specify the voice library version. Music generation models are used for automatic background music generation. Users can choose music generation models such as SunoAI or Mubert. If no background music is specified, the system will use a preset background music library for matching.

[0023] After the user completes the configuration and clicks the "Create Project" button, the system stores the global configuration information entered in a structured data format in the database of the project management module and generates a unique project identifier. This configuration information will be associated with the project and can be called by all subsequent drama series creation units. The system returns a project creation success message and automatically navigates the user to the project's main interface. The left side of the main interface displays the drama series list, the right side displays a summary of the project configuration, and provides an entry point for importing scripts, waiting for the user to perform step S2, script import and parsing.

[0024] In this embodiment, by establishing unified global configuration information, style type and video resolution are used as global constraints to ensure that all subsequent episodes maintain consistency in visual style and output format, avoiding style fragmentation caused by different episodes using different configurations; at the same time, it supports specifying different AI models and versions for different tasks, enabling the system to adapt to rapidly developing AI technologies. Users can flexibly adjust model combinations according to actual results, improving applicability.

[0025] S2, obtain the original script text, identify the original script text through the first text model, and combine it with global configuration information to split the original script text into multiple target drama units, and generate script data for the target drama units; Step S2 includes the following sub-steps: S21, Obtain the original script text and perform text data preprocessing to obtain the standard script text; It should be noted that users upload the original script text file through the front-end interface. Supported file formats include, but are not limited to: plain text, Word document, PDF, and structured script formats. The system parses the uploaded file format. For plain text files, the system directly reads the text content. For Word document files, the system uses a document parsing component to extract the text content and remove formatting symbols and style information. For PDF files, the system uses a PDF text extraction tool to extract the text content and identify paragraph structure. For structured script files, the system parses their XML or JSON structure and extracts dialogue, scene descriptions, and character information. After extraction, the system preprocesses the text content, including removing redundant spaces and line breaks, standardizing Chinese and English punctuation, identifying and retaining chapter titles as potential episode references, and filtering irrelevant comments or metadata. After preprocessing, the system obtains the standard script text.

[0026] S22, obtain the number of episodes in the global configuration information, and construct episode prompts to guide the first text model. The episode prompts include task instructions, plot analysis instructions, integrity constraint instructions and output format instructions. Specifically, the task instruction requires the model to act as a professional screenwriter, to perform in-depth content understanding and episode segmentation of the input script; for example: You are an experienced screenwriter, please break down the following novel text into several episodes, each of which should have a relatively complete plot arc.

[0027] The plot analysis instruction requires the model to identify plot turning points, chapter boundaries, character appearance and exit patterns, and plot climax locations in the script. For example: Please analyze the plot turning points, natural chapter breaks, character appearance and exit times, and plot climax locations in the text, and use these locations as candidate episode boundaries.

[0028] The integrity constraint requires the model to ensure that each episode unit has a relatively complete plot structure, including introduction, development, climax and resolution, and to avoid interruptions at key conflict points; for example: ensure that each episode has a relatively complete plot structure, including cause, development, climax and suspense, and should not be interrupted in the middle of key conflicts or dialogues.

[0029] The output format instruction requires the model to output the candidate set of boundaries and the corresponding plot points in a structured data format.

[0030] It also includes a plot integrity scoring instruction, which requires the model to generate a plot integrity score for each candidate set boundary. The higher the score, the more suitable the boundary is as a set point, that is, the closer it is to a natural plot turning point or chapter break, and the less likely it is to cut off important plot points.

[0031] S23, after concatenating the episode prompts with the standard script text, input them into the first text model to generate a candidate episode boundary set, and adjust the boundaries of the candidate episode boundary set according to the number of episodes to obtain the final episode boundary set; In this embodiment, the episode-specific prompt words are concatenated with the generated standard script text to form a complete input prompt. The designated first text model, Tongyi Qianwen-Max, is then invoked for reasoning. The model performs deep semantic understanding of the script content, analyzes the plot structure and narrative rhythm, and outputs a candidate episode boundary set C={c1,c2,...,c...} m}, each boundary c j Including boundary locations Plot Highlights and plot integrity rating After obtaining the candidate episode boundary set, the system adjusts and optimizes the candidate boundaries according to the number of episodes N in the global configuration information to obtain the final episode boundary set B.

[0032] Step S23, which involves adjusting the boundaries of the candidate episode boundary set based on the number of episodes to obtain the final episode boundary set, includes the following sub-steps: S231, Obtain the total number of words in the standard script text, and calculate the theoretical number of words per episode based on the number of episodes; S232, take the position corresponding to the theoretical number of words in each episode as the expected center point, and set an elastic offset range, and select all candidate episode boundaries that are within the elastic offset range from the candidate episode boundary set; Set the current episode start position pointer pos start =0, the current episode number i=1, initialize the final episode boundary set B=[]; For the current episode i, with pos start Starting from the position, let W be the theoretical word count per episode. target The corresponding position is taken as the expected center point pos center ;pos center =pos start +W target Set the elastic offset range [L,R]; Where L=pos start +W target ×(1-α), R=pos start +W target ×(1+α), where α is a preset elasticity coefficient, usually taken as 0.1-0.2, used to provide adjustment space for plot integrity while ensuring that the number of words is close; from the candidate set of episode boundaries C, all candidate boundaries located within the current elastic offset range are selected to form a candidate subset.

[0033] S233, if candidate episode boundaries exist after filtering, calculate the plot integrity score corresponding to each candidate episode boundary and the word count deviation value between the position of each candidate episode boundary and the expected center point. The expression for calculating the word count deviation value is:

[0034] In the formula, c j For the first j For each candidate diversity boundary, the smaller the word count deviation value, the closer the boundary is to the theoretical word count requirement; S234. Based on the plot integrity score and word count deviation value, calculate the comprehensive score of each candidate episode boundary after screening, and select the candidate boundary with the highest comprehensive score as the end boundary of the current target episode unit. It should be noted that for each candidate subset boundary, a comprehensive priority score is constructed, expressed as follows: P j =w1×(1- d j / W target )+w2×s j In the formula, w1 is the weighting coefficient for the word count deviation value, and w2 is the weighting coefficient for the plot completeness score; the weighting coefficients can be adaptively adjusted to balance word count proximity and plot completeness. P is selected as the weighting coefficient. j The candidate episode boundary c corresponding to the maximum value is taken as the actual end boundary of the current episode, and c is added to set B.

[0035] S235, if no candidate boundary is found after filtering, then the candidate boundaries adjacent to the elastic offset range before and after are obtained respectively as the front candidate boundary and the back candidate boundary. Obtain the word count deviation value and plot integrity score corresponding to the previous candidate boundary and the next candidate boundary. If the word count deviation of the previous candidate boundary or the next candidate boundary is less than the preset deviation threshold and the plot integrity score is greater than the preset integrity threshold, then select the candidate boundary with the smaller word count deviation value between the previous candidate boundary and the next candidate boundary as the end boundary of the current target episode unit. If the word count deviation values ​​of the previous candidate boundary and the subsequent candidate boundary are both greater than the preset deviation threshold, or the plot integrity scores are both less than the preset integrity threshold, then the semantic boundaries are searched before and after the expected center point, and the semantic boundaries are used as the end boundaries of the current target episode unit. S236, using the end boundary of the current target episode unit as the starting position of the next target episode unit, repeat steps S232 to S235 to obtain the final episode boundary set.

[0036] S24. The standard script text is divided into multiple series units according to the final episode boundary set. The text content of the multiple series units is re-input into the first text model for text optimization processing to obtain the script data of each target series unit. The script data includes the unique identifier of the target series unit, the series number, the plot points, the emotional tone, the list of characters, the list of scenes, the lines, the style tags, the text length, and the estimated duration.

[0037] It should be noted that the standard script text is segmented based on the final episode boundary set to obtain the original text content of N episode units. Each episode unit is assigned a unique identifier and episode number. The text content of each episode unit is then re-input into the first text model for optimization. The text optimization process includes dialogue polishing, scene description supplementation, plot conflict enhancement, emotional tone annotation, and style tag inheritance. Specifically, dialogue polishing involves adjusting written dialogue to colloquial expressions that conform to the character's setting, making the lines more natural and consistent with the character's personality. Scene description supplementation adds necessary visual description information to key plot points, enhancing visualization and facilitating subsequent storyboard generation. Plot conflict enhancement involves fine-tuning dialogue or descriptions to highlight the core conflict and enhance dramatic tension. Emotional tone annotation marks the main emotional trends for each episode unit, used for subsequent dubbing and background music selection. Style tag inheritance inherits the style type from the global configuration information or generates local style tags based on the content of the episode.

[0038] This embodiment uses the number of episodes in the global configuration information as a constraint, and combines a comprehensive scoring mechanism of elastic offset range, plot integrity score, and word count deviation value to ensure a balanced word count for each episode while minimizing the fragmentation of key plot points, thus solving the problems of low efficiency and strong subjectivity in manual episode segmentation. Through a multi-level fault-tolerant strategy of candidate boundary screening, extended search, and semantic boundary completion, it ensures that scripts of different lengths and structures can obtain reasonable episode segmentation results. Finally, structured script data is generated, providing a standardized data foundation for subsequent global asset extraction and parallel creation of multiple episodes, improving the automation level of long-form comic series production and the coherence of cross-episode plots.

[0039] S3, based on the script data of the target drama series unit, extracts the visual assets of each target drama series unit through the second text model and generates the corresponding first prompt word; Step S3 includes the following sub-steps: S31, acquire script data of all target drama units to generate script corpus, and construct visual asset extraction prompts to guide the second text model. Visual asset extraction prompts include task instructions, asset classification instructions, attribute extraction instructions, frequency statistics instructions, emotion extraction instructions and output format instructions. It should be noted that the task instruction for visual asset extraction prompts requires the model, acting as an expert in comic book character and scene art design, to identify all core elements with visual presentation value from the script. For example: You are an expert proficient in novel analysis and comic book character art design. Please extract all core characters, main scenes, and key props from the following script. The asset classification instruction requires the model to distinguish between three types of assets: characters, scenes, and props, and extract key attributes for each type of asset. The attribute extraction instruction requires extracting gender, age, physical characteristics, clothing features, and personality tags for characters; architectural style, environmental atmosphere, and color scheme for scenes; and appearance, material, and size for props. The frequency statistics instruction requires the model to count the episodes in which each asset appears. The emotion extraction instruction requires the model to identify the typical emotional atmosphere when an asset appears, for use in subsequent prompt generation. The output format instruction requires the model to output an asset list in a structured data format, for example: Please output in NDJSON format, with each asset record including asset type, asset name, attribute description, list of episodes in which it appears, frequency of appearance, and emotion tag.

[0040] S32, concatenate the asset extraction prompts with the script corpus and input them into the second text model to identify visual entities and output an initial asset list with attribute descriptions and frequency statistics for each visual entity; S33, calculate the similarity value between any two visual entity attribute descriptions based on semantic similarity. When the similarity value is greater than the preset similarity threshold, determine to merge the two assets corresponding to the two visual entity attribute descriptions, and update the list of episodes and frequency of occurrence of the merged assets. It should be noted that, due to the diversity of natural language, the same visual entity may be referred to by different expressions in the script. To solve this problem, the system performs synonym merging on the initial asset list. Specifically, for any two asset records in the initial asset list, their attribute description text is obtained. This text embedding model is used to convert the attribute descriptions into semantic vectors, and the cosine similarity between the two vectors is calculated to obtain a similarity value. If the similarity value is greater than a preset similarity threshold, the two assets are determined to point to the same visual entity. For two asset records identified as the same entity, a merge operation is performed: Merged asset name: Select the name with higher frequency or more complete description, or retain both as aliases; Merged attribute description: Take the union of the attribute descriptions of the two records, remove duplicates, and form a more complete attribute description; Merged list of appearing episodes: Take the union of the list of appearing episodes of the two records; Merged frequency of appearance: Update to the length of the merged list of appearing episodes; Merged sentiment tag: Take the union of the sentiment tags of the two records, or select the primary sentiment based on a weighted average of frequency of appearance. After the merge operation is completed, update the asset list and record the mapping relationship before and after the merge for easy traceability.

[0041] S34. Based on the global requirement of spanning at least two target series units, the merged asset list is filtered to retain assets with an occurrence frequency of not less than 2, and the final visual asset list is generated. S35, for each asset in the final visual asset list, extract visual features from the visual entity attribute description, obtain style type from global configuration information, extract emotion type from the emotion of the target drama unit with the highest asset frequency, combine visual features, style type and emotion type to form the first prompt word, and generate multiple versions of the first prompt word according to preset multi-angle requirements, including different perspectives, different expressions or different lighting conditions. It should be noted that visual feature extraction involves extracting core visual features from the asset's attribute description. For characters, this includes extracting gender, age, appearance features, clothing characteristics, hairstyle, etc.; for scenes, it includes extracting architectural style, environmental atmosphere, color tone, lighting, etc.; and for props, it includes extracting appearance, material, size, color, etc. Emotion type extraction involves obtaining the typical emotion when the asset appears from the asset's emotion tag field. If the asset appears in multiple episodes, the emotional tone corresponding to the episode with the highest frequency of appearance is selected, or a fusion of multiple emotions is taken. The above elements are combined into structured prompt text. Meanwhile, the system generates multiple versions of the first prompt word for the same asset based on preset multi-angle generation requirements, in order to meet the needs of different shot angles and lighting conditions in subsequent storyboard creation. The multi-angle requirements include different perspectives, different expressions, or different lighting conditions. The generated multiple versions of prompt words are uniformly associated with the same asset identifier and the version number is recorded.

[0042] S36. Store the list of visual assets and their corresponding first prompt words into the shared asset library. Each visual asset includes an asset identifier, asset type, asset name, attribute description, list of first prompt word versions, asset status, and asset version number.

[0043] In this embodiment, the problem of identifying different representations of the same entity is solved through a synonym merging mechanism, and the global reuse value of assets is guaranteed through frequency filtering. The introduction of emotion tags and the generation of multi-angle prompts inject emotional dimensions and perspective diversity into subsequent image generation, improving the richness and usability of the asset library. The finally established shared asset library provides a unified visual material foundation for the parallel creation of multiple series, fundamentally ensuring the visual consistency of cross-series content.

[0044] S4. Input the first prompt word into the image generation model to generate the corresponding visual asset image, perform image standardization processing, and store it in the shared asset library. It should be noted that in step S4, the system reads visual asset records in the pending generation state from the shared asset library, obtains the first prompt word corresponding to each asset and the preset multi-angle generation requirements; then, it sequentially inputs each first prompt word into the specified image generation model, calls the model to generate the corresponding visual asset image; for multi-angle requirements of the same asset, multiple versions of the image are generated through multiple calls or parameter adjustments; the generated images first undergo quality inspection, including image integrity inspection and sharpness inspection, and triggers a retry mechanism to regenerate images that fail the inspection. If the retry still fails, it is marked as a generation anomaly and manual processing is notified; images that pass the quality inspection... The process involves standardization, including cropping and scaling to a uniform size, format conversion, and standardized naming. The processed image files are then stored in the shared asset library's storage system, and the access path for each file is obtained. The system writes the access path back to the corresponding asset's image list, associating it with the asset identifier, asset type, attribute description, and prompt word version. Simultaneously, the asset status is updated to "generated" and the version number is recorded. Finally, a multi-dimensional search index is established for the shared asset library, including a classification index based on asset type, a keyword index based on asset name, and a semantic vector index based on attribute description. This enables subsequent steps to quickly retrieve matching visual assets based on semantic similarity.

[0045] S5, based on the visual asset images in the shared asset library, uses a third text big model for identification to generate the storyboard information and corresponding second prompt words for each target drama unit; Step S5 includes the following sub-steps: S51, acquire the script data of the target drama unit, and construct storyboard generation prompts to guide the third text model. The storyboard generation prompts include task instructions, asset reference instructions, storyboard element instructions, second prompt generation instructions, and output format instructions. It should be noted that the task instructions require the model to act as a film director and storyboard artist, designing detailed storyboards based on the script. For example: You are an experienced film director and storyboard artist. Please design detailed storyboards for this episode based on the following script and the available visual asset library; the asset referencing instruction requires that each storyboard must reference visual assets from the shared asset library. For example: each storyboard must reference character, scene, and prop assets from the asset library; the storyboard element instruction requires the output to include the storyboard number, shot size, camera movement, on-screen characters, props used, scene setting, plot description, and dialogue; for example: each storyboard should include: storyboard number, shot size, camera movement, on-screen characters, props used, scene setting, plot description, and dialogue; the second cue word generation instruction requires generating a second cue word for each storyboard for video generation, for example, generating one second cue word for each storyboard for video generation, which must include a description of the visual characteristics of the referenced assets and incorporate style and emotional information; the output format instruction requires output in a structured data format; for example: please output the complete storyboard list in JSON array format.

[0046] S52, after concatenating the storyboard generation prompts with the script data of the target series, input the third text model. The third text model recognizes the script content, generates a storyboard sequence, and retrieves matching visual assets from the shared asset library based on the character, scene and prop names in the plot description, and obtains the asset identifier of the corresponding visual asset. S53, for each scene, embed the retrieved asset identifier into the scene information, bind the scene to the shared asset library reference, and output the scene information, which includes scene number, shot type, camera movement method, on-screen character asset identifier, prop asset identifier, scene asset identifier, plot description text and lines. S54, based on the storyboard information and the referenced visual assets, the third text model generates a corresponding second cue word for each storyboard. The second cue word includes style prefix, image description, asset reference mark, mood label, and lighting and tone.

[0047] S6 generates dubbing audio files and corresponding timestamp information based on the storyboard information through a speech synthesis model; Step S6 includes the following sub-steps: S61, extract all dialogue text from the storyboard information of the target drama unit, and sort them according to the storyboard order to form a dialogue sequence. Each line of dialogue has a corresponding storyboard number and character identifier. S62, based on the character identifier of each line, matches the corresponding timbre parameters from the preset timbre library; S63: Input the dialogue text and timbre parameters into the speech synthesis model, generate corresponding dubbing audio segments one by one, and record the duration of each audio segment; S64, based on the duration and order of each audio segment, calculate the start and end times of each line in the entire audio stream, and generate a subtitle file with timestamp information; S65 stores the dubbed audio clips and their corresponding timestamps in the material library of the target drama unit and associates them with the corresponding storyboard number.

[0048] It should be noted that the process of obtaining storyboard dialogue information with timestamps and character identifiers involves selecting or cloning a matching voice from a preset voice library based on the character's identity in the dialogue (e.g., "the protagonist Lin Feng is a 20-year-old young male"). Then, it calls a highly realistic text-to-speech (TTS) model to generate a corresponding dubbing audio file for each line of dialogue and accurately marks its timestamp information for subsequent alignment with the video.

[0049] S7: Based on the second cue words corresponding to each storyboard information, the corresponding storyboard video clips are generated through the video generation model. Step S7 includes the following sub-steps: S71, obtain the second prompt word corresponding to each scene and the corresponding referenced visual asset identifier, and obtain the corresponding visual asset image access path from the shared asset library based on the visual asset identifier; It should be noted that the system reads the second cue word corresponding to each storyboard and the visual asset identifier referenced in the second cue word from the creation data table of the episode unit. Based on the asset identifier, it retrieves the corresponding visual asset image access path from the shared asset library. If the second cue word contains multiple asset references, it retrieves images of all relevant assets.

[0050] S72, taking the second cue word and the referenced visual asset image as input, generates corresponding video segments for each scene through a video generation model; It should be noted that for each storyboard, the second cue word and the referenced visual asset image are taken as input, and the specified video generation model (either Keling or RunwayGen-2 model can be selected) is called to generate the corresponding video segment. The model input methods include: taking the second cue word as the main input and the visual asset image as the reference image, and constraining the consistency between the main subject and the reference image in the generated video through the image guidance module; for storyboards that require precise control of the start and end frames, the asset image can be taken as the first frame or the last frame as input; for complex multi-character scenes, multiple character asset images can be taken as references at the same time, and the model output is a short video segment.

[0051] S73 stores video clips according to their storyboard numbers into the material library of the corresponding target drama unit and associates them with the corresponding storyboard information.

[0052] S8 splices all storyboard video clips in sequence, overlays the corresponding dubbing audio files according to the timestamp information, and outputs the complete comic book video of the target episode unit.

[0053] Step S8 includes the following sub-steps: S81: Obtain all storyboard video clips, sort them according to the storyboard number, and then splice the sorted storyboard video clips together to form a video stream. S82, obtain the dubbing audio file and timestamp information, and overlay the dubbing audio onto the corresponding time position of the video stream according to the timestamp; S83 generates subtitles based on the dialogue text and timestamp information, and then integrates the subtitles into the video; S84 retrieves the video resolution parameters from the global configuration information, encodes and encapsulates the output video file, and outputs a complete animated video file.

[0054] It should be noted that by setting global configuration information to uniformly constrain the project, a large text model is used to intelligently parse and segment the original script, generating structured script data for each episode unit. Then, a second text model extracts global visual assets such as characters, scenes, and props that run throughout the entire series from the multi-episode scripts, and calls an image generation model to generate standardized visual asset images and store them in a shared asset library. For any target episode, a third text model retrieves and references relevant visual assets from the shared asset library to generate structured storyboard information and corresponding second prompts. Then, a speech synthesis model and a video generation model are used to generate dubbing audio and storyboard video clips, which are finally automatically synthesized into a complete episode video. When multiple episodes are created in parallel, each episode unit accesses the same shared asset library and references the same visual asset images, ensuring visual consistency across episode content. This solves the technical problems of process fragmentation, poor visual consistency of multi-episode content, inability to reuse assets, and lack of support for multi-person parallel creation in existing technologies, improving the production efficiency, quality control, and team collaboration capabilities of long-form comics.

[0055] In the above process, once the project enters the storyboard generation stage, this method supports collaborative parallel creation by multiple creators. For example, creator A is responsible for episode 1, and creator B is responsible for episode 2. Their respective creation terminals send creation requests to the system for different episode units. After verifying permissions, the system allows them to simultaneously start their respective independent episode creation units. When these two units are running, they both access and strictly adhere to the global project data and visual assets in the shared asset library established in the steps, but execute their respective steps S5 to S8 without interference. This allows a long-form animated series project to quickly complete the production of all episodes in parallel, greatly shortening the project cycle.

[0056] Secondly, the present invention also provides a multi-episode comic book collaborative creation system, implemented using a multi-episode comic book collaborative creation method, including: The configuration settings module is used to set the global configuration information for the target comic book creation; The series segmentation module is used to obtain the original script text, identify the original script text through the first text model, and segment the original script text into multiple target series units based on global configuration information, and generate script data for the target series units. The visual asset generation module is used to extract the visual assets of each target drama unit based on the script data of the target drama unit through the second text model, and generate the corresponding first prompt words. The visual asset image generation module is used to input the first prompt word into the image generation model, generate the corresponding visual asset image, perform image standardization processing, and store it in the shared asset library. The storyboard information generation module is used to generate storyboard information and corresponding second prompt words for each target drama unit by recognizing visual asset images from the shared asset library through a third text big data model. The audio generation module is used to generate dubbing audio files and corresponding timestamp information based on the storyboard information and a speech synthesis model. The video generation module is used to generate corresponding storyboard video clips based on the second cue words corresponding to each storyboard information through a video generation model. The compositing output module is used to splice all the storyboard video clips in sequence, overlay the corresponding dubbing audio files according to the timestamp information, and output the complete comic book video of the target episode unit.

[0057] It should be noted that this system corresponds to the aforementioned method for collaborative creation of multi-episode comics. All implementation methods in the above method embodiments are applicable to the embodiments of this system and can achieve the same technical effect.

[0058] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0059] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0060] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0061] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0062] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0063] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0064] Furthermore, it should be noted that in the system and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or a combination thereof. This is something that those skilled in the art can achieve by using their basic programming skills after reading the description of the present invention.

[0065] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing system. The computing system can be a known general-purpose system. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0066] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for collaborative creation of multi-episode comics, characterized in that, Includes the following steps: S1 sets the global configuration information for creating the comic book; S2, obtain the original script text, identify the original script text through the first text model, and combine it with global configuration information to split the original script text into multiple target drama units, and generate script data for the target drama units; S3, based on the script data of the target drama series unit, extracts the visual assets of each target drama series unit through the second text model and generates the corresponding first prompt word; S4. Input the first prompt word into the image generation model to generate the corresponding visual asset image, perform image standardization processing, and store it in the shared asset library. S5, based on the visual asset images in the shared asset library, uses a third text big model for identification to generate the storyboard information and corresponding second prompt words for each target drama unit; S6 generates dubbing audio files and corresponding timestamp information based on the storyboard information through a speech synthesis model; S7: Based on the second cue words corresponding to each storyboard information, the corresponding storyboard video clips are generated through the video generation model. S8 splices all storyboard video clips in sequence, overlays the corresponding dubbing audio files according to the timestamp information, and outputs the complete comic book video of the target episode unit.

2. The multi-episode comic book collaborative creation method as described in claim 1, characterized in that, The global configuration information mentioned in step S1 includes project name, number of episodes, style type, content description and video resolution, and selects the model type and version used for creation from the model library. The model types include text model, image generation model, video generation model, speech synthesis model and music generation model.

3. The multi-episode comic book collaborative creation method as described in claim 2, characterized in that, Step S2, which involves obtaining the original script text, recognizing the original script text using a first text model, splitting the original script text into multiple target episode units based on global configuration information, and generating script data for the target episode units, includes the following sub-steps: S21, Obtain the original script text and perform text data preprocessing to obtain the standard script text; S22, obtain the number of episodes in the global configuration information, and construct episode prompts to guide the first text model. The episode prompts include task instructions, plot analysis instructions, integrity constraint instructions and output format instructions. S23, after concatenating the episode prompts with the standard script text, input them into the first text model to generate a candidate episode boundary set, and adjust the boundaries of the candidate episode boundary set according to the number of episodes to obtain the final episode boundary set; S24. The standard script text is divided into multiple series units according to the final episode boundary set. The text content of the multiple series units is re-input into the first text model for text optimization processing to obtain the script data of each target series unit. The script data includes the unique identifier of the target series unit, the series number, the plot points, the emotional tone, the list of characters, the list of scenes, the lines, the style tags, the text length, and the estimated duration.

4. The multi-episode comic book collaborative creation method as described in claim 3, characterized in that, Step S23, which involves adjusting the boundaries of the candidate episode boundary set based on the number of episodes to obtain the final episode boundary set, includes the following sub-steps: S231, Obtain the total number of words in the standard script text, and calculate the theoretical number of words per episode based on the number of episodes; S232, take the position corresponding to the theoretical number of words in each episode as the expected center point, and set an elastic offset range, and select all candidate episode boundaries that are within the elastic offset range from the candidate episode boundary set; S233, if candidate episode boundaries exist after screening, calculate the plot integrity score corresponding to each candidate episode boundary and the word count deviation value between the position of each candidate episode boundary and the expected center point; S234. Based on the plot integrity score and word count deviation value, calculate the comprehensive score of each candidate episode boundary after screening, and select the candidate boundary with the highest comprehensive score as the end boundary of the current target episode unit. S235, if no candidate boundary is found after filtering, then the candidate boundaries adjacent to the elastic offset range before and after are obtained respectively as the front candidate boundary and the back candidate boundary. Obtain the word count deviation value and plot integrity score corresponding to the previous candidate boundary and the next candidate boundary. If the word count deviation of the previous candidate boundary or the next candidate boundary is less than the preset deviation threshold and the plot integrity score is greater than the preset integrity threshold, then select the candidate boundary with the smaller word count deviation value between the previous candidate boundary and the next candidate boundary as the end boundary of the current target episode unit. If the word count deviation values ​​of the previous candidate boundary and the subsequent candidate boundary are both greater than the preset deviation threshold, or the plot integrity scores are both less than the preset integrity threshold, then the semantic boundaries are searched before and after the expected center point, and the semantic boundaries are used as the end boundaries of the current target episode unit. S236, using the end boundary of the current target episode unit as the starting position of the next target episode unit, repeat steps S232 to S235 to obtain the final episode boundary set.

5. The multi-episode comic book collaborative creation method as described in claim 4, characterized in that, Step S3, which involves extracting visual assets for each target drama unit based on the script data of the target drama unit using a second text model and generating corresponding first cue words, includes the following sub-steps: S31, acquire script data of all target drama units to generate script corpus, and construct visual asset extraction prompts to guide the second text model. Visual asset extraction prompts include task instructions, asset classification instructions, attribute extraction instructions, frequency statistics instructions, emotion extraction instructions and output format instructions. S32, concatenate the asset extraction prompts with the script corpus and input them into the second text model to identify visual entities and output an initial asset list with attribute descriptions and frequency statistics for each visual entity; S33, calculate the similarity value between any two visual entity attribute descriptions based on semantic similarity. When the similarity value is greater than the preset similarity threshold, determine to merge the two assets corresponding to the two visual entity attribute descriptions, and update the list of episodes and frequency of occurrence of the merged assets. S34. Based on the global requirement of spanning at least two target series units, the merged asset list is filtered to retain assets with an occurrence frequency of not less than 2, and the final visual asset list is generated. S35, for each asset in the final visual asset list, extract visual features from the visual entity attribute description, obtain style type from global configuration information, extract emotion type from the emotion of the target drama unit with the highest asset frequency, combine visual features, style type and emotion type to form the first prompt word, and generate multiple versions of the first prompt word according to preset multi-angle requirements, including different perspectives, different expressions or different lighting conditions. S36. Store the list of visual assets and their corresponding first prompt words into the shared asset library. Each visual asset includes an asset identifier, asset type, asset name, attribute description, list of first prompt word versions, asset status, and asset version number.

6. The multi-episode comic book collaborative creation method as described in claim 5, characterized in that, Step S5, which involves identifying visual asset images from the shared asset library using a third-party text model to generate storyboard information and corresponding second prompts for each target episode unit, includes the following sub-steps: S51, acquire the script data of the target drama unit, and construct storyboard generation prompts to guide the third text model. The storyboard generation prompts include task instructions, asset reference instructions, storyboard element instructions, second prompt generation instructions, and output format instructions. S52, after concatenating the storyboard generation prompts with the script data of the target series, input the third text model. The third text model recognizes the script content, generates a storyboard sequence, and retrieves matching visual assets from the shared asset library based on the character, scene and prop names in the plot description, and obtains the asset identifier of the corresponding visual asset. S53, for each scene, embed the retrieved asset identifier into the scene information, bind the scene to the shared asset library reference, and output the scene information, which includes scene number, shot type, camera movement method, on-screen character asset identifier, prop asset identifier, scene asset identifier, plot description text and lines. S54, based on the storyboard information and the referenced visual assets, the third text model generates a corresponding second cue word for each storyboard. The second cue word includes style prefix, image description, asset reference mark, mood label, and lighting and tone.

7. The multi-episode comic book collaborative creation method as described in claim 6, characterized in that, Step S6, which involves generating a dubbing audio file and corresponding timestamp information based on the storyboard information using a speech synthesis model, includes the following sub-steps: S61, extract all dialogue text from the storyboard information of the target drama unit, and sort them according to the storyboard order to form a dialogue sequence. Each line of dialogue has a corresponding storyboard number and character identifier. S62, based on the character identifier of each line, matches the corresponding timbre parameters from the preset timbre library; S63: Input the dialogue text and timbre parameters into the speech synthesis model, generate corresponding dubbing audio segments one by one, and record the duration of each audio segment; S64, based on the duration and order of each audio segment, calculate the start and end times of each line in the entire audio stream, and generate a subtitle file with timestamp information; S65 stores the dubbed audio clips and their corresponding timestamps in the material library of the target drama unit and associates them with the corresponding storyboard number.

8. The multi-episode comic book collaborative creation method as described in claim 7, characterized in that, Step S7, which involves generating corresponding storyboard video clips based on the second cue words corresponding to each storyboard information using a video generation model, includes the following sub-steps: S71, obtain the second prompt word corresponding to each scene and the corresponding referenced visual asset identifier, and obtain the corresponding visual asset image access path from the shared asset library based on the visual asset identifier; S72, taking the second cue word and the referenced visual asset image as input, generates corresponding video segments for each scene through a video generation model; S73 stores video clips according to their storyboard numbers into the material library of the corresponding target drama unit and associates them with the corresponding storyboard information.

9. The multi-episode comic book collaborative creation method as described in claim 8, characterized in that, Step S8, which involves sequentially splicing all storyboard video clips, overlaying corresponding audio files based on timestamp information, and outputting the complete animated video of the target episode unit, includes the following sub-steps: S81: Obtain all storyboard video clips, sort them according to the storyboard number, and then splice the sorted storyboard video clips together to form a video stream. S82, obtain the dubbing audio file and timestamp information, and overlay the dubbing audio onto the corresponding time position of the video stream according to the timestamp; S83 generates subtitles based on the dialogue text and timestamp information, and then integrates the subtitles into the video; S84 retrieves the video resolution parameters from the global configuration information, encodes and encapsulates the output video file, and outputs a complete animated video file.

10. A multi-episode comic book collaborative creation system, implemented using the multi-episode comic book collaborative creation method as described in any one of claims 1-9, characterized in that, include: The configuration settings module is used to set the global configuration information for the target comic book creation; The series segmentation module is used to obtain the original script text, identify the original script text through the first text model, and segment the original script text into multiple target series units based on global configuration information, and generate script data for the target series units. The visual asset generation module is used to extract the visual assets of each target drama unit based on the script data of the target drama unit through the second text model, and generate the corresponding first prompt words. The visual asset image generation module is used to input the first prompt word into the image generation model, generate the corresponding visual asset image, perform image standardization processing, and store it in the shared asset library. The storyboard information generation module is used to generate storyboard information and corresponding second prompt words for each target drama unit by recognizing visual asset images from the shared asset library through a third text big data model. The audio generation module is used to generate dubbing audio files and corresponding timestamp information based on the storyboard information and a speech synthesis model. The video generation module is used to generate corresponding storyboard video clips based on the second cue words corresponding to each storyboard information through a video generation model. The compositing output module is used to splice all the storyboard video clips in sequence, overlay the corresponding dubbing audio files according to the timestamp information, and output the complete comic book video of the target episode unit.

Images