A long-text-oriented adaptation-oriented comic script breakdown generation method and system

Abstract

The application discloses a long-text-oriented episodic screenplay generation method and system, and belongs to the technical field of artificial intelligence content generation and digital video production. The method forms multiple episode objects through voice-over role initialization, text semantic compression and episode segmentation, and extracts role candidate objects and location time sequence candidate objects; based on normalization rules, an identity register of roles and a location time sequence register are constructed respectively, and unified identification reuse or new registration is performed on the candidate objects; under the constraint of the register, scene planning results and structured screenplay constraint patterns are generated, and multiple generation agents are called to output screenplay objects; and then, through structure verification, field completion and unified field transformation, the standardized results of episodes, roles, locations and screenplays are output. The application can improve the consistency of cross-episode roles and locations, the stability of the screenplay structure and the reuse efficiency of downstream production.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence content generation, automatic adaptation of long texts, digital video production, and multi-agent collaborative control technology, specifically to a method and system for generating storyboards for comic book adaptations of long texts. Background Technology

[0002] Existing long text adaptation techniques typically employ a single large language model to perform a one-time inference on the entire text. This type of approach is limited by the length of the context window and is prone to problems such as plot compression, loss of context, character title drift, unbalanced scene segmentation, and loose output structure.

[0003] Especially in multi-episode adaptation scenarios, the names, appearance information, voice configurations, and narrative positioning of the same character in different episodes are prone to drift; the same location may be repeatedly created or incorrectly merged at different time periods, resulting in poor cross-episode identity consistency, which in turn affects downstream keyframe generation, video generation, dubbing synthesis, and resource reuse.

[0004] Furthermore, existing technologies lack a mechanism to uniformly constrain characters, locations, scenes, and storyboards into a standardized structure. Most solutions rely on free text output or weakly structured output, making it difficult to guarantee field integrity, original text alignment, and identifier consistency. This results in a significant amount of manual cleaning and mapping required during database import, task scheduling, and front-end canvas display.

[0005] Therefore, there is an urgent need for a generation method and system that can maintain consistency of characters and locations across series for long text adaptation tasks and improve the consumability of storyboard results. Summary of the Invention

[0006] This invention aims to solve the problems of context distortion, character and location identity drift, unstable storyboard structure, and difficulty in direct reuse by downstream systems in the existing long-text comic adaptation process. It provides a multi-agent storyboard generation method and system based on identity registry constraints for long-text comic adaptation.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: This invention constructs a multi-stage collaborative generation framework for adapting long-text comics. The framework includes at least a text semantic compression unit, a narrator character initialization unit, a series segmentation unit, a character identity registry, a location time sequence registry, a scene planning unit, and a structured storyboard generation unit.

[0008] Step S101: Receive the original narrative text, style control parameters and task identifier, call the narration initialization agent to perform long-term narrative character extraction on the original narrative text, and generate a narration configuration object including narration name, narration description, visual cues and text-to-speech parameters.

[0009] Step S102: Perform semantic compression rewriting and episode segmentation on the original narrative text, compress the redundant narrative into intermediate text suitable for structured processing, and split it into multiple episode objects according to the plot rhythm, text budget or narrative integrity requirements.

[0010] Step S103: Perform character extraction and location-time extraction for each of the aforementioned drama series objects to obtain a set of candidate characters and a set of candidate locations-time objects; wherein the candidate characters include at least the character name, character description, visual cues and text-to-speech parameters, and the candidate locations-time objects include at least the location name, time elements, environmental description and visual cues.

[0011] Step S104: Construct the character identity registry and the location time sequence registry based on the character name standardization rules, character alias mapping rules, and location name and time element combination rules, respectively; when a candidate object matches an existing normalized key, reuse the existing unified identifier, visual cue identifier, and compatible fields; when no match is found, create a new unified identifier and write it into the corresponding registry to form a reusable identity layer across series.

[0012] Step S105: Generate scene planning results using the character identity registry and location time sequence registry as constraint boundaries, and further construct a structured storyboard constraint pattern; the structured storyboard constraint pattern at least limits the legal character set, legal location set, maximum number of shots, maximum shot duration, maximum original text length per shot, field type rules, required field rules, and original text alignment rules. Under this constraint, call one or more generation proxies to output scene objects and storyboard objects.

[0013] Step S106: Perform structure verification, field completion, and unified field transformation on the generated results, converting scene objects and storyboard objects into episode arrays, character arrays, location arrays, and storyboard arrays for direct reuse in database entry, canvas display, keyframe generation, and video production.

[0014] Preferably, the role object adopts a dual-identification system, including at least a unified role identifier and a visual cue identifier; the location object adopts a unified location identifier and a location cue identifier. The dual-identification system enables the same role to share the same identity semantics in the dialogue analysis, visual generation, and speech generation links, while different generation agents achieve consistent output by referencing the corresponding identifiers.

[0015] Preferably, the storyboard generation unit adopts a strictly structured output mode. The storyboard object includes at least scene identifier, shot identifier, shot number, character set, original text fragment, environment description, lighting description, camera movement description, dialogue analysis, first frame cue information, video cue information, and duration field. Among them, the speaker identifier in the dialogue analysis can only be selected from the set of visual cue identifiers registered in the character identity registry, thereby suppressing the model's fabrication of characters and incorrect character assignment.

[0016] Preferably, when there is no identifiable dialogue segment in the storyboard object, or the dialogue is insufficient to carry the plot progression, the system writes the narration content corresponding to the narration configuration object into the storyboard object; when the original text segment corresponding to a single shot exceeds a preset length threshold, the system performs shot subdivision to maintain narrative continuity and shot granularity balance.

[0017] Beneficial effects: Compared with the prior art, the present invention has at least the following beneficial effects: 1. By establishing a unified identity layer across series through a character identity registry and a location time-series registry, issues such as character prompt drift, location name conflicts, and scene duplication can be significantly reduced.

[0018] 2. By incorporating character whitelists, location whitelists, shot thresholds, field rules, and original text alignment rules into the structured storyboard constraint mode, we can effectively reduce field omissions, structural misalignments, and out-of-bounds references caused by free generation, thereby improving the engineering consumability of the output results.

[0019] 3. By using multi-agent collaborative generation to replace single full-text reasoning, the stability of storyboards, narrative coherence, and success rate of downstream video production can be improved in long text adaptation scenarios.

[0020] 4. By performing structural validation, field completion, and unified field transformation during the result output stage, the costs of post-processing and manual cleaning can be reduced, and the data connection efficiency between database entry, canvas display, and video production links can be improved. Attached Figure Description

[0021] Figure 1 This is an architecture diagram of a multi-agent comic storyboard generation system based on identity registry constraints in an embodiment of the present invention.

[0022] Figure 2 This is a flowchart illustrating the long-text comic book adaptation and structured storyboard generation process in an embodiment of the present invention. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and preferred embodiments, but the scope of protection of the present invention is not limited thereto.

[0024] Example 1: Adaptation process of long text comics by multiple agents.

[0025] The system first receives the original long text, style control parameters, and task identifier, and then calls the narration initialization agent to generate a long-term stable narration configuration object. The narration configuration object includes the narration name, narration description, visual cues, and text-to-speech parameters, which are used for subsequent narration dubbing consistency control.

[0026] Subsequently, the system performs text semantic compression and episode segmentation. For the compressed intermediate text, the system generates corresponding episode identifiers according to the episode numbers and establishes episode sequence number, episode title, and episode text fields.

[0027] During the role and location extraction phase, the system generates a set of candidate roles and a set of candidate locations based on time sequence. For candidate roles, at least the role name, appearance description, visual cues, and text-to-speech parameters are generated; for candidate locations based on time sequence, at least the location name, time element, environmental description, and visual cues are generated.

[0028] During the registry construction phase, the system builds a character identity registry and a location time sequence registry based on rules for character name standardization, alias merging, and location name plus time element merging. For character objects, the system generates a unified character identifier and a visual cue identifier; for location objects, the system generates a unified location identifier and a location cue identifier. Subsequent scene planning and storyboard generation will no longer directly use free text names, but will instead reference the aforementioned unified identifiers.

[0029] During the scene planning phase, the system breaks down a single episode into multiple scene objects. Each scene object includes at least the following fields: episode identifier, scene sequence number, unified location identifier, location hint identifier, character set, and original text excerpt. These scene objects serve as internal planning layer objects, used only for subsequent storyboard generation boundary constraints, and are not directly used as the final delivery result.

[0030] During the storyboard generation phase, the system calls the storyboard generation agent for each scene object, requiring it to output an array of shots. Each shot object must include at least a scene identifier, shot identifier, shot number, character set, original text fragment, environment description, lighting description, camera movement description, dialogue analysis, first frame cue information, video cue information, and duration field; among which, the first frame cue information and video cue information are preferably cue content suitable for processing by the visual generation model, and the speaker identifier in the dialogue analysis must be selected from the set of legal characters in the current scene.

[0031] Example 2: Identity reuse across TV series.

[0032] Taking the character "Sect Leader" as an example, if the unified identifier and visual cue identifier of the character have been registered in the first episode, then when the second episode extracts aliases such as "Sect Leader", "Sect Leader Person", and "Sect Master", it will be uniformly mapped to the existing character object through the normalization key and alias mapping rules, thereby ensuring that the same visual identity is used for dialogue analysis, keyframe generation and video generation.

[0033] Example 3: Structured output transformation.

[0034] After generating all scenes and shots, the system performs structural repair, field completion, and unified field transformation on the original proxy output, ultimately outputting four types of structured results: episodes, characters, locations, and storyboards. The storyboard results can be directly mapped to downstream video production tasks, while the character and location results can be directly mapped to the resource library and canvas layer.

[0035] Example 4: Narration completion and shot subdivision.

[0036] When the system identifies that a shot lacks identifiable dialogue or the dialogue is insufficient to support the plot progression, it adds narration to the corresponding shot. When the system identifies that the original text segment corresponding to a shot exceeds a preset length threshold, it further splits the shot into multiple sub-shots and inherits the character and location boundaries of the original scene to ensure that the adaptation results take into account both narrative integrity and production feasibility.

Claims

1. A multi-agent storyboard generation method based on identity registry constraints for adapting long-text comics, characterized in that, Includes the following steps: S101: Obtain the original narrative text, style control parameters, and task identifier, and call the large language model to perform narrator character initialization, generating a narrator configuration object containing narrator name, narrator description, visual cue information, and text-to-speech parameters; S102, perform semantic compression and episode segmentation on the original narrative text to obtain multiple episode objects; S103, extract a set of candidate characters and a set of candidate locations and time sequences for each of the aforementioned drama series objects; S104, construct a role identity registry based on the role name normalization rule, construct a location time sequence registry based on the location name and time element combination normalization rule, and perform unified identifier reuse or new registration on the role candidate object and the location time sequence candidate object. S105, generate scene planning results based on the role identity registry and location time sequence registry, and generate a structured storyboard constraint mode; under the constraints of the structured storyboard constraint mode, call one or more generation proxy output storyboard objects; S106, perform structural validation, field completion, and unified field transformation on the storyboard object, and output the episode array, character array, location array, and storyboard array.

2. The method according to claim 1, characterized in that, Each role object in the role identity registry is associated with at least a unified role identifier, a visual cue identifier, a role name, a set of role aliases, a role description, and text-to-speech parameters; each location object in the location time-series registry is associated with at least a unified location identifier, a location cue identifier, a location name, a time element, an environmental description, and visual cue information; the normalization rules in step S104 include at least any one of the following or a combination thereof: case uniformity, punctuation removal, standardization of honorifics and titles, alias mapping, synonym mapping, and time element standardization; when a candidate object matches a normalization key in the registry, the existing unified identifier and cue identifier are reused; when no match is found, a new unified identifier is generated and registered in the corresponding registry.

3. The method according to claim 1, characterized in that, The scene planning result is an internal planning layer object, which includes at least the episode identifier field, scene sequence number field, location unified identifier field, location hint identifier field, character set field, and original text fragment field. It is used to limit the boundaries of subsequent storyboard generation, and is not output as the final external delivery object. The structured storyboard constraint mode includes at least the character whitelist, location whitelist, shot number threshold, shot duration threshold, single shot original text length threshold, original text alignment rules, field type rules, and required field rules. It restricts the generated proxy output to conform to the preset structure through structured constraint decoding, pattern verification, or a combination of both.

4. The method according to claim 1, characterized in that, The storyboard object includes at least a scene identifier, shot identifier, shot number, character set, original text segment, environment description, lighting description, camera movement description, dialogue analysis, first frame prompt information, video prompt information, and duration field; wherein, the speaker identifier in the dialogue analysis comes from the set of visual cue identifiers in the character identity registry.

5. The method according to claim 1, characterized in that, The method further includes a narration completion step: when there is no identifiable dialogue item in the storyboard object, or the identifiable dialogue is insufficient to carry the plot progression, the narration content corresponding to the narration configuration object is written into the storyboard object to maintain narrative continuity; and / or when the original text segment corresponding to a single shot exceeds a preset length threshold, shot subdivision is performed to maintain shot granularity balance.

6. The method according to claim 1, characterized in that, The plurality of generation agents include at least one of the following types of agents or a combination thereof: scene planning agent, role extraction agent, location and timing extraction agent, storyboard generation agent, and structure verification agent; the plurality of generation agents exchange information and perform boundary constraints through the role identity registry, location and timing registry, and structured storyboard constraint mode.

7. The method according to claim 1, characterized in that, The field completion and unified field transformation in step S106 include at least one of the following or a combination thereof: completing missing identifier fields, unifying field naming, unifying enumeration value expression, unifying time and sequence number formats, and completing the associated fields corresponding to the original text fragments, so as to improve the consumability of the generated results in database entry, task scheduling, canvas display and video production.

8. A multi-agent storyboard generation system based on identity registry constraints for adapting long-text comics, characterized in that, It includes a text semantic compression module, a narrator character initialization module, a series segmentation module, a character extraction module, a location and time sequence extraction module, a character identity registry management module, a location and time sequence registry management module, a scene planning module, a structured storyboard generation module, and a result structured output module; the modules work together to execute the method described in any one of claims 1 to 7.

9. An electronic device, comprising a processor and a memory, characterized in that, The memory stores a computer program that, when executed by the processor, implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 7.

Images