A device for creating creative works based on a database of lawsuits and cases categorized by type and facts input by users

Abstract

The present invention relates to a creative work creation device based on a database of lawsuits and cases classified by type and facts input by a user, the device comprising: a memory unit; and a processor unit, wherein the processor unit comprises: a fact extraction tool configured to extract case individual information based on at least one of audio data and text data related to the facts; a scenario generation tool configured to generate a text-based scenario script based on the extracted case individual information; a creative work creation tool configured to generate a creative work based on the generated text-based scenario script; and a verification and correction tool configured to receive user feedback input based on the generated creative work, the case individual information including first case individual information which is basic information indicating an outline of the case and second case individual information which is information on the progress of the case since its occurrence, the memory unit configured to store a fact database for recording the first case individual information and the second case individual information, and the verification and correction tool configured to update the fact database based on the received user feedback.

Description

[Technical Field]

[0001] The present invention relates to a creative creation device based on a database of lawsuits and cases categorized by type and facts input by a user, and to an apparatus and method for modifying the created creative creation based on user feedback. [Background technology]

[0002] Traditionally, in fields such as legal consultation, insurance review, and medical consultation, the process of gathering facts about a case has been carried out by lawyers, clients, doctors, and other personnel directly listening to oral or written materials from clients, organizing and documenting them. This manual method not only takes a lot of time, but also increases the risk of important information being omitted or misrepresented if the client's memory of the case is incomplete or their expressiveness is inadequate. In particular, in cases where accurate facts are important, such as traffic accidents, medical negligence, and insurance claims, there is a high risk of confusion during the consultation or review process due to factual errors or unclear statements.

[0003] Recently, technologies have emerged that use natural language processing (NLP) to analyze client conversations and automatically extract key information related to a case. This technology analyzes client statements to help structure key information, such as the time of the incident, location, people involved, the extent of damage, and type of accident. The system can also improve the efficiency of existing consultation methods by automatically detecting missing information and supplementing it through follow-up questions. For example, when a client enters a statement like, "A collision with an SUV occurred in XX-dong, XX-gu, Seoul, around 3 PM on January 2, 2025," AI can automatically classify the time of the incident ("3 PM on January 2, 2025"), the location of the incident ("XX-dong, XX-gu, Seoul"), and the type of vehicle ("SUV"), and generate follow-up questions regarding the extent of the damage ("Were there any injuries?", "How bad is the vehicle?").

[0004] However, because these natural language processing-based systems provide information in text format, they have limitations, making it difficult for users to visually recall the actual incident situation and making it difficult to easily identify inaccurate information.In particular, in complex incidents such as traffic accidents, visual elements such as the vehicle's movement path, collision location, and impact strength are important, but the problem remains that it is difficult to accurately reconstruct the incident using only text-based explanations.

[0005] Recently, AI video generation technology has been developing, which uses artificial intelligence to convert text information into 3D or 2D images. For example, by using AI video generation engines such as SORA and VEO2, it is possible to input a text-based scenario script and automatically generate a virtual environment, characters, vehicles, backgrounds, etc., to visually recreate an actual accident scene. When this technology is applied to the consultation and legal fields, it can generate videos similar to actual incidents based on the facts analyzed by AI, allowing users to easily review the facts and correct errors.

[0006] For example, in the case of a traffic accident, data analyzed by AI can be used to automatically convert the content, "An SUV collided from behind with a passenger car waiting at a traffic light at an intersection at around 3:50 PM," into an AI video and provide a simulated video so that the user can review it and correct any errors. Furthermore, while viewing the AI-generated video, the user can provide feedback such as "The SUV was black, not white," or "The vehicle shook more violently at the time of the collision," and the system can then reflect this and generate an updated video.

[0007] By utilizing this AI-based visualization technology, clients and staff can more intuitively examine the facts of a case, improving the accuracy of the case. Furthermore, the video-based fact-analysis system can be applied to a variety of fields beyond legal consultations, including insurance reviews, accident analysis, and medical record reviews, and can even be used as legal evidence.

[0008] The present invention provides an AI-based fact extraction and visualization system to overcome the limitations of existing technologies. The system of the present invention automatically extracts case-related information using natural language processing (NLP) and then visually converts that information using AI video generation technology. This allows users to directly check the progress of a case through video rather than through simple text analysis, and if the video does not match the actual facts, they can make corrections by requesting corrections. Summary of the Invention [Problem to be solved by the invention]

[0009] In the existing evidence gathering process, lawyers, insurance company employees, and other personnel must listen to the client's statements, manually document them, and supplement the information through additional question and answer sessions. This process consumes a lot of time and money, and slows down the speed of case analysis, which is a problem that the present invention aims to solve.

[0010] In the existing evidence collection process, if the client is unable to accurately remember the circumstances at the time of the incident or is not able to express himself / herself properly, important information may be omitted or distorted. For example, detailed information such as the time of the accident, the color of the vehicle, and the process of the accident may be conveyed incorrectly. This invention aims to solve this problem.

[0011] Existing incident analysis systems provide information based on text, which requires users to read and imagine the incident. It is particularly difficult to reconstruct the incident from text alone, especially in cases where spatial and temporal factors are important, such as traffic accidents. Furthermore, if the flow of the incident cannot be accurately conveyed, lawyers and insurance adjusters are more likely to make incorrect decisions. This invention aims to solve this problem.

[0012] When a user reads text data and examines the facts, it is difficult to find incorrect information, and when a correction is required, the user must manually correct the text, which is a hassle. When a correction is required, a new document must often be created from the existing document, which increases the repetitive work, and this is the problem that the present invention aims to solve.

[0013] Existing text-based incident analysis systems are limited to specific legal consultations and insurance reviews, making them difficult to use in various fields such as medicine and education. For example, in the analysis of medical accidents and educational simulations, the flow of events must be visually reproduced, but existing methods have limitations that make it difficult to perform accurate simulations, and this invention aims to solve this problem. [Means for solving the problem]

[0014] The present invention relates to a device for generating creative works based on a database of lawsuits and cases categorized by type and facts input by a user, and includes a fact extraction tool configured to extract case individual information based on at least one of audio data and text data related to the facts, a scenario generation tool configured to generate a text-based scenario script based on the extracted case individual information, a creative work generation tool configured to generate the creative work based on the generated text-based scenario script, and a verification and correction tool configured to receive feedback input from the user based on the generated creative work, wherein the scenario generation tool is configured to compare the database of lawsuits and cases categorized by type with the extracted case individual information, and to confirm missing information for generating the text-based scenario script based on the comparison.

[0015] In addition, the incident individual information is configured to include at least one of time individual information related to the facts, place individual information related to the facts, vehicle individual information related to the facts, accident type individual information related to the facts, victim individual information related to the facts, situation individual information related to the facts, person individual information related to the facts, and behavior individual information related to the facts.

[0016] In addition, the fact extraction tool is configured to generate questions to extract the individual incident information, and the scenario generation tool is configured to generate additional questions to be provided to the user to supplement the confirmed missing information.

[0017] The fact extraction tool and the scenario generation tool are also configured to utilize a natural language processing model to generate questions or follow-up questions for the user.

[0018] In addition, the creation creation tool is configured to generate an image that visually embodies the generated text-based scenario script by utilizing a deep learning-based image generation model.

[0019] In addition, the creative creation tool is configured to generate at least one of the outer shape of a person in the video, the facial expression of a person in the video, the posture of a person in the video, and the movement of a person in the video by utilizing a deep learning-based image generation model.

[0020] The creation tool is also configured to convert lines or narration information included in the text-based scenario script into voice.

[0021] The creation tool is also configured to perform emotion-aware TTS to reflect emotions in the converted voice.

[0022] In addition, the creative creation tool is configured to generate sound effects within the video by utilizing a deep learning-based sound synthesis model.

[0023] The present invention provides a creative work creation device based on a database of lawsuits and cases categorized by type and facts input by a user, the device comprising: a memory unit; and a processor unit, wherein the processor unit comprises: a facts extraction tool configured to extract case individual information based on at least one of audio data and text data related to the facts; a scenario generation tool configured to generate a text-based scenario script based on the extracted case individual information; a creative work creation tool configured to generate the creative work based on the generated text-based scenario script; and a verification and correction tool configured to receive feedback input from the user based on the generated creative work, wherein the case individual information includes first case individual information which is basic information indicating an outline of the case, and second case individual information which is information on the progress of the case since its occurrence, the memory unit is configured to store a facts database for recording the first case individual information and the second case individual information, and the verification and correction tool is configured to update the facts database based on the received user feedback.

[0024] In addition, the first incident individual information is configured to include at least one of time individual information related to the facts, place individual information related to the facts, vehicle individual information related to the facts, accident type individual information related to the facts, victim individual information related to the facts, situation individual information related to the facts, person individual information related to the facts, and behavior individual information related to the facts.

[0025] In addition, the second incident individual information is configured to include at least one of information on the victim's physical reaction related to the facts, information on the victim's emotional expression related to the facts, information on the perpetrator's initial reaction related to the facts, information on third party witness reaction related to the facts, information on the occurrence of additional damage related to the facts, information on whether the accident has escalated related to the facts, and information on on-site response measures related to the facts.

[0026] The scenario generation tool is also configured to perform a similarity analysis between the case information by type stored in the lawsuit and case type database and the case individual information.

[0027] In addition, the litigation and case type database stores case information by type, and the case information by type includes at least one of information on the conditions under which the case occurred, information on a summary of the case, information on the development of the case, information on the outcome of the case, and information on statements by interested parties.

[0028] The validation and correction tool is also configured to provide a user interface to receive the user's feedback input based on the generated creation.

[0029] The verification and correction tool is also configured to provide a frame-by-frame search function for the creation so as to receive feedback from the user based on the created creation.

[0030] The verification and correction tool is also configured to provide zoom and slow motion functions for the creation to receive feedback from the user based on the generated creation.

[0031] The verification and correction tool is also configured to provide an error checklist function and an error detection function for the creation so as to receive feedback from the user based on the created creation.

[0032] The verification and correction tool is configured to detect an error and output a warning message if the generated creation is different from the extracted incident individual information.

[0033] The verification and correction tool is also configured to provide a function that allows the user to input feedback based on the generated creation in at least one of a text input feedback method, a visual selection feedback method, and a voice input feedback method.

[0034] The present invention provides a method for verifying a creative work generated based on a database of lawsuits and cases classified by type and facts input by a user, the method comprising: a case individual information extraction step in which a processor unit extracts case individual information based on at least one of audio data and text data related to the facts; a scenario script generation step in which the processor unit generates a text-based scenario script based on the extracted case individual information; a step in which the processor unit generates the creative work based on the generated text-based scenario script; and a feedback reflection step in which the processor unit receives feedback input from the user based on the generated creative work, wherein the case individual information extraction step further comprises a first case individual information extraction step in which first case individual information is basic information indicating an outline of the case; a second case individual information extraction step in which second case individual information is information on the progress of the case since its occurrence; and a step in which a memory unit stores the first case individual information and the second case individual information in a facts database, and the feedback reflection step further comprises a step in which the processor unit updates the facts database based on the received user feedback.

[0035] In addition, the first incident individual information extraction step further includes a step of extracting at least one of time individual information related to the factual relationship, place individual information related to the factual relationship, vehicle individual information related to the factual relationship, accident type individual information related to the factual relationship, victim individual information related to the factual relationship, situation individual information related to the factual relationship, person individual information related to the factual relationship, and behavior individual information related to the factual relationship.

[0036] In addition, the second incident individual information extraction step further includes a step of extracting at least one of information from the victim's physical reaction information related to the facts, the victim's emotional expression information related to the facts, the perpetrator's initial reaction information related to the facts, the third party's eyewitness reaction information related to the facts, the occurrence of additional damage related to the facts, the information on whether the accident has escalated related to the facts, and the on-site response measures information related to the facts.

[0037] In addition, the scenario script generation step further includes a similarity analysis execution step in which the processor unit performs a similarity analysis between the type-specific case information stored in the lawsuit and case type database and the case individual information, and a step in which the processor unit checks for leaked information about the case individual information based on the performed similarity analysis.

[0038] In addition, the similarity analysis performing step further includes a step in which the processor unit performs a similarity analysis between at least one of the incident occurrence condition information, incident summary information, incident development information, incident result information, and stakeholder speech information stored in the litigation and incident type database and the incident individual information.

[0039] The feedback reflection step may further include the processor providing a user interface to receive feedback from the user based on the generated creation.

[0040] The feedback reflecting step may further include the step of the processor unit providing zoom and slow motion functions for the creation so as to receive feedback from the user based on the created creation.

[0041] The feedback reflecting step may further include the step of the processor providing an error checklist function and an error detection function for the creation so as to receive feedback from the user based on the created creation.

[0042] The feedback reflecting step may further include a step in which, if the generated creation differs from the extracted incident individual information, the processor unit detects an error and outputs a warning message.

[0043] The feedback reflecting step further includes a step in which the processor unit provides a function that allows the user to input feedback in at least one of a text input feedback method, a visual selection feedback method, and a voice input feedback method based on the generated creation. [Effects of the Invention]

[0044] The present invention converts text-based facts into a visual simulation format, allowing users to easily understand and review them.

[0045] This invention recreates the progression of an incident through AI-generated video, allowing users to check the incident more intuitively than with simple documents and easily identify discrepancies in the actual facts.

[0046] According to the present invention, if a user finds an inaccurate part while viewing a video, the user can request a correction by text input, voice command, visual selection, or the like.

[0047] In this invention, AI extracts facts, automatically searches for missing information, and generates follow-up questions, reducing repetitive question and answer processes. In addition, by using AI video generation technology to deliver information in a case simulation format, it is expected that the client's understanding will improve, consultation time will be shortened, and consultation satisfaction will increase.

[0048] The present invention stores the final data that reflects the user's review and revision requests and provides it for use by lawyers, insurance companies, medical staff, etc. This reduces the burden of document creation and provides more accurate information.

[0049] This invention can be used in a variety of fields, including not only legal consultations and insurance reviews, but also medical care, education, and case analysis. For example, in medical negligence cases, AI can visually reproduce and analyze changes in a patient's condition and treatment process, and in the field of education, it can improve learning by simulating historical events and experimental results. [Brief explanation of the drawings]

[0050] [Figure 1a] FIG. 1a is a diagram illustrating the configuration of a fact-based creative production system according to the present invention. [Figure 1b] FIG. 1b is a diagram illustrating first and second case individual information of a factual database according to an embodiment of the present invention. [Figure 2a] FIG. 2a is a diagram illustrating a case individual information extraction process of a factual relationship extraction tool according to an embodiment of the present invention. [Figure 2b] FIG. 2b is a diagram illustrating tagging of a factual relationship extraction tool according to an embodiment of the present invention. [Figure 2c] FIG. 2c is a diagram illustrating that a fact extraction tool according to an embodiment of the present invention stores case individual information in a standardized format. [Figure 3a]FIG. 3a is a diagram illustrating a text-based scenario script generation step of the scenario generation tool according to the present invention. [Figure 3b] FIG. 3b is a diagram for explaining a summary of facts generated by the scenario generation tool according to the present invention. [Figure 4a] FIG. 4a is a diagram illustrating an example of scene 1 of a factual summary generated by a scenario generation tool according to an embodiment of the present invention. [Figure 4b] FIG. 4b is a diagram illustrating an example of scene 2 of a factual summary generated by a scenario generation tool according to an embodiment of the present invention. [Figure 4c] FIG. 4c is a diagram illustrating an example of scene 3 of a factual summary generated by a scenario generation tool according to an embodiment of the present invention. [Figure 5a] FIG. 5a is a diagram illustrating an example of scene 1 of a text-based scenario script generated by a scenario generation tool according to an embodiment of the present invention. [Figure 5b] FIG. 5b is a diagram illustrating an example of scene 2 of a text-based scenario script generated by a scenario generation tool according to an embodiment of the present invention. [Figure 5c] FIG. 5c is a diagram illustrating an example of scene 3 of a text-based scenario script generated by a scenario generation tool according to an embodiment of the present invention. [Figure 6] FIG. 6 is a diagram showing a step in which the creation tool according to the present invention creates an image based on a factual summary and a scenario script. [Figure 7a] FIG. 7a shows a scene before the accident occurs in the video generated by the creation creation tool by inputting the accident occurrence script of scene 1 shown in FIG. 5a. [Figure 7b] FIG. 7b shows a scene immediately after the accident occurs in the video generated by the creation creation tool by inputting the accident occurrence script of scene 1 shown in FIG. 5a. [Figure 7c]FIG. 7c is a diagram showing a scene of a car crashing into a guardrail immediately after the accident occurs in a video generated by the creation generation tool by inputting the accident occurrence script of scene 1 shown in FIG. 5a. [Figure 7d] Figure 7d is a diagram showing the scene after the accident occurs, in which the SUV driver gets out of the vehicle and approaches the victim vehicle, in a video generated by the creative creation tool by inputting the accident occurrence script for scene 2 shown in Figure 5b. [Figure 7e] Figure 7e is a diagram showing a scene in which the rear bumper of the passenger car and the front bumper of the SUV are dented after the accident occurs, as generated by the creative creation tool after inputting the accident occurrence script for scene 2 shown in Figure 5b. [Figure 7f] Figure 7f is a diagram showing a scene in which insurance company officials and police gather at the scene of the accident after the accident occurs, in a video generated by the creative creation tool by inputting the accident occurrence script for scene 3 shown in Figure 5c. [Figure 8] FIG. 8 is a diagram illustrating a modification request and scenario script update step of the verification and modification tool according to the present invention. [Figure 9a] FIG. 9a is a diagram illustrating a scene where a landlord and a tenant meet and greet each other on May 1, 2024 at Happy Real Estate, generated by a creation creation tool according to an embodiment of the present invention. [Figure 9b] FIG. 9b is a diagram showing a dialogue scene between a lessor and a lessee generated by a creation creation tool according to an embodiment of the present invention. [Figure 9c] FIG. 9c is a diagram showing a brokerage scene of a happy real estate agent, generated by a creation generation tool according to an embodiment of the present invention. [Figure 9d] FIG. 9d is a diagram illustrating a scene in which a lessor makes an utterance requesting that the down payment be made by May 10, 2024, generated by a creation creation tool according to an embodiment of the present invention. [Figure 9e] FIG. 9e is a diagram illustrating a scene in which a tenant and a happy real estate agent speak in response to a tenant's utterance, generated by a creation creation tool according to an embodiment of the present invention. [Figure 9f]FIG. 9f is a diagram illustrating a scene in which the lessor accepts the tenant's utterance, generated by the creation creation tool according to an embodiment of the present invention. [Figure 9g] FIG. 9g is a diagram showing a scene in which a lease agreement is created by a lessor and a lessee using a creation creation tool according to an embodiment of the present invention. [Figure 9h] FIG. 9h is a diagram showing the final scene of the cartoon of FIGS. 9a to 9g, generated by the creation generation tool according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0051] Specific details of the embodiments are included in the detailed description and the drawings.

[0052] The advantages and features of the present invention, as well as methods for achieving them, will become clearer with reference to the following detailed description of the embodiments in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, and can be embodied in various different forms. The present embodiments are provided merely to complete the disclosure of the present invention and to fully convey the scope of the invention to those skilled in the art to which the present invention pertains. The present invention is defined only by the scope of the claims. The same reference numerals refer to the same elements throughout the specification.

[0053] FIG. 1a is a diagram illustrating the configuration of a fact-based creative production system according to the present invention.

[0054] As shown in the figure, the fact-based creation creation system 10 may include a user terminal 100 and a server device 200. Specifically, the user terminal 100 includes a sensor unit 110, a communication unit 120, a processor unit 130, a memory unit 140, and an output unit 150. Although not shown in FIG. 1a, the server device 200 of the present invention may also include a memory unit, a communication unit, and a processor unit.

[0055] User terminal 100

[0056] Sensor unit 110

[0057] According to the present invention, the sensor unit 110 can collect factual data in various forms, such as a user's voice, gesture, video, image, or text input. For example, if a user describes the details of an incident through voice to the user terminal 100, the sensor unit 110 can convert the collected voice data into a digital signal and transmit it to the processor unit 130. In addition, the sensor unit 110 can include a camera, a touch sensor, a position sensor (GPS), environmental sensors such as illuminance, air pressure, temperature, and humidity, and a facial expression or gesture recognition sensor, and can detect additional real-world information other than voice provided by the user in the process of describing the incident.

[0058] For example, when a user takes a photo of an accident at a specific location or uploads documents such as a contract or medical certificate, the sensor unit 110 acquires the image and document data and collects associated temporal meta information such as the time of the photo and location, thereby providing insight into the context of the facts. In addition, even if the user cannot clearly remember the time or weather information at the time of the incident, the sensor unit 110 can estimate background elements necessary for generating an incident scenario by referring to the real-time location and weather information sensed by the sensor unit 110.

[0059] According to an embodiment of the present invention, the sensor unit 110 can detect a user's real-time reactions and indirectly capture non-verbal reactions such as a sense of strangeness, surprise, or inconvenience while reviewing a video or cartoon result, even if the user does not explicitly input feedback. That is, the sensor unit 110 can improve the accuracy of factual relationships and the ability to reproduce reality by collecting objective and contextual on-site information in addition to subjective statements input by the user.

[0060] Communications Department 120

[0061] According to the present invention, the communication unit 120 can transmit and receive data between the user terminal 100 and the server device 200. In the present invention, the communication unit 120 functions as a medium for exchanging data required for each step, such as extracting factual information, generating a scenario, visualizing a video or cartoon, and reflecting user feedback, with the server device 200. Specifically, first event individual information 2-1 and second event individual information 2-2 input by the user in the form of voice, text, image, or document are collected via the sensor unit 110 and then transmitted to the server device 200 via the communication unit 120. The server device 200 can perform advanced computations, such as natural language processing, individual name recognition, structuring factual information, and question generation, based on the first event individual information 2-1 and second event individual information 2-2 transmitted from the communication unit 120. In addition, the communication unit 120 can transmit feedback information, such as error detection and correction requests, input by the user regarding the video or cartoon they have viewed to the server device 200.

[0062] Processor section 130

[0063] According to the present invention, the processor unit 130 located in the user terminal 100 mainly performs pre-processing work on user input and can be responsible for lightweight calculation or pre-filtering functions at a stage before advanced calculation processing in the server device 200. For example, when a user describes the details of an incident through voice via the user terminal 100, the processor unit 130 can perform a speech-to-text voice recognition function that converts voice data received through the sensor unit 110 into text in real time, or can detect specific keywords or sentence patterns and tag them as analysis targets.

[0064] In addition, the processor unit 130 can analyze images and documents captured in the user terminal 100 and perform file format conversion, size adjustment, simple optical character recognition, etc., and transmit the refined data to the server device 200, thereby reducing the network load and improving the processing efficiency of the server. In addition, the processor unit 130 can temporarily store and optimize rendering of text summaries and scenario results received from the server device 200, and provide smooth and stable visual output to the user through the output unit 150. In the present invention, the processor unit 130 delegates many main calculations to the server device 200, but can also be designed to function as a lightweight processing component to improve input responsiveness and real-time user experience.

[0065] memory section 140

[0066] According to the present invention, the memory unit 140 is a component that temporarily stores or caches various data generated in the user terminal 100, and can contribute to improving the computational efficiency and responsiveness of the user interface of the fact-based creation creation system 10. For example, the memory unit 140 may store pre-processed data, such as voice, text, image, or document data input by a user, collected via the sensor unit 110 and before being transmitted to the server device 200. In addition, the memory unit 140 may temporarily store a part or all of the first case individual information 2-1 and the second case individual information 2-2, which are information directly related to the facts of the user, among the analysis results received from the server device 200.

[0067] According to the present invention, the memory unit 140 can also store dialogue session information, details of revision requests, recently viewed video clips, and last confirmed factual items necessary to maintain context between the user's input and the fact-based responses of the creative creation system 10. This allows the user to naturally continue the interaction from the point where the transfer process was interrupted. Meanwhile, large amounts of external reference information, such as the litigation and case type database 3, are stored in the server device 200, and the memory unit 140 can partially store only a temporary summary of the data or citation reference results.

[0068] Output section 150

[0069] According to the present invention, the output unit 150 is a component responsible for visually or audibly outputting information generated via the user terminal 100, and performs a function of intuitively conveying the results of analysis and visualization of factual relationships to the user. The output unit 150 may include, for example, a display device, a speaker, a haptic feedback device, etc., and is used to provide the user with a summary of factual relationships, a scenario script, 2D / 3D-based video snippets, a cartoon-style visualization image, etc., which are the results of implementing the present invention.

[0070] According to the present invention, the output unit 150 renders the resultant product processed and transmitted by the server device 200 so that it can be visually displayed within the user terminal, and if the resultant product is video or audio information, it can output audio such as a voice narration or dialogue between characters through a speaker. For example, a fragmented accident video generated by the creation generation tool 133, 230 or a cartoon resultant product of a contract negotiation scene generated by the creation generation tool 133, 230 is played back through the output unit 150, and the user can recognize the difference from the actual facts while watching it and provide feedback.

[0071] The output unit 150 also plays an important role in the user feedback process, helping the user identify errors or compare them through operations such as pausing, repeating, zooming in, and slowing down playback of a specific scene during playback of a video or cartoon, and providing an interactive UI related to the verification and correction tools 134, 240 according to user actions such as clicking on a specific section or entering a comment during playback.

[0072] Server device 200

[0073] Fact Extraction Tools 131,210

[0074] According to the present invention, the factual information extraction tool 131, 210 can organize the factual information of a case as structured information from various types of data input by a user, such as voice, text, document, image, etc. The factual information extraction tool 131, 210 is mainly located in the server device 200, but some simple calculations or user response-based processing may be performed within the user terminal 100.

[0075] According to the present invention, the factual information extraction tool 131, 210 can extract basic information constituting a case, such as the time of occurrence of the case, location, related persons, type of case, details of damage, main actions and statements, from the statement or materials provided by the user, and can generate first case individual information 2-1 and second case individual information 2-2 based on the extracted information. Thereafter, the factual information extraction tool 131, 210 can store the first case individual information 2-1 and the second case individual information 2-2 in the factual information database 2. In addition, the factual information extraction tool 131, 210 can perform the function of searching for information omitted from the user's statement or items that may become legal issues in connection with the lawsuit and case type database 3, and based on the searched information, generate supplementary questions in the supplementary question generation step (S113-2).

[0076] Scenario generation tool 132,220

[0077] According to the present invention, the scenario generation tool 132, 220 can generate text content in the form of a visualized scenario based on the case information structured by the fact extraction tool 131, 210. The scenario generation tool 132, 220 can refer to the first case individual information 2-1 and the second case individual information 2-2 to determine whether necessary information has been omitted based on the case type and legal context, and can prompt the user for additional input or questions to supplement the information (S200). Then, based on the acquired case information, a summary of the facts (S210) is generated that condenses and organizes the flow of the case, and the summary of the facts is provided in a format that can be reviewed by the user or an expert in text format. Furthermore, the scenario generation tool 132, 220 can generate a text-based scenario script (S220) divided into stages, such as the occurrence of the incident, the damage situation, and follow-up measures, based on the summary of the facts.

[0078] Creative creation tools133,230

[0079] According to the present invention, the creative product generation tools 133, 230 can receive a text-based scenario script (such as that shown in FIG. 4a) generated by the scenario generation tools 132, 220 and perform the function of constructing a visual image of the facts of the incident in scene units. The creative product generation tools 133, 230 extract individual elements, such as the location, time period, people, vehicles, and weather, from the scenario and generate fragmented images in the form of 2D or 3D graphics, virtual avatars, or simulation animations by linking with an AI-based image generation engine such as SORA or VEO2. The creative product generation tools 133, 230 also realistically reproduce the facts of the scenario through voice synthesis, character dialogue implementation, and sound effect insertion. The generated images are subject to user review and correction feedback and can then be used as an iterative generation structure. Therefore, the creative product generation tools 133, 230 can provide the effect of enhancing user understanding by actually visually implementing incident information in the present invention.

[0080] Verification and correction tools 134,240

[0081] According to the present invention, the verification and correction tools 134, 240 can verify and correct the video or cartoon generated by the creative creation tools 133, 230. Specifically, after viewing or browsing the visualization result generated by the fact-based creative creation system 10, a user can point out errors or fill in missing information in parts that are not in accordance with the actual facts, and the verification and correction tools 134, 240 can reflect the user's feedback in a text-based scenario script such as that shown in Figure 4a. The creative creation tools 133, 230 can then regenerate the video or cartoon based on the corrected text-based scenario script, allowing the user to repeatedly review and correct it.

[0082] This cyclical structure of "primary generation → user review → revision request → reflection of text-based scenario script → regeneration of video or cartoon" can act as the operating mechanism for user feedback in the present invention. Through this cyclical process, the present invention systematically reflects user experience-based feedback within the fact-based creation creation system 10, thereby gradually reducing errors in the initial facts and ultimately achieving highly accurate and reliable visualization results. The present invention dramatically reduces the possibility of factual errors through this iterative improvement structure based on interaction between the verification and revision tools 134, 240 and users, and provides the effect of enabling objective and specific factual verification.

[0083] Creative creation tools133,230

[0084] According to the present invention, the creative work generation tools 133 and 230 are components that convert facts into static view content, i.e., cartoon-style scene images, based on the text-based scenario script generated by the scenario generation tools 132 and 220. The creative work generation tools 133 and 230 visualize the flow of events in a continuous scene format, reflecting the lines, emotions, actions, and background of the characters, providing a representation method suitable for narrative information transmission that focuses on dialogue rather than audio. In particular, in case types where the flow of dialogue and statements between characters is more important than spatial context, such as contract disputes, medical explanations, and insurance consultations, the creative work generation tools 133 and 230 function as a visualization means that can enhance the user's understanding and clearly convey the facts. The generated cartoon is provided to the user via the output unit 150, and it can be repeatedly revised and supplemented based on feedback.

[0085] FIG. 2a is a diagram illustrating a case individual information extraction process of a factual relationship extraction tool according to an embodiment of the present invention.

[0086] Fact Extraction Tools 131,210

[0087] Pre-processing step (S100)

[0088] According to one embodiment of the present invention, the fact extraction tool 131, 210 can analyze user-provided voice and text data using natural language processing (NLP) techniques to structure information such as time, place, people, an outline of the incident, and issues. For example, a user can provide voice data such as, "A collision with an SUV occurred at around 3:00 PM on January 2, 2025, in XX-dong, XX-gu, Seoul. I was waiting at a traffic light when the SUV suddenly hit me from behind." The fact extraction tool 131, 210 can perform preprocessing on the voice data provided by the user.

[0089] Speech-to-Text (STT) processing step (S101)

[0090] According to an embodiment of the present invention, the fact extraction tool 131, 210 can perform speech-to-text (STT) processing on voice data provided by a user. Specifically, when a user describes an incident through voice, a process of converting the voice into text is required. To this end, the fact extraction tool 131, 210 can utilize an STT engine, such as Google Speech API or Whisper (OpenAI), to filter unnecessary utterances such as "yeah," "um," and "so." For example, if a user says, "Um... so, on January 2, 2025, at around 3:00 PM, there was an accident in XX-dong, XX-gu, Seoul. An SUV suddenly crashed into my car," the fact extraction tool 131, 210 can convert this to "On January 2, 2025, at around 3:00 PM, there was an accident in XX-dong, XX-gu, Seoul. An SUV suddenly crashed into my car." using the STT engine.

[0091] Sentence Segmentation Step (S102)

[0092] According to an embodiment of the present invention, the factual relationship extraction tool 131, 210 can perform sentence refinement and segmentation. Specifically, the factual relationship extraction tool 131, 210 can divide a user's input into semantic units and organize the sentence to more effectively extract incident information. To this end, the factual relationship extraction tool 131, 210 can segment sentences based on punctuation marks such as ".", "?", and "!", and can segment sentences based on meaning using conjunctions such as "and" and "therefore." For example, the factual relationship extraction tool 131, 210 can perform analysis by dividing the sentence "The SUV suddenly crashed, and there was damage" into "The SUV suddenly crashed" and "There was damage."

[0093] Stopword Filtering & Keyword Extraction Step (S103)

[0094] According to an embodiment of the present invention, the factual relationship extraction tool 131, 210 may perform stopword filtering and keyword extraction. The factual relationship extraction tool 131, 210 may remove expressions unrelated to the incident, leaving only meaningful information. The factual relationship extraction tool 131, 210 may remove particles such as "wa," "ga," and "wo," unnecessary adverbs such as "kurai," "mashita," "ah," "um," and "dakara," and exclamations such as "kurai," "mashita," "ah," "um," and general verbs such as "had" and "was" depending on the purpose of analysis. For example, the factual relationship extraction tool 131, 210 may remove unnecessary expressions from a sentence such as "There was an accident in XX-dong, XX-gu, Seoul, around 3:00 PM on January 2, 2025. An SUV suddenly crashed into my car," leaving only important words, such as "Accident in XX-dong, XX-gu, Seoul, at 3:00 PM on January 2, 2025. SUV suddenly crashed into my car."

[0095] Normalization & Structuring Step (S104)

[0096] According to an embodiment of the present invention, the fact extraction tool 131, 210 may perform normalization and data conversion (normalization and structuring). Specifically, the fact extraction tool 131, 210 may convert extracted data into a standardized format. For example, the fact extraction tool 131, 210 may convert date information such as "January 2, 2025, 3:00 PM" into a standard time format such as "2025-01-02 15:00," and standardize location information such as "XX-dong, XX-gu, Seoul" into "XX-dong, XX-gu, Seoul." Furthermore, the fact extraction tool 131, 210 may convert an expression such as "an SUV suddenly crashes" into a fixed incident type such as "rear-end collision." The standardized data may be stored in JSON format or a table format and used as basic data for subsequent AI image generation and incident analysis.

[0097] Important information extraction stage (S110)

[0098] Named Entity Recognition (NER) Application Stage (S111)

[0099] According to one embodiment of the present invention, the entity recognition (NER) technique is a technology that extracts key information related to a case from a user's statement and classifies it by entity type. In this embodiment, the process by which the fact extraction tool 131, 210 classifies case information using a deep learning-based NER model such as BERT-NER, spaCy, or Stanza will be described below.

[0100] According to an embodiment of the present invention, when a user describes an incident, the fact extraction tool 131, 210 can analyze the statement and identify information. For example, if the user says, "A collision with an SUV occurred at around 3 PM on January 2, 2025, in XX-dong, XX-gu, Seoul. While waiting at a traffic light, the SUV suddenly hit me from behind. My car was severely crushed and my neck hurt a lot," the fact extraction tool 131, 210 can preprocess this and refine it into a form such as "A rear-end collision with an SUV occurred at 3 PM on January 2, 2025, in XX-dong, XX-gu, Seoul. While waiting at a traffic light, the SUV hit me from behind. My car was severely damaged and I had a lot of neck pain."

[0101] Extraction of incident information (S112)

[0102] First incident individual information and second incident individual information

[0103] FIG. 1b is a diagram illustrating first and second case individual information of a factual database according to an embodiment of the present invention.

[0104] As shown in FIG. 1b, the factual database 2 includes first incident individual information 2-1 and second incident individual information 2-2.

[0105] In the present invention, the first incident individual information 2-1 and the second incident individual information 2-2 are concepts that provide different levels of information in the process of analyzing and recording an accident. The first incident individual information 2-1 is basic information that shows a general overview of the accident, and plays a role in defining when, where, and what type the accident occurred. On the other hand, the second incident individual information 2-2 is data that analyzes the accident progression and impact in more detail than the first incident individual information 2-1, and includes detailed progress status since the accident occurred and whether or not there was secondary damage.

[0106] Specifically, the first incident individual information 2-1 basically includes information such as the time of the accident, location, and incident type. For example, if the accident time is "3:00 PM on February 15, 2025," the accident location is "Dogok-dong Intersection, Gangnam-gu, Seoul," and the incident type is "rear-end collision," this corresponds to the first incident individual information 2-1. The first incident individual information 2-1 may also include information that the vehicles involved in the accident were an SUV and a passenger car, and that the victim vehicle suffered rear bumper damage. In other words, the first incident individual information 2-1 may be used to grasp the overall outline of the accident and to classify the incident. However, the first incident individual information 2-1 alone may lack detailed information such as how the accident progressed, whether additional damage was caused, and what measures were taken after the accident.

[0107] According to the present invention, the second incident individual information 2-2 refers to information on the progress of the incident after it occurred. Specifically, the second incident individual information 2-2 may include at least one of information on the victim's physical reaction, information on the victim's emotional expression, information on the assailant's initial reaction, information on third-party witness reaction, information on the occurrence of additional damage, information on whether the accident escalated, and information on on-site response measures. Specifically, the victim's physical reaction information may include the victim's actions, such as saying "my neck hurts" or getting out of the car and hugging themselves, immediately after the accident. The emotional expression information may include information indicating the victim's emotional state, such as the victim being surprised or angry, yelling loudly, or crying. The assailant's initial reaction information may include the assailant's actions, such as getting out of the car and saying "I'm sorry" immediately after the accident or attempting to leave the scene. The third-party witness reaction information may include the actions of nearby witnesses who reported the accident or helped rescue the victim.

[0108] In addition, the additional damage information may include additional material damage such as when the victim vehicle was pushed into a secondary collision with another vehicle or structure, or when luggage was thrown out and spread on the road, and the information on whether the accident escalated may include information related to the scale of the accident such as risk factors such as fuel leakage, fire risk, and electrical short circuit, and whether airbags were deployed. The on-site response measures information may include information related to administrative and medical measures such as the arrival time of 119, police, insurance companies, etc., whether first aid was administered, whether breathalyzer tests were conducted by the police, and whether the victim was transported to a hospital.

[0109] The second incident individual information 2-2 can also reflect the flow of information unfolding chronologically after the incident. For example, time information such as "the victim's vehicle is pushed forward immediately after the collision," "the assailant gets out of the vehicle 10 seconds later," and "the police arrive 2 minutes later" are key elements in creating a scenario for the progression of an incident. Such information can be directly reflected in the scene composition and character behavior of creative works generated as videos or cartoons, and can be supplemented or modified based on user feedback.

[0110] Furthermore, the second incident individual information 2-2 may not simply record the accident as a "rear-end collision," but may include information organized by time about how the accident unfolded and what the results were. For example, the second incident individual information 2-2 may include information such as, "At 2:45 PM on February 15, 2025, an SUV rear-ended a passenger car waiting at a traffic light, and at 2:47 PM, the impact caused the passenger car to collide with a guardrail, resulting in a secondary accident." The second incident individual information 2-2 may also include the time when the driver of the at fault vehicle got out of the vehicle and attempted to communicate with the driver of the victim vehicle immediately after the accident, the time when the police and insurance company arrived at the scene, whether the victim was taken to a hospital, etc.

[0111] According to the present invention, the second incident information 2-2 goes beyond a simple accident summary and records subsequent events after the accident, enabling more accurate analysis. For example, if a rear-end collision does not end as a simple collision but records whether a secondary accident occurred due to the impact, this can be important evidence in legal disputes and insurance reviews. If the victim's vehicle collided with a guardrail and the trunk was deformed, the damage would be greater than in a simple rear-end collision, which could change the degree of fault and compensation liability.

[0112] Utilizing this information system allows for more precise accident analysis and provides a more objective basis for legal and insurance decisions. For example, if the extent of rear bumper damage is a matter of debate, the second incident information 2-2 can be used to analyze the strength of the impact and the location of the damage at the time of the accident. By comparing it with existing similar accident cases, it is also possible to identify the damage patterns that typically occur when an SUV rear-ends a stationary passenger car. This allows for a more precise analysis of whether only the bumper was damaged in a rear-end collision, whether the trunk was deformed, or whether the impact reached the interior of the vehicle, and can be used as important information for insurance companies and courts to evaluate the scale of the accident.

[0113] Litigation and Case Type Database3

[0114] 1b, the lawsuit and case type database 3 includes case type information 3-1. Specifically, the case type information 3-1 may include case occurrence condition information 3-1-1.

[0115] Incident occurrence condition information 3-1-1

[0116] According to the present invention, the incident occurrence condition information 3-1-1 is information that structures and stores the temporal, spatial, and environmental conditions that led to the occurrence of a past incident. The incident occurrence condition information 3-1-1 encompasses the external context that existed before the unfolding of the incident and is used to search for missing information through comparison with incident individual information or to guide scenario generation based on similar conditions. Specifically, the incident occurrence condition information 3-1-1 may include the date and time of the incident, the location of the incident, the environmental conditions at the time of the incident, the initial state of the incident participants, and situational context information immediately before the incident. The date and time of the incident may include information such as the year, month, date, day of the week, time period (e.g., morning / afternoon, regular time / night), or seasonal information, and may be stored in the form of, for example, "Sunday, May 12, 2024, around 9:00 AM." The location of the incident may be expressed in detailed location categories such as road name and address, area type (urban / suburban), intersection, parking lot, alley, etc., and environmental condition information includes natural background elements at the time of the incident such as weather (sunny, rainy, snowy), illumination (daytime / nighttime), and temperature. The initial state of participants involved in the incident may also be included in incident occurrence condition information 3-1-1, which may indicate the type of vehicle, whether it was parked or moving, driving speed, number of drivers, or whether there were passengers. Additionally, the context of the situation immediately before the accident may be stored as natural language-based situation information, such as "the vehicle was stopped at a traffic light" or "the vehicle was turning right in a narrow alley."

[0117] Case summary information 3-1-2

[0118] According to the present invention, the case summary information 3-1-2 may include information describing the background and structural framework of the case as a sub-item of the case-specific case information 3-1 contained in the lawsuit and case-specific database 3. Specifically, the case summary information 3-1-2 is a component that stores the direct causes and components that actually led to the occurrence of the case, i.e., the roles and relationships of the people involved, the initial circumstances of the case, and the causal factors that led to the occurrence of the case, in a structured manner. More specifically, the case summary information 3-1-2 may first include character composition information, which is used to clarify the roles and relationships of the leaders involved in the case. For example, legal and practical relationships can be explained not only by primary classification such as "Driver A (victim)" and "Driver B (perpetrator)," but also by case type such as "lessor, tenant, intermediary" or "doctor, patient, guardian." This character structure serves as the basis for subsequent developments and responsibility determinations.

[0119] According to one embodiment of the present invention, the incident summary information 3-1-2 may include information about the initial situation before or immediately before the incident began. Specifically, the initial situation before or immediately before the incident refers to a static state immediately before the incident or a situation where the seeds of a problem were apparent, such as "the situation in which the offending vehicle closely followed the vehicle in front," "the situation in which the lessee was reviewing the contract," or "the state before the patient took medication." Furthermore, the direct cause information is defined as the trigger, triggering factor, or initial problematic behavior that led to the incident, such as "the vehicle behind failed to recognize a stop signal and rear-ended the vehicle," "the lessor verbally changed the terms of the contract," or "the doctor failed to thoroughly check the vehicle's past medical history."

[0120] According to one embodiment of the present invention, the incident summary information 3-1-2 is more intrinsic and causal than the incident occurrence condition information 3-1-1, and acts as an element that explains the essential framework and cause of the incident beyond simple time, place, and environment. Therefore, when the initial situation or problem occurrence structure of the first incident individual information 2-1 and the second incident individual information 2-2 is unclear, the scenario generation tool 132, 220 can complement the missing prerequisites or generate questions by comparing them with the incident summary information.

[0121] Incident development information 3-1-3

[0122] According to the present invention, the case development information 3-1-3 is information that structures and stores the specific flow of development of a situation in chronological order since the occurrence of an incident as a sub-item of the case information by type 3-1 contained in the lawsuit and case type database 3. Specifically, the case development information 3-1-3 provides a substantial foundation for constructing a scenario and visually reproducing it by recording in detail the context of individual cases and the process of development over time, focusing on the fact that an incident is not the result of a single point in time but is made up of an accumulation of continuous actions and reactions.

[0123] According to one embodiment of the present invention, the incident development information 3-1-3 may first include information on the victim's reaction. This information is a chronological record of the victim's behavioral and psychological reactions immediately after the accident, such as "My neck hurts," "I got out of the car to check the scene," and "My face turned pale with fear." Second, the incident development information 3-1-3 includes information on the perpetrator's behavior, such as "sudden braking," "getting out of the vehicle to apologize," and "trying to leave the scene before the police arrive," and is structured to include the time and context of each action. Third, the incident development information 3-1-3 may include whether a secondary accident or a chain reaction occurred. For example, it may record whether or not there was additional incident escalation, such as "After a rear-end collision, the victim's vehicle was pushed forward and collided again with the vehicle in front," "The vehicle hit a guardrail," or "The vehicle was stopped in the center of the road, creating a risk of secondary damage." Fourth, incident development information 3-1-3 may be information that organizes the actual response procedures and flow along a timeline, such as the time when the police, insurance companies, 119, and other agencies arrived at the scene after receiving the report, whether first aid was administered, breathalyzer tests, and dashcam checks, etc., regarding the process of rescue requests and on-site response.

[0124] Incident result information 3-1-4

[0125] According to the present invention, case outcome information 3-1-4 is a sub-item of case information by type 3-1 contained in the lawsuit and case type database 3, and includes information constituting the outcome of the case, such as the results of administrative and legal procedures actually carried out after the incident occurred, details of compensation for damages, and diagnosis results. Specifically, case outcome information 3-1-4 is not just the history of the accident, but is official follow-up action data that reflects the resulting judgment and records, and will serve as an important reference for future comparison with user input or for deriving predicted questions or scenario completion structures.

[0126] According to an embodiment of the present invention, the case result information 3-1-4 may include a legal judgment result, which refers to a judgment made through official procedures such as the results of a police investigation, whether or not the prosecution filed an indictment, the content of a civil lawsuit judgment, a court's determination of the degree of fault and the scope of liability for damages, and may be structured and stored in a form such as "determination that the at-fault vehicle was 100% at fault" or "determination that civil damages amount to 9 million won."

[0127] According to one embodiment of the present invention, the case result information 3-1-4 includes administrative decision records directly linked to the accident, such as the insurance company's review result, accident classification, whether liability insurance was processed, whether auto repair costs were paid, whether actual medical expenses were approved, etc. For example, "automobile total loss treatment," "hospitalization costs paid in full," etc. may also be used as criteria for determining the scope of compensation with the case information entered by the user.

[0128] According to one embodiment of the present invention, the medical diagnosis results and disability assessment information in the case result information 3-1-4 may be stored as structured data of the victim's official medical treatment results and injury level at the hospital in the form of "cervical sprain diagnosis 2 weeks later," "recommendation for future outpatient treatment," "grade 10 disability certification," etc. In other words, the case result information 3-1-4 may be used to determine the scale of damage, follow-up insurance procedures, and estimate compensation standards.

[0129] According to one embodiment of the present invention, the case outcome information 3-1-4 may also include whether or not an agreement has been reached between the parties or whether or not the dispute is continuing. For example, cases such as "an agreement was drawn up on-site" and "civil litigation proceeded after an agreement failed" may be used to determine the persistence and seriousness of the dispute.

[0130] Stakeholder Speech Information 3-1-5

[0131] According to the present invention, the stakeholder speech information 3-1-5 refers to linguistic information such as remarks, lines, emotional expressions, questions, or statements actually made by stakeholders involved in a case as a sub-item of the case information by type 3-1 contained in the lawsuit and case type database 3. Specifically, the stakeholder speech information 3-1-5 is used to add a sense of realism and emotional context to scenarios and visual creations by structuring and storing linguistic data extracted from on-site conversations, statements, emotional reactions, etc. made at the time of or immediately after the incident.

[0132] According to an embodiment of the present invention, the stakeholder utterance information 3-1-5 may be classified into victim, perpetrator, witness, police, insurance company employee, lawyer, medical professional, etc. by subject, and the time of utterance and emotional state may be recorded for each subject. For example, if a victim says, "My neck hurts so much. I hit you suddenly!" immediately after the accident, or if the perpetrator says, "I'm sorry. I braked too late...", the utterance may be stored in the lawsuit and case type database 3 along with the time, emotion, and speaker's role.

[0133] According to one embodiment of the present invention, the stakeholder speech information 3-1-5 may be not only a text quotation but also annotated with linguistic context, tone of voice, and emotional bases (fear, anger, confusion, sadness, etc.). That is, when the creation creation tool 133, 230 applies text-to-speech technology, it may be used to adjust the strength of the tone of the dialogue, distinguish between screams and calm speech, and express tension. In the case of a cartoon creation tool, it may also be used as a reference for reflecting speech bubbles, facial expressions, and voice volume.

[0134] In the present invention, the second case individual information 2-2 performs a role beyond simply describing facts and is used to precisely analyze the legal context of a case through linkage with the lawsuit and case type database 3. The factual relationship extraction tools 131, 210 analyze the user's statement to structure the first case individual information 2-1 and the second case individual information 2-2, and can search for similar cases by comparing and contrasting the second case individual information 2-2 with the lawsuit and case type database 3. For example, if the factual relationship extraction tools 131, 210 structure the statement "an SUV collided from behind," the information is classified as belonging to the "rear-end collision" category.

[0135] Thereafter, the fact extraction tools 131, 210 can search for precedents and cases of past rear-end collisions from the litigation and case type database 3 and confirm that information such as whether there was a secondary collision, the possibility of vehicle avoidance, the damage pattern of the victim vehicle, and the criteria for determining the degree of fault was frequently discussed. As a result, the fact extraction tools 131, 210 can determine that the user's input does not contain the relevant information and generate questions such as "Did the victim vehicle also collide with the guardrail?" and "Did additional damage occur immediately after the accident?" to guide the user's response.

[0136] Unlike existing natural language processing technologies that simply summarize and classify text, the fact extraction tools 131, 210 are technologically differentiated in that they can extract case-specific information that requires legal judgment by case type and actively search for and inquire about missing information. In addition, the fact extraction tools 131, 210 hierarchically structure the first case-specific information 2-1 and the second case-specific information 2-2 so that user statements can be converted into a legally interpretable form.

[0137] According to one embodiment of the present invention, the factual relationship extraction tool 131, 210 may extract first incident individual information 2-1 related to an incident using a pre-trained individual name recognition model. The factual relationship extraction tool 131, 210 may use an NER model to tag each word or phrase as first incident individual information 2-1, thereby classifying time individual information 2-1-1, location individual information 2-1-2, vehicle individual information 2-1-3, accident type individual information 2-1-4, victim individual information 2-1-5, situation individual information 2-1-6, person individual information 2-1-7, and behavior individual information 2-1-8. For example, the factual relationship extraction tool 131, 210 may tag "January 2, 2025, 3:00 PM" as time individual information 2-1-1 and tag "Dogok-dong, Gangnam-gu, Seoul" as location individual information 2-1-2. In addition, the factual relationship extraction tools 131 and 210 can classify "SUV" as individual information of the offending vehicle, and "waiting at a traffic light" and "SUV hit from behind" as individual information of the situation at the time of the accident 2-1-6. In addition, the factual relationship extraction tools 131 and 210 can extract "vehicle severely damaged" and "occurrence of neck pain" as individual information of the victim 2-1-5 in relation to the scale of the damage.

[0138] According to an embodiment of the present invention, the factual relationship extraction tool 131, 210 can perform a standardization process on the extracted individual information. For example, the factual relationship extraction tool 131, 210 can convert date information from "January 2, 2025, 3:00 PM" to a standard time format such as "2025-01-02 15:00," and can organize vehicle information from "SUV vehicle" to "offending vehicle: SUV." Information on the scale of damage can also be converted and stored by converting "severe vehicle damage" to "serious vehicle damage" and "occurrence of neck pain" to "minor injury (additional medical treatment required)."

[0139] FIG. 2b is a diagram illustrating tagging of a factual relationship extraction tool according to an embodiment of the present invention.

[0140] According to one embodiment of the present invention, the fact extraction tool 131, 210 can perform tagging as shown in FIG. 2b.

[0141] FIG. 2c is a diagram illustrating that a fact extraction tool according to an embodiment of the present invention stores case individual information in a standardized format.

[0142] According to one embodiment of the present invention, the fact extraction tool 131, 210 can store the case individual information in a standardized format as shown in FIG. 2c.

[0143] Structured data storage and additional query generation step (S113)

[0144] Structured Data Storage Stage (S113-1)

[0145] According to one embodiment of the present invention, if the factual information extraction tool 131, 210 recognizes the first incident individual information using NER, it can store the first incident individual information in a standardized format based on the first incident individual information. For example, as shown in Figure 2b, the factual information extraction tool 131, 210 can extract information such as time individual information, place individual information, vehicle individual information, accident type individual information, behavior individual information at the time of the accident, and damage scale individual information, organize the information, and store it in a database (DB) by incident type in JSON format as shown in Figure 2c.

[0146] Additional question generation stage (S113-2)

[0147] According to an embodiment of the present invention, the fact extraction tool 131, 210 may search for information that has been omitted since the structured data was stored and generate additional questions to supplement the information in the additional question generation step (S113-2). At this time, the fact extraction tool 131, 210 may utilize natural language understanding (NLU) and text generation models to grasp and supplement incomplete information in the user's accident statement.

[0148] Example of the leaked first incident individual information search

[0149] According to one embodiment of the present invention, the factual relationship extraction tool 131, 210 can use a BERT-based NER model to determine whether required first incident individual information is included in structured data and search for missing first incident individual information. Specifically, the factual relationship extraction tool 131, 210 analyzes a user's statement to recognize required first incident individual information, such as time individual information, location individual information, vehicle individual information, accident type individual information, and damage scale individual information, and can tag any missing items as "no information." For example, if damage scale individual information is leaked, the factual relationship extraction tool 131, 210 sets this as "no damage scale information." Then, the factual relationship extraction tool 131, 210 can detect that damage scale individual information has been leaked using a GPT-based question generation model and generate an appropriate question.

[0150] Additional Question Generation Embodiment

[0151] According to an embodiment of the present invention, the factual relationship extraction tool 131, 210 can generate natural questions to supplement missing information by utilizing a GPT-based question generation model. That is, the factual relationship extraction tool 131, 210 can generate questions that fit the context by utilizing existing natural language processing models (T5, BART, etc.) and large-scale language models such as GPT-4. First, the factual relationship extraction tool 131, 210 can analyze missing individual information searched for by the BERT model and determine individual information that requires additional questions.

[0152] FIG. 3a is a diagram illustrating a text-based scenario script generation step of the scenario generation tool according to the present invention.

[0153] Case information comparison and second case individual information supplementation step (S200)

[0154] According to one embodiment of the present invention, the scenario generation tool 132, 220 can compare the first case individual information and the second case individual information stored in the structured data storage step (S113-1) with the lawsuit and case type database 3. Here, the lawsuit and case type database 3 is a structured data set that can be used to supplement missing information by comparing it with existing cases at the time of an accident. Specifically, when the scenario generation tool 132, 220 lacks second case individual information including detailed information on the circumstances at the time of the accident, such as whether the rear bumper was damaged, the strength of the vehicle impact, and the road environment, the scenario generation tool 132, 220 can supplement the second case individual information by using the lawsuit and case type database 3.

[0155] Litigation and Case Type Database3

[0156] According to an embodiment of the present invention, the scenario generation tool 132, 220 can analyze information on the time and location of an accident by utilizing the lawsuit and incident type database 3. The lawsuit and incident type database 3 may include various incident types and their associated characteristics and legal and insurance issues.

[0157] According to one embodiment of the present invention, the lawsuit and case type database 3 may include accident characteristics by vehicle type. For example, if an SUV rear-ends a compact car, the SUV is likely to suffer only minor damage, but the compact car is likely to suffer a dent in its rear bumper and deformation of its trunk. On the other hand, if a large truck rear-ends a parked passenger car, the truck is likely to suffer little damage, but the passenger car's trunk is likely to suffer a severe dent and even deformation of its internal structure.

[0158] According to an embodiment of the present invention, the lawsuit and case type database 3 may include damage ranges for each case type. For example, in rear-end collisions, damage to the rear bumper is the most common, and if the impact is strong, even the trunk may be deformed. In side collisions, the vehicle's side panels and doors are likely to be dented, and the side airbags are likely to be deployed. In head-on collisions, the front bumper, hood, and headlights are likely to be damaged, and if the impact is strong, even the radiator and engine compartment may be deformed.

[0159] According to an embodiment of the present invention, the lawsuit and case type database 3 may include damage patterns according to the circumstances of the accident. For example, if a vehicle stopped at a traffic light is hit from behind, the vehicle may be pushed and hit the vehicle in front, resulting in a secondary accident. If a rear-end collision occurs while a vehicle is stopped on a highway, the impact may cause the vehicle to crash into the median strip or lead to a chain reaction accident.

[0160] According to one embodiment of the present invention, the lawsuit and case type database 3 may include driver and passenger injury types. For example, in a low-speed rear-end collision, minor neck pain such as cervical sprains is common, while in a high-speed collision, neck bone and disc herniation injuries are more likely to occur. In a side collision, passengers often suffer shoulder and rib fractures, and in a frontal collision, facial injuries may occur due to airbag deployment.

[0161] According to one embodiment of the present invention, the lawsuit and case type database 3 may also include elements that may become legal issues. For example, in the case of a rear-end collision that occurs after a sudden stop, the issue of whether the sudden stop of the vehicle in front was justified may become an issue. In the case of a collision accident during a lane change, important factors may include whether the vehicle changing lanes turned on its turn signal and whether it changed lanes after maintaining a sufficient distance.

[0162] Embodiment of second case individual information supplement

[0163] According to one embodiment of the present invention, the scenario generation tool 132, 220 can perform a similarity analysis with type-specific case information 3-1 stored in a database 3 for lawsuits and cases based on case individual information extracted from the user, i.e., the first case individual information 2-1 and the second case individual information 2-2.

[0164] According to one embodiment of the present invention, the scenario generation tool 132, 220 can express the type-specific case information 3-1 in the form of a multidimensional vector based on predefined attribute items (feature set) of incident occurrence condition information 3-1-1, incident summary information 3-1-2, incident development information 3-1-3, incident outcome information 3-1-4, and stakeholder speech information 3-1-5 in order to perform similarity analysis. As a result, the scenario generation tool 132, 220 can perform comparative quantification of the first incident individual information 2-1, the second incident individual information 2-2, and the type-specific case information 3-1 by applying Euclidean distance, cosine similarity, or a deep learning-based embedding similarity method to the type-specific case information 3-1.

[0165] According to one embodiment of the present invention, the scenario generation tool 132, 220 performs basic matching on the time individual information 2-1-1 and place individual information 2-1-2 corresponding to the first incident individual information 2-1, and then detects inconsistencies or empty fields based on whether specific information items are missing, such as the detailed progress status since the accident occurred and whether there was secondary damage corresponding to the second incident individual information 2-2. For example, if the information entered by the user only specifies the incident type of "rear-end collision," but similar past cases require information such as "whether a secondary collision occurred," "the driver's injury status," and "whether on-site rescue was requested," the scenario generation tool 132, 220 can detect that the information is missing and generate additional questions.

[0166] According to one embodiment of the present invention, in the case of a rear-end collision, the first incident individual information includes basic information such as the time of the accident (3:00 PM on February 15, 2025), location (Dogok-dong intersection, Gangnam-gu, Seoul), incident type (rear-end collision), offending vehicle (SUV), and victim vehicle (passenger car). However, this information alone may be insufficient to clearly grasp the overall progression of the accident and the extent of the damage, and may be insufficient to generate a summary of the facts. Therefore, the scenario generation tool 132, 220 can check whether important information is included in a typical rear-end collision by comparing it with a database by lawsuit type and incident type, and if any information is missing, it can generate additional questions to supplement the missing information.

[0167] According to one embodiment of the present invention, in a rear-end collision, it may be confirmed that the information on whether the victim vehicle was pushed by the impact and an additional collision (secondary accident) occurred is not recorded. Furthermore, the specific extent of damage to the victim vehicle, such as whether only the rear bumper was damaged, whether the trunk was deformed, or whether the rear window was broken, may be insufficient. Furthermore, the victim vehicle driver's injuries and hospital visit records may be omitted. Because such information is essential for constructing a scenario, the scenario generation tool 132, 220 can generate additional questions for this purpose.

[0168] For example, the scenario generation tools 132 and 220 can generate questions such as, "After the SUV collided from behind, was the vehicle pushed and collided with another object (guardrail, car in front, etc.)?", "In addition to the rear bumper of the passenger car, was the trunk deformed or the rear window broken?", "Did you experience neck or back pain due to the impact of the collision? Did you visit a hospital?" If the user inputs additional information through such questions, the scenario generation tools 132 and 220 can update the second incident individual information to reflect this.

[0169] According to one embodiment of the present invention, after reflecting the user's response, the second incident individual information includes more specific details about the accident progression and damage situation. For example, in addition to the time of the accident, time sequences such as the time when the SUV rear-ended a passenger car waiting at a traffic light, the time when the impact caused the victim vehicle to hit a guardrail, resulting in a secondary accident, the time when the driver of the offending vehicle got out of the vehicle and attempted to communicate with the driver of the victim vehicle, and the time when the police and insurance company were dispatched to investigate the accident scene may be added. Furthermore, information may be added that the rear bumper of the victim vehicle was severely damaged, the trunk was deformed, but the rear window was undamaged. Information may also be added that the driver of the victim vehicle visited a hospital complaining of neck pain, and that the driver of the offending vehicle was uninjured and on the phone with his insurance company.

[0170] Factual summary generation stage (S210)

[0171] According to one embodiment of the present invention, in the facts summary generation step (S210), the scenario generation tool 132, 220 performs the role of summarizing the details of the accident based on the stored first incident individual information and second incident individual information to generate a structured summary. The purpose of the facts summary generation step (S210) is to generate a summary in the form of pre-organized data to be used in the text-based scenario script generation step (S220), including basic information such as the time of the accident, location, incident type, involved vehicles, and scale of damage, as well as the main details of the accident progression process.

[0172] According to one embodiment of the present invention, the fact summary generation step (S210) serves to summarize only important information rather than providing a detailed description of the progress of the case, and is a process of preparing for the creation of a natural and specific description in the subsequent text-based scenario script generation step (S220). That is, in the fact summary generation step (S210), the scenario generation tool 132, 220 organizes the outline of the case, and based on this, processes information so that a storytelling-style scenario script can be generated in the text-based scenario script generation step (S220).

[0173] According to one embodiment of the present invention, in the case of a rear-end collision, the first incident individual information may include basic information such as the time of the accident (3:00 PM on February 15, 2025), the location of the accident (Dogok-dong intersection, Gangnam-gu, Seoul), the type of incident (rear-end collision), and the vehicles involved (SUV and passenger car). However, this information alone cannot clearly explain the progression of the accident and the scale of the damage. Therefore, the scenario generation tool checks against a database of lawsuit types and incident types to see if any important information about the rear-end collision has been omitted, and then performs additional questions to supplement the second incident individual information (the progression of the accident and detailed damage situation).

[0174] Additional information obtained in this process may include whether the impact pushed the victim vehicle and caused an additional collision with the guardrail, whether the trunk and rear window of the victim vehicle were deformed or broken, whether the driver of the victim vehicle was injured in the accident and visited a hospital, etc. Based on this supplemented information, the scenario generation tool 132, 220 generates a summary of the facts, which serves as basic data for conversion into a scenario script in the next step (S220).

[0175] FIG. 3b is a diagram for explaining a summary of facts generated by the scenario generation tool according to the present invention.

[0176] For example, the summary of facts may be organized in the format shown in Figure 3b. The summary of facts organized as shown in Figure 3b is converted into a more natural and specific storytelling format in the text-based scenario script generation step (S220), and can be used in a variety of subsequent applications, such as legal consultation, insurance review, and AI video generation.

[0177] Text-based scenario script generation step (S220)

[0178] Figure 4a is a diagram showing an example of scene 1 of a factual summary generated by a scenario generation tool according to an embodiment of the present invention, Figure 4b is a diagram showing an example of scene 2 of a factual summary generated by a scenario generation tool according to an embodiment of the present invention, and Figure 4c is a diagram showing an example of scene 3 of a factual summary generated by a scenario generation tool according to an embodiment of the present invention.

[0179] Figure 5a is a diagram showing an example of scene 1 of a text-based scenario script generated by a scenario generation tool according to an embodiment of the present invention, Figure 5b is a diagram showing an example of scene 2 of a text-based scenario script generated by a scenario generation tool according to an embodiment of the present invention, and Figure 5c is a diagram showing an example of scene 3 of a text-based scenario script generated by a scenario generation tool according to an embodiment of the present invention.

[0180] As shown in Figures 4a to 5c, the scenario generation tool 132, 220 can generate a scenario script.

[0181] FIG. 6 is a diagram showing a step in which the creation tool according to the present invention creates an image based on a factual summary and a scenario script.

[0182] Scenario script analysis stage (S300)

[0183] According to one embodiment of the present invention, the creative creation tool 133, 230 receives a factual summary (S210) and a scenario script (S220) and performs a process of visually recreating the accident situation in conjunction with an AI image generation engine. Through this, the present invention simulates the accident scene in an image to enable a more intuitive understanding, and can be used for various purposes such as legal consultation, insurance review, and use as evidence in court.

[0184] According to one embodiment of the present invention, the creative product generation tool 133, 230 analyzes input text data to extract key elements of an accident and, based on the extracted elements, creates image entity information necessary for image generation. The creative product generation tool 133, 230 identifies the type of incident and analyzes environmental factors, such as the location, time of day, and weather, of the accident, and then identifies objects involved in the accident, such as the vehicle, driver, and pedestrian. For example, if a scenario script is input stating, "A passenger car was waiting at a traffic light at the Dogok-dong intersection in Gangnam-gu, Seoul at 2:45 PM on February 15, 2025, when it was hit from behind by an SUV," the creative product generation tool 133, 230 analyzes the input data and classifies the accident type entity information as a rear-end collision. It then sets the background to an intersection in downtown Seoul, adjusts the weather to sunny, and adjusts the time of day to afternoon. The creative product generation tool 133, 230 can also determine the color and damage level of the passenger car and SUV, and analyze the driver's condition, such as whether they were injured, to determine additional animation elements.

[0185] Scene-specific graphic elements and environment composition step (S310)

[0186] According to an embodiment of the present invention, the creative creation tool 133, 230 can generate graphic elements in conjunction with an AI image generation engine such as SORA or VEO2 based on the analyzed information. The creative creation tool 133, 230 can use the AI ​​image generation engine to set up a 3D environment based on information defined in a scenario and perform background modeling of roads, buildings, lighting, etc. that match the accident scene. Specifically, the creative creation tool 133, 230 can generate offending and victimized vehicles by reflecting the type and color of the vehicles and place characters such as drivers and pedestrians.

[0187] According to an embodiment of the present invention, the creation generation tool 133, 230 can apply animations tailored to each scene using an AI image generation engine. For example, in the case of a rear-end collision, a scene is simulated in which an SUV is unable to slow down and collides with the victim vehicle, and an animation is applied in which the vehicle shakes and the rear bumper is dented during the collision. If the impact of the accident is severe, the creation generation tool 133, 230 can add a scene in which the vehicle is pushed forward and collides with a guardrail. The creation generation tool 133, 230 can also simulate the driver's reaction, which may include a scene in which the driver of the victim vehicle clutches his neck in pain.

[0188] AI voice (TTS) and narration output stage (S320)

[0189] According to an embodiment of the present invention, the creation creation tool 133, 230 can use Text-to-Speech (TTS) or AI voice generation technology to realize the lines of characters as audio. For example, a scene may be included in which an SUV driver frantically utters, "Sorry, I didn't brake fast enough!" immediately after the accident occurs, and a line from the driver of the victim vehicle saying, "My neck hurts so much. I think I should go to the hospital." Furthermore, when the police and insurance company arrive at the accident scene and begin investigating the accident, they may add a narration such as, "After reviewing the dashcam footage, it is clear that an SUV crashed into me from behind."

[0190] According to an embodiment of the present invention, the creative creation tool 133, 230 can adjust the production of each scene by reflecting the progression and context of the accident, rather than simply visually converting text data. For example, if the accident occurs at night, the creative creation tool 133, 230 can adjust the lighting using an AI engine to embody a dark environment and adjust the brightness of street lamps and vehicle headlights. If it is raining, a raindrop effect can be added to reflect the wet surface of the road.

[0191] FIG. 7a shows a scene before the accident occurs in the video generated by the creation creation tool by inputting the accident occurrence script of scene 1 shown in FIG. 5a.

[0192] As shown in Figure 7a, the creative creation tool 133, 230 can create a video of the script portion, "The driver was listening to the radio in the car while waiting at a traffic light. The road was deserted and the vehicle in front was also stopped."

[0193] According to an embodiment of the present invention, the creation creation tool 133, 230 can analyze text information in a scenario script and then, based on the analyzed text information, visually realize the scene in question in conjunction with an AI image generation engine and insert appropriate narration, thereby enabling the image to provide a more intuitive explanation of the accident situation together with audio information rather than being limited to mere visual elements.

[0194] According to an embodiment of the present invention, the creation generation tool 133, 230 can extract text from a scenario script that matches a scene and match it with a video. For example, if a script such as "The driver was listening to the radio in his car while waiting at a traffic light. The road was deserted, and the car in front was also stopped," is input, the creation generation tool 133, 230 can analyze it and determine which scene this sentence corresponds to. If it is determined that this scene matches the driver looking at the road from inside the car, the creation generation tool 133, 230 prepares to apply the script.

[0195] According to an embodiment of the present invention, the creative product generation tool 133, 230 may utilize Text-to-Speech (TTS) or AI voice technology to convert text information into audio. Specifically, after analyzing the sentence, "The driver was listening to the radio in his car while waiting at a traffic light," the creative product generation tool 133, 230 may apply an appropriate voice style using an AI voice synthesis engine. The creative product generation tool 133, 230 may adjust the tone and speed of the narration to generate a natural voice appropriate to the situation. For example, in a scene in which the driver is quietly waiting at a traffic light, the creative product generation tool 133, 230 may generate a quiet, natural narration voice rather than a very fast, emotional voice. The creative product generation tool 133, 230 may insert the generated audio file into a video in conjunction with a video editing tool. For example, as shown in FIG. 7a, a narration voice such as, "The driver was listening to the radio in his car. The roads were quiet and the weather was good," may be output along with the video.

[0196] According to one embodiment of the present invention, after the audio is inserted, the creative product generation tool 133, 230 can perform a process of adjusting the background sound so that the narration and video blend naturally. Since the scene is set inside a vehicle, the creative product generation tool 133, 230 can set the radio sound to be faintly audible and add the faint engine noise inside the vehicle. The creative product generation tool 133, 230 can also adjust light external noises such as wind noise and the sound of passing cars to reflect the road environment, thereby creating an overall realistic atmosphere. The creative product generation tool 133, 230 can also perform an audio mixing process to adjust the volume of the background sound and narration so that the narration can be clearly heard.

[0197] According to an embodiment of the present invention, the creative product creation tool 133, 230 can synchronize the timing of video and narration. For example, when a line begins, "The driver was listening to the radio in his car while waiting at a traffic light," the creative product creation tool 133, 230 can adjust the timing so that a scene of the driver looking ahead is naturally connected. Next, when the line, "The road was deserted, and the car ahead was also stopped," appears, the creative product creation tool 133, 230 can edit the video to emphasize the scene of the car ahead being stopped. The creative product creation tool 133, 230 can also insert subtitles and display text that matches the narration at the bottom of the screen. This provides both visual and audio information, making the video more intuitive to understand.

[0198] FIG. 7b shows a scene immediately after the accident occurs in the video generated by the creation creation tool by inputting the accident occurrence script of scene 1 shown in FIG. 5a.

[0199] As shown in Figure 7b, the creative creation tool 133, 230 can create a video of the script portion, "(SUV vehicle: approaching quickly from behind) An SUV vehicle was approaching quickly from behind. (The vehicle shakes violently with a sudden collision sound), (The driver of the victim vehicle is confused as he leans forward due to the impact), (The SUV vehicle suddenly stops, and the sound of tires scraping is heard)."

[0200] According to an embodiment of the present invention, the creative product generation tool 133, 230 analyzes the factual summary (S210) and the scenario script (S220) and visually embodies the scene in conjunction with an AI image generation engine, and inserts narration and sound effects. In this case, the creative product generation tool 133, 230 may provide a more realistic accident simulation by including text-to-speech (TTS) and sound effects that describe the driver and vehicle movements. In the scene of Figure 7b, the moment an SUV rapidly approaches and collides with the vehicle in front is captured, and based on this, the following narration may be inserted:

[0201] Scenario script analysis and scene matching

[0202] According to an embodiment of the present invention, the creation creation tool 133, 230 can analyze text that matches a scene in a scenario script and match it with a video. The creation creation tool 133, 230 can also analyze a script such as Table 1 to check whether it matches a scene in the video.

[0203] [Table 1]

[0204] According to an embodiment of the present invention, the creation generation tool 133, 230 can analyze parts of the script that correspond to the scene, match them with the actual video, and perform the application.

[0205] Scene and script matching process

[0206] According to one embodiment of the present invention, the current video captures the moment when an SUV crashes into a victim vehicle at high speed, which matches the part of the scenario that reads, "The SUV was approaching quickly from behind," so the creation generation tool 133, 230 can match the scene with the script. In addition, since the current video includes a scene in which the vehicle shakes after the collision, the creation generation tool 133, 230 can match the script that reads, "The vehicle shakes violently along with the sound of a sudden collision."

[0207] AI voice (TTS) generation and application

[0208] According to an embodiment of the present invention, the creation generation tool 133, 230 can convert the script into voice using Text-to-Speech (TTS) or AI voice technology. Specifically, the creation generation tool 133, 230 can generate a narration (TTS) for the phrase, "An SUV was approaching fast from behind." The creation generation tool 133, 230 can also apply a car driver's TTS line for the phrase, "What?! Did that just hit me from behind?" The creation generation tool 133, 230 can also apply an SUV driver's TTS line for the phrase, "Damn... I braked too late!" In this case, the creation generation tool 133, 230 can set the car driver's line to a tone that reflects a surprised emotion and adjust the SUV driver's line to a panicked voice. The creation generation tool 133, 230 can also generate a narration in an objective, calm voice.

[0209] Added sound effects and environmental sounds

[0210] According to an embodiment of the present invention, the creative product generation tool 133, 230 can perform a process of adding background sound effects, collision sounds, and tire scuffing sounds so that narration and dialogue blend naturally with the video. Specifically, the creative product generation tool 133, 230 can insert a strong impact sound effect in the section, "The vehicle shakes violently with the sound of a sudden collision." Furthermore, the creative product generation tool 133, 230 can insert the sound of sudden braking in the section, "The SUV suddenly stops, tire scuffing sounds are generated." Furthermore, the creative product generation tool 133, 230 can add the sound of the driver's movements occurring inside the vehicle after a collision and the sound of the safety belt being pulled. Furthermore, the creative product generation tool 133, 230 can insert the sound of wind and background noise while the vehicle is moving, creating a momentary silence after the collision and emphasizing a tense atmosphere.

[0211] Adjusting the timing of the video and narration

[0212] According to an embodiment of the present invention, the creative product generation tool 133, 230 can perform synchronization adjustment so that narration and sound effects accurately match the scene in the video. Specifically, the creative product generation tool 133, 230 can start the narration at the moment an SUV approaches quickly. The creative product generation tool 133, 230 can insert the narration, "An SUV was approaching quickly from behind," at the beginning of the scene. The creative product generation tool 133, 230 can also insert the lines of the car driver just before the collision scene. For example, the line, "What?! Why are you approaching so fast?" can be applied just before the collision. Then, the creative product generation tool 133, 230 can insert a strong collision sound effect at the moment of the collision and then apply the lines of the SUV driver. The creative product generation tool 133, 230 can insert the line, "Damn... I braked too late!" just after the collision. Finally, the creative tools 133, 230 can create tension by inserting interior noise and short periods of silence when the vehicle shakes after a collision.

[0213] FIG. 7c shows a scene in which the car crashes into a guardrail immediately after the accident occurs in a video generated by the creation creation tool by inputting the accident occurrence script of scene 1 shown in FIG. 5a.

[0214] As shown in FIG. 7c, the creation generating tool 133, 230 can create a part of the script "(the victim vehicle is pushed forward by the impact and collides with the guardrail)" as an image.

[0215] Scenario script analysis implementation example

[0216] According to an embodiment of the present invention, the creative product generation tool 133, 230 analyzes the factual summary and the scenario script, and in conjunction with an AI video generation engine, visually embodies the accident scene and can perform the role of creating a more realistic accident reenactment video, including the driver's lines, the portrayal of a tense situation, screams, and impact sounds. In the scene of Figure 7c, it is necessary to recreate the moment when the victim's vehicle is pushed forward by the collision of the SUV and collides with the guardrail, and during this process, it is necessary to include the driver of the victim's vehicle screaming in a tense voice.

[0217] According to an embodiment of the present invention, the creation generation tool 133, 230 can analyze elements appropriate for a given scene in a scenario script and match them with an AI engine to implement visual and auditory elements. For example, if a script stating, "The victim vehicle is pushed forward by an impact and collides with a guardrail," is input, the creation generation tool 133, 230 analyzes the script and recognizes the need for background effects that reflect the vehicle's movement, the collision process, the driver's reaction, and the surrounding environment, and can generate animations based on the analysis. In this case, the creation generation tool 133, 230 can create natural-looking scenes, including the victim vehicle being pushed forward quickly and turning the steering wheel hard, the driver screaming and shaking inside the vehicle, an animation of the vehicle denting when it hits the guardrail, and the driver gasping for breath as the vehicle stops after the collision.

[0218] According to an embodiment of the present invention, the creation generation tool 133, 230 may realistically embody vehicle movement by utilizing a deep learning model that performs physics-based collision simulation. Specifically, the creation generation tool 133, 230 may physically calculate the interaction with a guardrail during a vehicle collision by utilizing a physics simulation engine model such as PhysX, Bullet Physics, or NVIDIA FleX, thereby applying vehicle deformation (dents), impact recoil, and glass breakage effects. Furthermore, the creation generation tool 133, 230 may generate facial expressions and movements of a driver inside a vehicle when impacted by an impact using GAN-based AI image generation models such as StyleGAN, VQ-VAE-2, and DALL·E 2. Furthermore, the creation generation tool 133, 230 may realistically render guardrails and surrounding backgrounds to reflect the actual road environment by utilizing scene reconstruction technologies such as NeRF and 3D Gaussian Splatting, thereby embodying scenes in which a vehicle moves and surrounding objects react.

[0219] Serif Generation Implementation

[0220] According to an embodiment of the present invention, the creative product generation tool 133, 230 can perform a task of naturally generating and inserting the driver's lines once video production is complete. Specifically, the creative product generation tool 133, 230 can use a Text-to-Speech (TTS) model such as Tacotron2, FastSpeech2, or Wav2Vec2 to synthesize voice that reflects the driver's emotions, and can apply Emotion AI technology to reflect emotions such as fear and surprise. In this case, the driver's lines may include a nervous voice such as "Ah! No! I have to stop!" just before a collision, a loud scream such as "Whoa!" at the moment of impact, and a breathless voice such as "Ah...ah...what...what's going on..." after the impact. Therefore, the creative product generation tool 133, 230 can use a Text-to-Speech (TTS) model to more naturally express the driver's emotions. In addition, the creative creation tools 133, 230 can adjust the driver's mouth shape to naturally synchronize with the voice by utilizing Lip Sync AI (Wav2Lip, DeepFaceLive).

[0221] Implementation of adding sound effects

[0222] According to an embodiment of the present invention, the creative product generation tool 133, 230 may perform a process of adding collision sound effects. Specifically, the creative product generation tool 133, 230 may generate the sounds of metal hitting each other and glass breaking by using an audio synthesis model such as Google Magenta, WaveNet, or Jukebox AI to maximize the realism of an accident. The creative product generation tool 133, 230 may also use an Audio Source Separation (Spleeter, Demucs) model to separate audio from background noise and optimize audio so that the sounds of collisions and screams are clearly audible. Furthermore, in the sound effect application process, the creative product generation tool 133, 230 may insert a strong collision sound when the SUV pushes the victim vehicle, reflect the sound of skidding tires during sudden stops and turns, and add the sound of metal breaking when the vehicle frame dents when it hits a guardrail. At this time, the creative creation tool 133, 230 can also include the sound of a vehicle window breaking depending on the strength of the collision, and can also emphasize the tense voice of the driver breathing heavily after the collision to further enhance the sense of realism after the accident.

[0223] Video and Audio Adjustment Embodiments

[0224] According to one embodiment of the present invention, the creative product generation tool 133, 230 can precisely adjust the generated video and audio to create a natural-looking scene, like an actual accident scene. To this end, the creative product generation tool 133, 230 can start the driver's scream at the moment of collision and time it so that the collision sound and scream occur simultaneously as the vehicle is pushed. In addition, the creative product generation tool 133, 230 can insert the sound of the vehicle shaking after colliding with the guardrail, and end the scene with the driver gasping for breath in shock.

[0225] Figure 7d is a diagram showing a scene in which the SUV driver gets out of the vehicle and approaches the victim vehicle after the accident occurs, in a video generated by the creative creation tool by inputting the accident occurrence script for scene 2 shown in Figure 5b.

[0226] As shown in FIG. 7d, the creation generation tool 133, 230 can create a video of the script portion "(the SUV driver gets out of the vehicle and approaches the victim vehicle)."

[0227] An embodiment of element analysis related to the scene in the scenario script

[0228] According to an embodiment of the present invention, the creative product generation tool 133, 230 can analyze elements related to a scene in a scenario script and match them with an AI engine to adjust the scenario script to reflect realistic movements and emotional expressions. Specifically, the creative product generation tool 133, 230 can extract content from the scenario script, such as "The SUV driver gets out of the vehicle and approaches the victim vehicle," "The driver quickly gets out of the vehicle with a surprised expression and runs away," "The driver's panicked expression after seeing the car that crashed into the guardrail," "The remains of the vehicle reflect under the light of a street lamp," and "The sound of a light breeze and footsteps gradually approaching on the road," and then perform a process of creating a scene based on the extracted content. For example, the creative product generation tool 133, 230 can generate an animation of an SUV driver getting out of the vehicle and can naturally embody the action of the car door opening and the driver quickly jumping out. In addition, the creative product generation tool 133, 230 can reflect the motion of the driver accelerating his walking and balancing his body on the road, and can also create the action of looking at the vehicle that crashed into the guardrail.

[0229] Implementation of deep learning-based motion analysis and behavior generation model

[0230] According to an embodiment of the present invention, the creation generation tool 133, 230 may adjust the driver's movements using a deep learning-based motion analysis and behavior generation model to naturally create a scene in which an SUV driver moves quickly and shows a surprised expression. Specifically, the creation generation tool 133, 230 may track a person's movements using a Pose Estimation & Motion Capture model such as OpenPose, DeepLabCut, or AlphaPose to naturally generate a fast running motion. The creation generation tool 133, 230 may analyze the running movements of real people to train an AI character to move similarly, and may apply a physics-based character animation model such as DeepMimic, NVIDIA PhysX, or Unity ML-Agents to reflect the physical movements of an SUV driver running quickly and balancing.

[0231] According to an embodiment of the present invention, the creation generation tool 133, 230 can enhance the immersiveness of a scene by applying natural animations using a physics engine to realize fast-paced gait, body sway, and accelerated movements. Furthermore, the creation generation tool 133, 230 can naturally express the surprised expression of an SUV driver by using a GAN-based Facial Expression Synthesis model such as StyleGAN, FaceFormer, or First Order Motion Model. In this case, the creation generation tool 133, 230 can add subtle facial muscle movements and adjust eye movements when the driver panics upon seeing the vehicle, thereby expressing emotions more realistically.

[0232] Lighting and Background Adjustment Embodiments

[0233] According to an embodiment of the present invention, the creative product generation tool 133, 230 can adjust the lighting and background in a scene in which an SUV driver approaches a victim vehicle at night. The creative product generation tool 133, 230 can adjust the reflected light of road streetlights and vehicle headlights using a night-time scene enhancement model such as Pix2Pix, CycleGAN, or HDRNet to make people and vehicles appear more clearly in a nighttime environment. Furthermore, the creative product generation tool 133, 230 can generate a natural effect of streetlight light reflecting off the SUV and road surface using a 3D scene reconstruction model such as NeRF or 3D Gaussian Splatting, creating an environment in which vehicle debris moves minutely due to the wind. Furthermore, the creative product generation tool 133, 230 can adjust the color and intensity of streetlight lighting to realistically embody a road environment and reflect shadows and reflections along the driver's path, creating a more realistic nighttime accident scene.

[0234] Embodiment of sound effect generation

[0235] According to an embodiment of the present invention, the creation generation tool 133, 230 should include footsteps, wind, vehicle debris, and the driver's breathing in a scene where a driver is rushing toward them. Specifically, the creation generation tool 133, 230 can use a Footstep Sound Generation model such as Google Magenta, WaveNet, or Jukebox AI to naturally generate footstep sounds that match the road surface and adjust the frequency of the footsteps depending on the running speed. In addition, the creation generation tool 133, 230 can use an Emotion-Based Speech Synthesis model such as Tacotron2, FastSpeech2, or Wav2Vec2 to generate lines spoken by an SUV driver who is out of breath and adjust the breathing sounds and speaking style to reflect emotion.

[0236] Example of sound effect application

[0237] According to one embodiment of the present invention, the creative product generation tool 133, 230, in the sound effect application process, can create the feeling of an SUV driver getting out of the vehicle and getting closer to the victim vehicle by adding the sound of light wind on the road to enhance the realism of an outdoor environment. The creative product generation tool 133, 230 can also create a tense atmosphere by adding the subtle background sound of a vehicle passing by in the distance and the sound of the driver breathing heavily as he drives. The creative product generation tool 133, 230 can also maximize the tension of the accident situation by adding a scene in which the SUV driver approaches and says in a panic, "Oh, damn! Are you okay?!", and then whispers in a panicked voice, "What is this... what should I do..." upon seeing the victim vehicle.

[0238] Video and Audio Adjustment Embodiments

[0239] According to an embodiment of the present invention, the creative product generation tool 133, 230 can precisely adjust the generated video and audio to create a natural-looking scene resembling an actual accident scene. Specifically, in the timing adjustment process, the creative product generation tool 133, 230 can play a scene in which the SUV stops, the door opens, and the driver jumps out, and can apply a scene in which the footsteps and wind sounds get louder as the driver drives away. In addition, the creative product generation tool 133, 230 can insert dialogue in a scene in which the driver's breathing increases as the driver approaches the vehicle that has collided with the guardrail, and he looks panicked when he sees the vehicle, and can end the scene with the driver arriving at the vehicle, checking for injuries, and catching his breath.

[0240] Figure 7e is a diagram showing a scene in which the rear bumper of the passenger car and the front bumper of the SUV are dented after the accident occurs, as generated by the creative creation tool after inputting the accident occurrence script for scene 2 shown in Figure 5b.

[0241] Scenario script analysis implementation example

[0242] According to one embodiment of the present invention, the creative product generation tool 133, 230 can analyze a scenario script and, in conjunction with an AI-based video generation engine, create a realistic accident scene by reflecting the emotional expressions, tone of speech, facial expressions, gestures, voice synthesis, and background environment of the passenger car driver and SUV driver after the accident. In the scene shown in Figure 7e, the creative product generation tool 133, 230 can add the SUV driver approaching with an apologetic attitude and a troubled expression, and the passenger car driver reacting with irritation due to the accident and rubbing his neck in discomfort. In addition, the creative product generation tool 133, 230 can naturally express the flow of real-time dialogue and changes in emotions to create a video similar to an actual accident scene.

[0243] Embodiments of Reflecting Driver Movements and Emotional Expressions

[0244] According to an embodiment of the present invention, the creation creation tool 133, 230 can analyze elements related to the scene in the scenario script and match them with the AI ​​engine to perform adjustments so that the driver's movements and emotional expressions are reflected naturally. Specifically, the creation creation tool 133, 230 may include in the scenario script, "The SUV driver gets out of the vehicle and approaches the victim vehicle," "SUV driver (approaching suddenly): 'Are you okay? I'm so sorry! I was too late to notice the car in front of me...'," "Car driver opens the door, gets out and checks the rear of the vehicle," "Car driver (in an irritated voice): Oh, really... the rear bumper is completely dented! The trunk won't even close," "SUV driver (hastily looking closely at the vehicle): Oh... my car has a dent in the front bumper too," "There's a small scratch. Oh, what should we do about that?", "Car driver rubs his neck and shakes his head", "Car driver (looking a little distressed): By the way... my neck really hurts. It seems the impact was severe", "SUV driver (worried): Uh... maybe you should go to the hospital? Should I call the insurance company?", "Car driver (sighing and taking out his cell phone): We need to call the insurance company and the police. We also need to check the dashcam." Content such as these can be extracted and the process of directing the scene can proceed based on this.

[0245] According to an embodiment of the present invention, the creation creation tool 133, 230 can simulate the SUV driver hurrying up and making a worried expression, and the passenger car driver seeing the damage from the accident and making an irritated expression. The creation creation tool 133, 230 can also simulate the SUV driver making anxious hand gestures and bowing his head to express regret, and the passenger car driver checking the vehicle and sighing or rubbing his neck to express inconvenience. In this case, the creation creation tool 133, 230 can adjust facial and body animations using a deep learning-based emotion expression model to naturally embody the SUV driver making an apologetic and confused expression and the passenger car driver making an irritated expression.

[0246] Facial animation embodiment

[0247] According to an embodiment of the present invention, the creation generation tool 133, 230 may use a Facial Expression Synthesis model such as StyleGAN, First Order Motion Model, EmoGAN, or FaceFormer for facial animation to generate facial expressions that express embarrassment, regret, or worry for the SUV driver, and may embody facial expressions that express irritation, inconvenience, or pain for the car driver. Additionally, the creation generation tool 133, 230 may use a Pose Estimation & Gesture Animation model such as OpenPose, AlphaPose, or DeepLabCut to apply the motion of the SUV driver raising their hand to express regret and looking closely at the vehicle, and add a gesture of the car driver rubbing their neck to express inconvenience.

[0248] According to one embodiment of the present invention, the creation creation tool 133, 230 can create an expression in which the SUV driver hastily gets out of the vehicle, looks worried, frowns in apologetic manner, and opens his mouth slightly to more effectively embody such emotional expressions. On the other hand, the creation creation tool 133, 230 can create an expression in which the car driver looks behind the vehicle, frowns, and rubs his neck with one hand to express pain, emphasizing the damage caused by the accident. The creation creation tool 133, 230 can create natural dialogue scenes by linking the generated facial expressions and gestures with emotion-based voice synthesis.

[0249] Embodiment of speech synthesis model application

[0250] According to an embodiment of the present invention, the creation generation tool 133, 230 may apply a speech synthesis model so that the SUV driver speaks in a concerned and apologetic tone, while the passenger car driver responds with a mixed feeling of irritation. Specifically, the creation generation tool 133, 230 may utilize an emotion-based speech synthesis model, such as Tacotron2, FastSpeech2, Wav2Vec2, or RVC, to generate natural-sounding dialogue speech that reflects emotional tones. The creation generation tool 133, 230 may also adjust the synchronization between speech and facial expressions by applying a lip sync AI model, such as Wav2Lip or DeepFaceLive, to naturally align speech and facial expressions. For example, when the SUV driver says, "Are you okay? I'm so sorry!", the creation generation tool 133, 230 may apply a breathless tone to reflect a sense of urgency. Additionally, the creative generator 133,230 can apply an annoyed voice and a sigh when the car driver says, "Oh my gosh... my rear bumper is totally dented!"

[0251] Implementation of adding sound effects

[0252] According to an embodiment of the present invention, the creation generation tool 133, 230 may add vehicle wreckage, road noise, vehicle horns, and the voice of an insurance company phone call to enhance the realism of an accident scene. Specifically, the creation generation tool 133, 230 may use a sound synthesis model such as WaveNet, Google Magenta, or Jukebox AI to generate vehicle crash sounds, road noise, and mobile phone operation sounds. The creation generation tool 133, 230 may also use an audio source separation model such as Spleeter or Demucs to naturally mix audio and ambient noise. In this case, the creation generation tool 133, 230 may insert the sound of an SUV driver getting out of the car and opening the door, and add the sound effects of the SUV driver's approaching footsteps, road noise, and vehicle wreckage gently rustling in the wind.

[0253] Video and Audio Adjustment Embodiments

[0254] According to an embodiment of the present invention, the creative product generation tool 133, 230 can precisely adjust the generated video and audio to accurately match the driver's facial expressions, emotional expressions, voice, and gestures. In the timing adjustment process, the creative product generation tool 133, 230 can play a scene in which the SUV stops, the driver jumps out the moment the door opens, and apply a performance in which the footsteps and wind sounds get louder as the SUV driver speeds away. In addition, the creative product generation tool 133, 230 can adjust the voice of the car driver checking the vehicle and speaking in an irritated voice, adjust the voice of the SUV driver speaking with a confused expression, and then end the video by switching to a scene in which the insurance company and police are called.

[0255] Figure 7f is a diagram showing a scene in which insurance company officials and police have gathered at the scene of the accident after the accident occurred, in a video generated by the creative creation tool by inputting the accident occurrence script for scene 3 shown in Figure 5c.

[0256] Scenario script analysis implementation example

[0257] According to one embodiment of the present invention, the creative work generation tool 133, 230 can analyze elements related to the scene in the scenario script and match them with an AI engine to adjust the movements and dialogue of the police and insurance company employees to reflect natural behavior. Specifically, the scenario script includes a scene in which a police officer approaches with a notebook and pen, asking, "We've received a report of an accident. Are you both okay?" The driver of the passenger car responds, clutching his neck, "My neck hurts a little. I was hit from behind suddenly..." The SUV driver admits to the police, "Yes... I hit the rear end. I braked too late." The police then ask the SUV driver, "So, was the driver of the at-fault vehicle unable to maintain a safe distance?" The SUV driver nods apologetically and replies, "Yes, it seems I was largely at fault." The insurance company employee then reviews and analyzes the dashcam footage, explaining, "After checking the dashcam, we found that the victim vehicle was stopped and the SUV suddenly braked. It's highly likely that the SUV driver is 100% at fault." The SUV driver sighs and asks, "Yes, it was definitely my mistake. What should we do?" The police leave a record and explain, "First, the driver of the victim vehicle needs to be examined at a hospital. The driver's degree of fault will be determined after a formal investigation, but at this point, it's highly likely that the SUV bears 100% responsibility." The car driver sighs and says, "Ugh... My whole schedule for today is ruined. I'll go from the hospital." He then takes the car to the hospital in an ambulance, the police leave the scene of the accident and proceed with towing the vehicle, and the insurance company completes the accident report and begins the compensation process. The scene ends with the scene.

[0258] According to one embodiment of the present invention, the creative creation tool 133, 230 can naturally depict the lines and facial expressions of police and insurance company employees, the process of reviewing dashcam footage, and vehicle towing procedures. The creative creation tool 133, 230 can depict police holding a notebook and pen and recording the accident situation, insurance company employees reviewing dashcam footage and analyzing the accident details, the passenger car driver clutching his neck and explaining whether or not there were any injuries, and the SUV driver sighing with an apologetic expression. The creative creation tool 133, 230 can add lines showing the police objectively conducting and recording the accident investigation while the police and insurance company employees are taking measures to resolve the accident, and the insurance company employees analyzing the accident footage with an expert attitude and explaining the compensation procedure to the driver.

[0259] Facial animation embodiment

[0260] According to an embodiment of the present invention, the creative product generation tool 133, 230 can naturally express such scenes by naturally embodying the calm expressions of the police and insurance company employees while investigating the incident, the analytical attitude as they grasp the accident situation, the frustrated reaction of the driver of the victim vehicle, and the troubled expression of the driver of the at-fault vehicle. Specifically, the creative product generation tool 133, 230 can use a Facial Expression Synthesis model, such as StyleGAN, First Order Motion Model, EmoGAN, or FaceFormer, for facial animation to generate the expression of the police calmly but carefully recording the incident, and the expression of the insurance company employee concentrating on analyzing the dashcam. In addition, the creative product generation tool 133, 230 can generate the expression of the SUV driver nodding apologetically and the expression of the driver of the victim vehicle sighing and making a frustrated expression. Additionally, the creative tool 133,230 can utilize Pose Estimation & Gesture Animation models such as OpenPose, AlphaPose, and DeepLabCut to create gestures for police officers taking notes and asking questions, and add gestures for insurance company personnel operating dashcams and analyzing data.

[0261] Embodiment of speech synthesis model application

[0262] According to an embodiment of the present invention, the creation generation tool 133, 230 can apply a speech synthesis model to make the lines of the police and insurance company employees sound realistic. Specifically, the creation generation tool 133, 230 can generate a video in which the police ask questions in a calm yet authoritative manner and the insurance company employee explains in a professional tone. The creation generation tool 133, 230 can also generate a video in which the SUV driver maintains an apologetic tone and the passenger car driver expresses an irritated reaction and complains of injury. To this end, the creation generation tool 133, 230 can utilize an emotion-based speech synthesis model such as Tacotron2, FastSpeech2, Wav2Vec2, or RVC.

[0263] Audio and Expression Sync Adjustment Embodiments

[0264] According to an embodiment of the present invention, the creation generation tool 133, 230 can adjust the synchronization between voice and facial expression by applying a Lip Sync AI model such as Wav2Lip or DeepFaceLive to naturally align voice and facial expression. The creation generation tool 133, 230 can also apply a calm and authoritative voice when a police officer says, "We've received a report of an accident. Are you both okay?" The creation generation tool 133, 230 can also apply an analytical tone when an insurance company employee explains, "We checked the dashcam and found that the victim's vehicle was parked and the SUV suddenly braked." The creation generation tool 133, 230 can also reflect a flustered and apologetic emotion when an SUV driver says, "Yes, it seems I was largely at fault," and can also apply a sighing and irritated voice when a car driver says, "Ugh... my whole schedule is ruined today."

[0265] Implementation of adding sound effects

[0266] According to one embodiment of the present invention, the creative product generation tool 133, 230 can add vehicle debris, road noise, police radio sounds, vehicle towing sounds, and dashcam footage playback sounds to enhance the realism of an accident scene. Specifically, the creative product generation tool 133, 230 can use a sound synthesis model such as WaveNet, Google Magenta, or Jukebox AI to generate vehicle wreckage sounds, road noise, and cell phone operation sounds, and an audio source separation model such as Spleeter or Demucs to naturally mix voice and ambient noise. The creative product generation tool 133, 230 can also insert the sound of police writing notes when recording an incident, a soft "click" sound when dashcam footage is played, and the sound of a tow truck moving a vehicle.

[0267] Video and Audio Adjustment Embodiments

[0268] According to one embodiment of the present invention, the creative creation tool 133, 230 can precisely adjust the generated video and audio to naturally portray the investigation process between the police and insurance company staff. Specifically, the creative creation tool 133, 230 can insert a scene in which the police arrive at the accident scene and begin questioning the SUV driver, who responds apologetically, and a process in which the insurance company staff checks and analyzes the dashcam. The creative creation tool 133, 230 can then add a scene in which the police record the accident, order the vehicle to be towed, and transport the driver of the car to the hospital.

[0269] FIG. 8 is a diagram illustrating a modification request and scenario script update step of the verification and modification tool according to the present invention.

[0270] Video provision and user verification step (S400)

[0271] Video viewing stage (S401)

[0272] According to one embodiment of the present invention, in the video viewing step (S401), the verification and correction tool 134, 240 provides the accident video generated by the creation generation tool 133, 230 to the user, allowing the user to directly view and compare it with the incident. Specifically, the verification and correction tool 134, 240 may provide the accident video in a streaming format and provide a UI interface so that the user can review the entire video from beginning to end. The user can freely search for a specific scene using the video play and pause functions, and can also rewatch a desired scene using the rewind and fast-forward functions.

[0273] According to an embodiment of the present invention, the verification and correction tool 134, 240 can provide a user with a frame-by-frame search function to individually analyze specific frames. Furthermore, the verification and correction tool 134, 240 can display incident-related information, such as the time of the accident, the vehicle color, and the road conditions, on the screen to allow the user to easily review the video. For example, if the time of the accident is displayed incorrectly or the color of the SUV is displayed differently from the actual color while the user is viewing the accident video, the verification and correction tool 134, 240 can check and correct this.

[0274] User convenience function provision step (S402)

[0275] According to an embodiment of the present invention, in the user convenience function providing step (S402), the verification and correction tool 134, 240 may provide various auxiliary functions to enable the user to more accurately review the video. Specifically, the verification and correction tool 134, 240 supports zoom and slow motion functions, allowing the user to zoom in on specific parts of the accident scene and review them, or to slowly replay the moment of impact to analyze the detailed progression of the accident. In addition, the verification and correction tool 134, 240 may provide a comparison function, allowing the user to compare the original dashcam video and the AI-generated video side by side, and may configure a UI that allows the user to view the two videos simultaneously and identify differences.

[0276] According to one embodiment of the present invention, the verification and correction tool 134, 240 provides an error checklist function, allowing a user to check factors such as the time of the accident, vehicle color, road conditions, and vehicle speed. For example, the verification and correction tool 134, 240 allows a user to use a slow-motion function to accurately confirm the point at which an SUV braked. In addition, the verification and correction tool 134, 240 uses a comparison function to allow a user to compare dashcam footage with AI-generated footage and perform an investigation, such as discovering that the vehicle speed was set faster than it actually was.

[0277] Error or missing information confirmation step (S403)

[0278] According to an embodiment of the present invention, in the error or missing information checking step (S403), the verification and correction tool 134, 240 allows the user to review the accident video, check for errors or missing information, organize them, and request corrections. The verification and correction tool 134, 240 provides an error detection function and can detect basic errors and output a warning message to the user. For example, if the time of an accident is recorded in the video as 3:50 PM but the user states otherwise, the verification and correction tool 134, 240 can provide a message saying, "The recorded accident time is incorrect. Do you want to correct it?" The verification and correction tool 134, 240 can also provide a user feedback input function and a UI that allows the user to directly input errors or missing information.

[0279] Step of inputting correction request (S410)

[0280] According to an embodiment of the present invention, the correction request input step (S410) may be a step in which a user reviews an accident video generated by AI and then inputs parts that do not match the actual accident into the system to correct them. In the correction request input step (S410), the verification and correction tool 134, 240 may collect the user's input, structure it, store it in a database, and then compare it with existing scenario scripts and metadata to reflect the changes. The user may input a correction request in various ways, such as text input, visual selection, or voice input, and the verification and correction tool 134, 240 may analyze it to find areas in the video that need to be changed and perform the correction work.

[0281] User Input Method Selection Embodiment

[0282] According to an embodiment of the present invention, the verification and correction tool 134, 240 may provide three input methods for a user to request correction of a portion in which an error is found. Specifically, the verification and correction tool 134, 240 may provide a text input method, allowing a user to directly input correction content in text form. For example, the verification and correction tool 134, 240 may submit a correction request by inputting content such as "The accident occurred at 4:10 PM, not 3:50 PM." In addition, the verification and correction tool 134, 240 may provide a visual selection method, allowing a user to directly click on a specific element in the image, such as the color of a vehicle or the road environment, to request a change.

[0283] For example, the verification and correction tool 134, 240 allows a user to easily select desired elements using a graphical user interface (GUI), such as "The color of the SUV vehicle should be changed from black to white." The verification and correction tool 134, 240 can also provide a voice input method. Thus, if a user verbally transmits a request for correction, the verification and correction tool 134, 240 can recognize the voice and convert it into text, and then perform analysis. For example, if a user says, "The driver hesitated for several seconds after the accident, rather than immediately getting out of the car," the verification and correction tool 134, 240 can analyze this and reflect the request for correction.

[0284] Embodiments of Records in Factual Database 2

[0285] According to an embodiment of the present invention, the verification and correction tool 134, 240 can record a correction request entered by a user in the factual database 2. The verification and correction tool 134, 240 can identify changes by comparing it with existing scenario scripts and metadata. At this time, the verification and correction tool 134, 240 can convert the user's input into structured data and organize it. For example, if the time of an accident is changed, the verification and correction tool 134, 240 can compare the existing data and store the changed information, and the verification and correction tool 134, 240 can determine whether to change it by comparing "existing data: 3:50 PM" with "user input: 4:10 PM." The verification and correction tool 134, 240 then analyzes the existing scenario script and metadata to determine whether the correction request actually needs to be reflected, preventing unnecessary changes and updating only the necessary parts.

[0286] Scenario script and video update stage (S420)

[0287] According to one embodiment of the present invention, in the scenario script and video update step (S420), the verification and editing tool 134, 240 searches for specific scenes in the scenario script and video and reflects the changes based on a revision request input by the user. The verification and editing tool 134, 240 analyzes the user's request, compares it with the existing scenario, and identifies and corrects parts that need to be changed. The verification and editing tool 134, 240 then works in conjunction with the creative creation tool 133, 230 to generate an updated video. At this time, the verification and editing tool 134, 240 may utilize AI models such as natural language processing (NLP)-based text correction, image object search and modification, and voice data reflection.

[0288] Partial identification of a request for correction

[0289] According to an embodiment of the present invention, the verification and correction tool 134, 240 can compare an existing scenario script with the existing script and update it to reflect a user's correction requests. The verification and correction tool 134, 240 can analyze the text, selected visual elements, and voice data entered by the user to organize the correction requests and then compare them with existing data to identify the parts that need to be changed. For example, if a user enters, "The accident occurred at 4:10 PM, not 3:50 PM," the verification and correction tool 134, 240 can compare the existing data to organize the corrections and update the corresponding information in the scenario script. The verification and correction tool 134, 240 can also use natural language processing (NLP) technology to search for and correct sentences that need to be changed in the existing scenario. For example, the sentence "An SUV crashed into a car from behind" can be updated to "An SUV crashed into a car waiting at a traffic light from behind at 4:10 PM." For this purpose, the verification and correction tools 134, 240 can leverage NLP models such as BERT, T5, and GPT-3.

[0290] According to an embodiment of the present invention, the verification and correction tool 134, 240 can identify specific scenes (Scene IDs) of a video that require modification based on an existing scenario script and metadata. For example, if a user requests, "The color of the SUV vehicle must be changed from black to white," the verification and correction tool 134, 240 can analyze all scenes in which the vehicle color appears and identify frames that require modification. The verification and correction tool 134, 240 then updates the metadata to reflect the changes, generating data such as "Scene_04: Change vehicle color - black → white." In this case, the verification and correction tool 134, 240 can utilize object search models such as YOLO and Faster R-CNN and Scene Segmentation AI (ShotDetect, DeepLab) models.

[0291] Embodiment of video data correction

[0292] According to an embodiment of the present invention, the verification and correction tool 134, 240 can identify scenes that require modification and then modify the video data based on the identified scenes. The verification and correction tool 134, 240 can also modify text overlays within the video (such as the time of the accident, location, and vehicle information). For example, the verification and correction tool 134, 240 can change the time of the accident from "3:50 PM" to "4:10 PM." The verification and correction tool 134, 240 can also utilize AI models such as StyleGAN, Pix2PixHD, and SPADE to modify object attributes such as vehicle color and road environment. For example, if a request to change the color of a vehicle is received, the verification and correction tool 134, 240 can analyze the scene, change the vehicle color from black to white, and then reflect this in the video. Furthermore, if the weather information is incorrect, the verification and correction tool 134, 240 can change the road environment to rainy weather using GAN-based video editing technology.

[0293] Corrected image generation stage

[0294] According to an embodiment of the present invention, the verification and correction tool 134, 240 can ultimately reflect subsequently changed scenarios and video elements and generate a new video based on the modified information. The verification and correction tool 134, 240 can apply a partial update method for only the modified parts in conjunction with the creative creation tool 133, 230, and can provide the updated video to the user after verifying whether the modified scene has been accurately reflected. In addition, the verification and correction tool 134, 240 can also generate a final, completed accident video by correcting text subtitles, time display, weather information, etc. in the video. For example, if a user views a video generated by the creative creation tool 133, 230 and inputs a correction request such as, "The color of the SUV is white, not black," the verification and correction tool 134, 240 can analyze the request and compare it with existing data to find scenes that need to be changed. The verification and correction tool 134,240 can then use Scene Segmentation AI to find all frames in which the vehicle appears, and use StyleGAN and Pix2PixHD to change the color of the SUV to white.

[0295] Review of the revised video (S430)

[0296] According to one embodiment of the present invention, in the revised video review step (S430), the verification and correction tool 134, 240 can finally verify whether the video generated by reflecting the user's revision requests accurately matches the actual accident situation. At this time, the user reviews the changes, and if there are any additional revision requests, the verification and correction tool 134, 240 can reflect them and generate a final, confirmed video. Specifically, once the revised video is provided, the user can review it again and verify whether the changes have been accurately reflected. The verification and correction tool 134, 240 can provide a video that reflects the user's revision requests and provide a side-by-side comparison function with a previous version so that the user can compare it with the existing video.

[0297] UI Provisioning Implementation Example

[0298] According to an embodiment of the present invention, the verification and correction tool 134, 240 may provide a UI that allows a user to input feedback via text input, visual selection, or voice input if they find a portion that requires additional correction after reviewing a specific scene in a video. For example, if a user inputs, "The color of the SUV has changed to white, but it was actually closer to ivory. Please adjust the color," the verification and correction tool 134, 240 may compare the existing correction details and analyze the portion that requires additional changes. The verification and correction tool 134, 240 may process the user's input using natural language analysis models such as GPT-4 and BERT, and may use algorithms such as Levenshtein Distance and Siamese Network to compare the existing correction details with the new request.

[0299] Application of changes

[0300] According to an embodiment of the present invention, the verification and correction tool 134, 240 may update existing data to reflect a user's additional correction request and apply the changes. At this time, the verification and correction tool 134, 240 may compare the existing correction details with the new request to verify whether the additional changes conflict with the existing changes. For example, if a user requests a color adjustment from "white to ivory," the verification and correction tool 134, 240 may analyze the existing change details from "black to white" and apply the additional color adjustment. The verification and correction tool 134, 240 may then search for the scene in the video and correct the color values ​​to meet the new request. The verification and correction tool 134, 240 may modify the existing vehicle color data to an ivory tone, using StyleGAN, Pix2PixHD, or SPADE models for this purpose. Additionally, the verification and correction tools 134 and 240 apply Color Transfer AI techniques to correct pixel colors within a scene, allowing a color similar to the actual ivory color to be expressed.

[0301] Modified Video Presentation Embodiments

[0302] According to an embodiment of the present invention, the verification and correction tool 134, 240 may provide the final corrected video to the user after the changes are reflected, allowing the user to review the video. The verification and correction tool 134, 240 may configure a UI that allows the user to review the changes and select "corrections complete" or "request additional changes" after the final video is provided. If the user does not request additional changes, the verification and correction tool 134, 240 stores the finalized video in a database for use in subsequent legal consultations, insurance reviews, accident analysis, etc. At this time, the verification and correction tool 134, 240 may utilize the Siamese Network for final comparison and review, and a Super Resolution AI (ESRGAN) model for video quality correction.

[0303] For example, if a user views a video generated by the verification and correction tool 134, 240 and inputs a correction request, such as, "The color of the SUV has changed from black to white, but it was actually closer to ivory," the verification and correction tool 134, 240 analyzes the request and compares it with existing data to find scenes that need to be changed. The verification and correction tool 134, 240 then uses Scene Segmentation AI to find all frames in which the vehicle appears, and can correct the color of the SUV to ivory using StyleGAN and Pix2PixHD. Finally, the corrected video is provided to the user, who reviews the changes and then gives their final approval.

[0304] FIG. 9a is a diagram showing a scene in which a landlord and a tenant meet and greet each other on May 1, 2024 in Happy Real Estate, which is created by a creation creation tool according to an embodiment of the present invention.

[0305] The present invention provides visualization results in various forms, such as 3D images, 2D scenario animations, and cartoon-style images, to more intuitively convey the facts of a case to users. These diverse visualization methods are not merely formal alternatives; they maximize information transmission and user comprehension by selectively providing the most appropriate form based on the case type and the characteristics of the interpreter. For example, in cases where spatial layout, vehicle movement path, and collision location are important, such as traffic accident disputes, 3D-based visualization is effective for spatially reconstructing and simulating the on-site situation. On the other hand, in cases where document-based dialogue and the flow of statements are important, such as lease agreements, medical consent procedures, and insurance policy disputes, cartoon-style and 2D scenario animations, which can naturally convey the emotions, lines, and contextual flow of characters, are more suited to user understanding.

[0306] In this way, the present invention is designed to enable strategic selection of visualization methods according to case type, and is not limited to a single visual representation, but has a structure that allows repeated generation and supplementation of user-customized visualization results, thereby enhancing usability and scalability in practical situations such as consultation, dispute resolution, and case explanation. The functions of the creation creation tools 133 and 230 are described below.

[0307] Creative creation tools133,230

[0308] The creation creation tools 133 and 230 of the present invention can perform the function of generating a storyboard and character dialogue in the form of a cartoon based on the results of the important information extraction step (S110) extracted by the factual relationship extraction tools 131 and 210.

[0309] Fact Extraction Tools 131,210

[0310] According to one embodiment of the present invention, if a user inputs a statement such as "I signed a contract with Happy Real Estate," the factual relationship extraction tool 131, 210 can analyze the statement and generate structured data. At this time, the input text may be divided into sentence units, and unnecessary words such as "You did..." may be removed. The factual relationship extraction tool 131, 210 can then tag time individual information 2-1-1, location individual information 2-1-2, person individual information 2-1-7, etc. using a Named Entity Recognition (NER) technique. For example, the factual relationship extraction tool 131, 210 can classify the statement input by the user as follows: time individual information 2-1-1 is "May 1, 2024," location individual information is "Happy Real Estate," and person individual information is "landlord, tenant, real estate agent." The factual relationship extraction tool 131, 210 can also extract the content spoken by each person and organize it using dialogue information. For example, we can organize information such as the landlord saying, "Hello, my house is very livable. Have you looked around carefully with the real estate agent?", the tenant saying, "The house looks very livable. I'm glad I found a good tenant," and the real estate agent saying, "Thank you for using our real estate agency. We'll prepare the contract for you!"

[0311] Scenario generation tool 132,220

[0312] According to one embodiment of the present invention, the subsequent scenario generation tool 132, 220 can generate a summary of the facts. In this case, the scenario generation tool 132, 220 can divide scenes into scenes taking into account the flow of dialogue and set the behavior and emotions of characters in each scene. For example, the scenario generation tool 132, 220 can generate a scene in which three people gather at the Happy Real Estate Company in Scene 1 of FIG. 9a, a scene in which the landlord and tenant interact in Scene 2 of FIG. 9b, and a scene in which the Happy Real Estate Agent proceeds with the contract in Scene 3 of FIG. 9c. The scenario generation tool 132, 220 can also set the facial expressions and behaviors of each character, so that Landlord A can greet Landlord A with a friendly expression and an outstretched hand, Tenant B can greet Landlord A with a bright smile, and Happy Real Estate Agent C can appear welcoming.

[0313] Scene-Based Generation Embodiments

[0314] According to an embodiment of the present invention, the creation generation tool 133, 230 can generate cartoon scenes based on a scenario script. The creation generation tool 133, 230 can generate characters and backgrounds. The creation generation tool 133, 230 can analyze an input scenario and generate cartoon-style character images that reflect the role and characteristics of each character. For example, the landlord can be represented as a middle-aged woman dressed in neat clothes, the tenant as a young man dressed in casual clothes, and the real estate agent as a young woman dressed in a suit. The creation generation tool 133, 230 can also generate a real estate office with a sign for "Happy Real Estate" and add detailed elements such as a window and a table with a contract on it. The creation generation tool 133, 230 can generate characters and backgrounds using deep learning models such as Stable Diffusion and DALL·E 3, and create cartoon-style character images using StyleGAN 3 and Toonify AI.

[0315] Embodiment of speech bubble generation

[0316] According to an embodiment of the present invention, the creative product generation tool 133, 230 may convert the generated scene into a cartoon style and add speech bubbles to complete the final output. Specifically, the creative product generation tool 133, 230 may first apply a Toonify filter to enhance the colors and contours of the cartoon style and adjust the clarity of the character and background. The creative product generation tool 133, 230 may then insert the character's lines into the speech bubbles using OCR-based AI. The creative product generation tool 133, 230 may position the speech bubbles near the character's face and apply a font specifically designed for cartoons to improve readability. Furthermore, the creative product generation tool 133, 230 may adjust the character's emotional expressions naturally using PoseGAN and GANimator, for example, to make the tenant appear to smile and hold out their hand, or to appear to be impressed.

[0317] According to the present invention, the generated cartoon is configured to allow a user to intuitively understand the facts and can communicate the flow of events more effectively than a text-based explanation. For example, the final output cartoon may depict a scene in Scene 1, with a sign for Happy Real Estate visible in the background, in which a real estate agent welcomes the tenant and renter, saying, "Thank you for using our real estate service. We'll prepare the contract for you!" The tenant extends his hand and says, "Hello, my house is comfortable to live in. Have you had a good look around with the real estate agent?" The tenant smiles and says, "The house looks very comfortable to live in. I'm glad I found a good tenant."

[0318] The creative work generation tools 133, 230 of the present invention can convert the information collected by the fact extraction tools 131, 210 into a cartoon format for intuitive understanding. This allows users to understand the flow of the case more easily than with text-based explanations and intuitively review any necessary corrections. Furthermore, the cartoon-style dialogue representation can be applied to various fields such as legal consultations, education, and insurance reviews, thereby overcoming the limitations of existing text-centric case analysis methods. In other words, the creative work generation tools 133, 230 can utilize a GPT-based NLP model, a GAN-based image conversion model, and an OCR-based text analysis model, thereby effectively visualizing the contract process, accident occurrence status, legal consultation content, etc.

[0319] FIG. 9b is a diagram showing a dialogue scene between a lessor and a lessee generated by a creation creation tool according to an embodiment of the present invention.

[0320] The creation creation tools 133 and 230 of the present invention can perform a function of expressing a negotiation scene between Lessor A and Lessee B in the form of a cartoon based on the information analyzed by the fact extraction tools 131 and 210. For example, the user can say, "I, the lessor, would like to rent to the lessee at Happy Real Estate. How about proceeding with the contract for a 30 million won deposit and 600,000 won rent? Considering the condition of my house, 600,000 won rent is low compared to the market price." After hearing my explanation, the lessee said, "I understand that 600,000 won rent is low compared to the market price. However, the deposit is a burden. Is it possible to conclude the contract for a 20 million won deposit and 650,000 won rent?" The user can then say, "I wanted to receive a larger deposit, but Lessee B felt it was a burden. In the end, I decided to accept the proposal, taking Lessee B's burden into consideration."

[0321] Fact Extraction Tools 131,210

[0322] According to an embodiment of the present invention, the fact extraction tool 131, 210 may analyze the content of a conversation entered by a user and extract case individual information. The fact extraction tool 131, 210 may divide the user's input into sentence units and extract only case individual information by removing unnecessary emotional expressions such as "That's really unfortunate." In this case, the fact extraction tool 131, 210 may tag time individual information, place individual information, and person individual information using a Named Entity Recognition (NER) model. Through this, the fact extraction tool 131, 210 may organize the flow of a conversation in which Lessor A and Lessee B are negotiating a contract at a place called Happy Real Estate and adjusting the deposit and rent amount. The fact extraction tool 131, 210 may extract, as second case individual information, important elements of the conversation, such as the draft terms of a 30 million won deposit and a 600,000 won rent, and the terms changed to a 20 million won deposit and a 650,000 won rent at the tenant's suggestion. In addition, the fact extraction tools 131 and 210 can also reflect emotional elements, such as Lessor A wanting to receive a larger deposit but accepting Lessee B's proposal in consideration of the burden on Lessee B, in the second case individual information.

[0323] An embodiment of cartoon generation

[0324] According to one embodiment of the present invention, the creative product generation tool 133, 230 can visually represent a negotiation scene between a landlord and a tenant based on a scenario script. The creative product generation tool 133, 230 can generate cartoon-style characters, with Landlord A dressed neatly and with a careful expression, and Tenant B dressed casually and with a persuasive expression. The creative product generation tool 133, 230 can change the character's facial expression according to the flow of the dialogue, so that Landlord A maintains a calm expression when first proposing the terms of the contract, changes to a worried expression when Tenant B requests an adjustment to the security deposit, and eventually sighs and accepts.

[0325] FIG. 9c is a diagram showing a brokerage scene of a happy real estate agent generated by a creation generation tool according to an embodiment of the present invention.

[0326] The creative work generation tools 133, 230 of the present invention can perform the function of expressing a negotiation scene between a landlord and a tenant in the form of a cartoon based on the information analyzed by the fact extraction tools 131, 210. The scene in Figure 9c includes a scene in which Happy Real Estate Agent C is arbitrating a contract, and Landlord A wants to receive a larger deposit but is willing to listen to the agent's explanation for the time being, while Tenant B actively proposes a condition to reduce the deposit and increase the rent. In this case, the creative work generation tools 133, 230 can generate a cartoon that reflects the emotions and attitudes in such a conversation, and to achieve this, natural language processing (NLP), emotion analysis, image generation, and style conversion AI can be combined.

[0327] Example of Named Entity Recognition (NER) model utilization

[0328] According to an embodiment of the present invention, the creative creation tool 133, 230 may perform a natural language processing (NLP)-based analysis to analyze a user's dialogue and generate a storyboard. The creative creation tool 133, 230 may divide the user's input into sentence units and tag the characters (landlord A, tenant B, agent C) and main content using a Named Entity Recognition (NER) model. Through this, the creative creation tool 133, 230 may understand the content of the agent's mediation, "The tenant wants to lower the security deposit and increase the rent," and recognize that the landlord's response is not simply positive but is reluctant but willing to listen. To this end, the creative creation tool 133, 230 may utilize a sentiment analysis model. For example, the creative creation tool 133, 230 may classify emotional elements in text using an NLP model such as BERT or RoBERTa and predict emotional changes among characters as the dialogue flows using a model such as DialogueRNN.

[0329] Facial Expression and Pose Generation Embodiments

[0330] According to an embodiment of the present invention, the creative product generation tool 133, 230 may generate facial expressions and poses to visually express the emotions and attitudes of a character. Because Lessor A wants to receive an increased security deposit, the creative product generation tool 133, 230 may generate facial expressions that show her looking slightly worried but listening attentively to the agent's explanation. On the other hand, the creative product generation tool 133, 230 may generate facial expressions that show Lessee B making a proactive proposal with a confident expression, and Intermediary C maintaining a neutral and persuasive facial expression and pose. To this end, the creative product generation tool 133, 230 may convert text-based emotions into visual expressions using DeepFace, AffectNet, or FER+ as emotion analysis models, and may generate poses using PoseGAN and OpenPose to adjust the character's gestures and postures. Furthermore, the creative product generation tool 133, 230 may generate natural facial expressions and movements according to emotional changes using GANimator or EmoGAN.

[0331] Example of office interior background generation

[0332] According to an embodiment of the present invention, the creative creation tool 133, 230 may generate a background for the interior of a real estate office using an image generation model such as Stable Diffusion and DALL·E 3. The creative creation tool 133, 230 may also convert to a cartoon style using StyleGAN 3 and Toonify AI, and apply specific styles and perform detailed corrections using SPADE and Pix2PixHD models. The creative creation tool 133, 230 may convert characters and backgrounds to a cartoon style and perform adjustments to maintain clear contours and simple colors. The creative creation tool 133, 230 may naturally position the position of broker C holding the contract, and the relative sizes and positions of seated tenant A and tenant B, creating a balanced scene.

[0333] Embodiment of speech bubble arrangement

[0334] According to an embodiment of the present invention, the creative product generation tool 133, 230 automatically arranges speech bubbles and dialogue. Specifically, the creative product generation tool 133, 230 can appropriately arrange the size and position of speech bubbles taking into account the character's face and speaking order, recognize dialogue text using a model such as Tesseract OCR or EasyOCR, and determine the optimal speech bubble arrangement using a model such as DeepLayout or LayoutLM. The creative product generation tool 133, 230 can also add a voice assistance function using a Text-to-Speech (TTS) model such as Tacotron2. In particular, in this scene, the agent's dialogue is relatively long, so the agent's speech bubble should be arranged in a large size, and the short responses of the landlord and tenant are arranged in smaller speech bubbles to improve readability.

[0335] FIG. 9d is a diagram illustrating a scene in which a lessor, generated by a creation creation tool according to an embodiment of the present invention, makes a speech requesting that the down payment be made by May 10, 2024.

[0336] According to an embodiment of the present invention, the creation generation tool 133, 230 may perform a function of generating a scene in which Lessor A reluctantly accepts a request for a security deposit adjustment based on information analyzed by the fact extraction tool 131, 210. In the scene of Fig. 9d, Lessor A may sigh and make a sad expression, while Lessee B may smile, emphasizing the scene. In other words, the creation generation tool 133, 230 may perform a function of analyzing emotional elements in the dialogue and naturally expressing the character's expressions and actions.

[0337] Embodiments of storyboard generation reflecting emotions and attitudes

[0338] According to one embodiment of the present invention, the creative product creation tool 133, 230 can analyze a dialogue input by a user and generate a storyboard that reflects the user's emotions and attitudes. Specifically, the creative product creation tool 133, 230 can extract the characters (landlord A, tenant B, agent C) appearing in the dialogue and their emotional states using a natural language processing (NLP)-based emotion analysis model. For example, the creative product creation tool 133, 230 can detect hesitation and worry from the landlord's statement, "Hmm... I guess you want me to lower the security deposit," and acceptance but dissatisfaction from the expression, "I understand." In contrast, the tenant's facial expression is relatively positive, and the creative product creation tool 133, 230 can express the landlord smiling when agreeing to the security deposit adjustment.

[0339] Example of using sentiment analysis models (BERT, RoBERTa, DialogueRNN)

[0340] According to an embodiment of the present invention, the creation creation tool 133, 230 may extract emotional elements from text using an emotion analysis model (BERT, RoBERTa, DialogueRNN) to reflect such emotional expressions, and may generate character expressions using a facial emotion recognition model such as DeepFace, AffectNet, or FER+. In this case, the creation creation tool 133, 230 may generate an expression for Lessor A in which the eyebrows are slightly raised or frowning, or the lips are kept in a straight line, as if sighing. In contrast, the creation creation tool 133, 230 may generate an expression for Lessee B in which the mouth is slightly raised and the mouth is smiling.

[0341] An embodiment of character pose adjustment

[0342] According to an embodiment of the present invention, the creative product generation tool 133, 230 can adjust the poses of characters naturally. Specifically, the creative product generation tool 133, 230 can use a pose generation model such as PoseGAN or OpenPose to make Lessor A appear to sigh with his shoulders slightly slumped and one hand raised, and Lessee B appear to sit with a relaxed posture and smile. Furthermore, the creative product generation tool 133, 230 can maintain the appearance of Agent C standing in a neutral posture and explaining the terms of the contract. In other words, the creative product generation tool 133, 230 can learn such pose data and generate gestures appropriate for the emotions and lines of each character.

[0343] Background and Character Styling Embodiments

[0344] According to an embodiment of the present invention, the creative product generation tool 133, 230 may apply background and character styling. Specifically, the creative product generation tool 133, 230 may generate a background for a real estate office using an image generation model such as Stable Diffusion or DALL·E 3, and convert it into a cartoon style using StyleGAN 3 or Toonify AI. In addition, the creative product generation tool 133, 230 may optimize the character style to a cartoon format using SPADE or Pix2PixHD models, and perform adjustments to clearly show the character's facial expressions and poses. In this case, the creative product generation tool 133, 230 may add background elements such as a table with a contract on it, windows and decorations that create the atmosphere of the office interior, etc., to enhance the sense of realism.

[0345] Embodiment of speech bubble arrangement

[0346] According to an embodiment of the present invention, the creative work generation tool 133, 230 may perform the task of arranging dialogue and speech bubbles. Specifically, the creative work generation tool 133, 230 may recognize text using Tesseract OCR or EasyOCR, and naturally arrange dialogue in speech bubbles using DeepLayout or LayoutLM. In the scene shown in FIG. 9d, the speech bubble for landlord A is the largest, while a smaller speech bubble may be appropriate for tenant B, whose reaction is brief. In addition, the creative work generation tool 133, 230 may add an oval dotted line or a blurred font effect to the inside of the speech bubble for the tenant to emphasize hesitation, such as "Hmm..."

[0347] FIG. 9e is a diagram illustrating a scene in which a tenant and a happy real estate agent speak in response to a tenant's utterance generated by a creation creation tool according to an embodiment of the present invention.

[0348] Embodiment of character facial expression generation

[0349] As shown in Figure 9e, the creative creation tools 133, 230 can detect emotions including surprise and worry from Lessee B's words ("Oh... 10 million won? That's a lot of money") and analyze the calm and persuasive attitude from Agent C's words ("5 million won is a reasonable contract amount.") by utilizing natural language processing (NLP)-based emotion analysis models such as BERT, RoBERTa, and DialogueRNN. In addition, the creative creation tools 133, 230 can express Lessor A as if he is making a worried expression even though he does not say anything.

[0350] The creation generation tools 133 and 230 can then generate character expressions based on the emotion analysis results. Specifically, the creation generation tools 133 and 230 can use facial emotion recognition models such as DeepFace, AffectNet, and FER+ to set each character to have an appropriate expression. For example, the creation generation tools 133 and 230 can create an expression where tenant B has round, wide eyes and a slightly open mouth, broker C has a calm expression, and tenant A has a worried expression. To this end, the creation generation tools 133 and 230 can use models such as GANimator and EmoGAN to naturally generate facial expressions and subtle gesture changes that accompany emotional changes.

[0351] Pose adjustment embodiment

[0352] According to an embodiment of the present invention, the creation generation tool 133, 230 can adjust the poses of characters. Specifically, the creation generation tool 133, 230 can use a pose generation model such as PoseGAN or OpenPose to create a pose where tenant B raises his hand to his chest in a worried pose, where agent C spreads his hands in a persuasive motion, and where tenant A folds his arms or places his hand on his chin in a worried pose. In this case, the creation generation tool 133, 230 can adjust the positions and interactions between characters so that they appear natural, taking into account emotions and the flow of dialogue.

[0353] Figure 9f is a diagram showing a scene in which a lessor accepts a tenant's utterance generated by a creation creation tool according to an embodiment of the present invention. Figure 9g is a diagram showing a scene in which a lessor and tenant create a lease agreement generated by a creation creation tool according to an embodiment of the present invention. Figure 9h is a diagram showing the final scene of a cartoon generated by a creation creation tool according to an embodiment of the present invention.

[0354] The scope of the present invention is not limited to the above-described embodiments, but may be embodied in various forms within the scope of the appended claims. Any person skilled in the art to which the invention pertains may make various modifications without departing from the gist of the present invention as defined in the claims. [Explanation of symbols]

[0355] 10: Fact-based creative creation system 100: User terminal 110: Sensor unit 120: Communications Department 130: Processor section 131: Fact Extraction Tools 132: Scenario generation tool 133: Creation generation tools 134: Verification and correction tools 140: Memory section 150: Output section 200: Server device 210: Fact Extraction Tools 220: Scenario generation tool 230: Creation generation tools 240: Verification and correction tools

Claims

1. In a creation creation device based on a database of lawsuits and cases by type and facts input by a user, a fact extraction tool configured to extract case individual information based on at least one of audio data and text data related to the facts; a scenario generation tool configured to generate a text-based scenario script based on the extracted incident individual information; a creation creation tool configured to create a creation based on the generated text-based scenario script; a validation and correction tool configured to receive feedback input from the user based on the generated creation; The scenario generation tool includes: The database of lawsuits and cases is configured to compare the extracted individual case information with the database of lawsuits and cases; and determining information leaks for constructing the text-based scenario script based on the comparison. A device for generating creative works based on a database of lawsuits and cases categorized by type and facts input by the user.

2. The incident individual information is The information is configured to include at least one of time individual information related to the factual relationship, place individual information related to the factual relationship, vehicle individual information related to the factual relationship, accident type individual information related to the factual relationship, victim individual information related to the factual relationship, situation individual information related to the factual relationship, person individual information related to the factual relationship, and behavior individual information related to the factual relationship, 10. A device for creating a creative work based on the database of lawsuits and cases classified by type and facts input by a user according to claim 1.

3. The fact extraction tool is configured to generate a query to extract the case individual information; The scenario generation tool includes: and generating additional questions to present to the user to supplement the identified leak information.

10. A device for creating a creative work based on the database of lawsuits and cases classified by type and facts input by a user according to claim 1.

4. The fact extraction tool and the scenario generation tool are configured to utilize a natural language processing model to generate a question or follow-up question for the user; 4. A device for creating a creative work based on the database of lawsuits and cases classified by type and facts input by a user according to claim 3.

5. The creation creation tool comprises: and generating an image that visually embodies the generated text-based scenario script by utilizing a deep learning-based image generation model.

10. A device for creating a creative work based on the database of lawsuits and cases classified by type and facts input by a user according to claim 1.

6. The creation creation tool comprises: By utilizing a deep learning-based image generation model, configured to generate at least one of an outer shape of a person in the video, a facial expression of a person in the video, a posture of a person in the video, and a movement of a person in the video; 6. A device for creating a creative work based on the database of lawsuits and cases classified by type and facts input by a user according to claim 5.

7. The creation creation tool comprises: The script is configured to convert lines or narration information contained in the text-based scenario script into voice.

6. A device for creating a creative work based on the database of lawsuits and cases classified by type and facts input by a user according to claim 5.

8. The creation creation tool comprises: The system is configured to perform emotion-aware TTS (Text-to-Speech) to reflect emotions in the converted voice.

8. A device for creating a creative work based on the database of lawsuits and cases classified by type and facts input by a user according to claim 7.

9. The creation creation tool comprises: Configured to generate sound effects within the video using a deep learning-based audio synthesis model, 6. A device for creating a creative work based on the database of lawsuits and cases classified by type and facts input by a user according to claim 5.

Images