Video copywriting generation method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a video copywriting generation method and device, equipment and a storage medium, and the method comprises the steps: obtaining a to-be-processed video, and determining a minute-level abstract of the to-be-processed video and line information corresponding to the minute-level abstract; generating a marketing reference copywriting of the to-be-processed video based on the minute-level abstract and the line information by using a pre-trained first large language model; according to the embodiment of the invention, the first large language model is trained based on the samples including the marketing subject samples in the video marketing sample videos, and the first large language model is utilized to automatically generate the marketing reference copywriting of the to-be-processed video. Therefore, the marketing reference copywriting can be automatically provided as the creation inspiration and creation direction reference of the marketing copywriting when the user performs video marketing short video creation, thereby improving the video marketing short video creation efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a video script generation method and device, an electronic device and a storage medium. BACKGROUND

[0002] With the rise of short video platforms, the creation and dissemination of marketing short videos of videos such as films and television dramas play an increasingly important role in film and television marketing. For example, through video marketing short videos (also known as seeding short videos), key scenes of a film and television drama are edited and spliced and promoted to attract the viewing interest of the audience, thereby driving the viewing traffic of the film and television drama.

[0003] However, the content creation of video marketing short videos needs to fully consider the essence of the plot of different videos and the interest points of the audience, and users need to spend a lot of effort to think and summarize the marketing scripts of video marketing short videos, resulting in a low creation efficiency of video marketing short videos, which is difficult to meet the creation needs of a large number of video marketing short videos. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide a video script generation method to solve the problem of how to automatically generate marketing reference scripts of videos, so as to provide creation inspiration and creation direction reference for the marketing scripts when users create video marketing short videos, and thereby improve the creation efficiency of video marketing short videos. The specific technical solutions are as follows:

[0005] In the first aspect of the present application, a video script generation method is first provided, which comprises:

[0006] obtaining a to-be-processed video;

[0007] determining a minute-level abstract of the to-be-processed video and dialogue information corresponding to the minute-level abstract; wherein the minute-level abstract is a plot of each minute of the video described in the form of text;

[0008] generating a marketing reference script of the to-be-processed video based on the minute-level abstract and the dialogue information by using a pre-trained first large language model; wherein the marketing reference script is used to provide reference information for users when creating a marketing video of the to-be-processed video;

[0009] wherein the first large language model is obtained by training based on a marketing theme sample in a video marketing sample video, a minute-level abstract sample in an original video corresponding to the video marketing sample video and matching the video marketing sample video, dialogue information samples in the original video corresponding to the minute-level abstract sample and a plot description sample corresponding to the video marketing sample video.

[0010] In a second aspect of the present invention, a video copy generation device is provided, comprising:

[0011] A module for obtaining videos to be processed, used to obtain videos to be processed;

[0012] A determination module, configured to determine a minute-level summary of the video to be processed and line information corresponding to the minute-level summary; wherein the minute-level summary is a textual description of the plot of each minute of the video;

[0013] a marketing reference copy generation module, utilizing the pre-trained first language model to generate a marketing reference copy for the video to be processed based on the minute-level summary and the line information; wherein the marketing reference copy is used to provide reference information for users when creating a marketing video for the video to be processed;

[0014] Among them, the first language model is trained based on marketing theme samples in video marketing sample videos, minute-level summary samples in the original video corresponding to the video marketing sample videos that match the video marketing sample videos, dialogue information samples in the original video that match the minute-level summary samples, and plot description samples corresponding to the video marketing sample videos.

[0015] In the third aspect of the implementation of the present invention, an electronic device is also provided, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement any of the above-mentioned video copy generation methods when executing the program stored in the memory.

[0016] In a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium is run on the computer, the computer executes any of the above-mentioned methods for generating video text.

[0017] In a fifth aspect of the implementation of the present invention, a computer program product comprising instructions is further provided, which, when executed on a computer, enables the computer to execute any of the above-described methods for generating video text.

[0018] The video copy generation method provided by the embodiment of the present invention trains a first language model based on samples including marketing theme samples in video marketing sample videos, and uses the first language model to automatically generate marketing reference copy for the video to be processed, so that when the user creates a video marketing short video, the marketing reference copy can be automatically provided as a reference for the creation inspiration and creation direction of the marketing copy, thereby improving the efficiency of video marketing short video creation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced.

[0020] Figure 1 is a step flow chart of a video script generation method embodiment of the present application;

[0021] Figure 2 is an original video matching flow chart example of a video script generation method embodiment of the present application;

[0022] Figure 3 is a whole execution flow chart schematic diagram of a video script generation method embodiment of the present application;

[0023] Figure 4 is a structure block diagram of a video script generation device embodiment of the present application;

[0024] Figure 5 is a structure schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.

[0026] Referring to Figure 1 is a step flow chart of a video script generation method embodiment of the present application, including the following steps:

[0027] Step 101, obtaining a video to be processed;

[0028] In step 101, the video to be processed can be obtained first, which can be a movie, a TV series, etc., and the present application does not make specific limitation on this.

[0029] Step 102, determining a minute-level summary of the video to be processed and script information corresponding to the minute-level summary; wherein the minute-level summary is a plot of each minute of the video described in the form of text;

[0030] The minute-level summary of the video can describe the plot of each minute of the video in the form of text, providing basic data for applications such as plot search and recommended script generation. For example, the plot of each single episode of a TV series is generally pre-stored in the form of a minute-level summary, which can describe the plot in different scenes in the form of text. As shown in Table 1, it is an example of a video minute-level summary record format:

[0031] Table 1

[0032] time Abstract <![CDATA[(t1,t2)]]> Abstract 1 <![CDATA[(t3,t4)]]> Abstract 2 … …

[0033] The minute-level summary generally contains the time period information (for example, start and end time stamps) of the plot to which it belongs in the video, so the corresponding dialogue information of the minute-level summary can be determined according to the time period information. For example, the time stamp information in the time period information can be compared with the time stamp information corresponding to the dialogue data in the video dialogue file (for example, a subtitle file or a speech recognition file) to find a dialogue paragraph with the same start and end time stamp range in the dialogue data. Then, all dialogue texts within the time stamp information range can be automatically extracted in the dialogue data and spliced into a paragraph text as the corresponding dialogue information of the minute-level summary.

[0034] In step 102, since the minute-level summary and the corresponding dialogue information of the minute-level summary can fully represent the complete plot of the video and are highly related to the content to be expressed by the video, when there is a creation demand for a video marketing short video, one or more minute-level summaries can be manually selected by the user from the to-be-processed video, and the time period information corresponding to the minute-level summary in the original video is determined. Then, the time stamp information in the time period information can be compared with the time stamp information corresponding to the dialogue data in the original video dialogue file (for example, a subtitle file or a speech recognition file) to find a dialogue paragraph with the same start and end time stamp range in the dialogue data. Then, all dialogue texts within the time stamp information range can be automatically extracted in the dialogue data and spliced into a paragraph text as the corresponding dialogue information of the minute-level summary, so as to automatically generate a marketing reference script of the video in the subsequent steps to provide the user with the creation inspiration and creation direction reference of the marketing script.

[0035] Without manual operation of the user, a to-be-processed video can be randomly selected, and one or more minute-level summaries can be randomly selected from the selected to-be-processed video. The corresponding dialogue information of the minute-level summary is determined according to the minute-level summary in the video, so as to automatically generate a marketing reference script of the video in the subsequent steps to provide the user with a large amount of creation inspiration and creation direction reference of the marketing script in advance.

[0036] In step 103, a first large language model is used to generate a marketing reference script of the to-be-processed video based on the minute-level summary and the dialogue information. The marketing reference script is used to provide reference information for the user when creating a marketing video of the to-be-processed video.

[0037] The first large language model is trained based on a marketing theme sample in a video marketing sample video, a minute-level summary sample in an original video corresponding to the video marketing sample video and matching the video marketing sample video, dialogue information samples in the original video matching the minute-level summary sample, and a plot description sample corresponding to the video marketing sample video.

[0038] In some examples, the marketing video of the video to be processed may be a short video, and the video length of the marketing video of the video to be processed may be shorter than the video length of the video to be processed.

[0039] In step 103, since the marketing copy of a video marketing video is generally the core factor in attracting the audience's interest in watching, and the user needs to ensure that the content of the video marketing video accurately corresponds to the plot in the original video when creating a video marketing video, the marketing theme samples in the video marketing sample video, the minute-level summary samples in the original video corresponding to the video marketing sample video that match the video marketing sample video, the lines information samples in the original video that match the minute-level summary samples, and the plot description samples corresponding to the video marketing sample video can be used to form a training data set, and the first largest language model is trained to obtain the first largest language model, so as to use the first largest language model to take the minute-level summary and lines information of the video to be processed as input data to generate the marketing reference copy for the video to be processed.

[0040] In an embodiment of the present invention, the video marketing sample video can be an existing popular video marketing video with a high playback volume. Since the marketing theme samples in the video marketing sample video are fully utilized in the training process of the first language model, and the minute-level summary samples, the lines information samples matching the minute-level summary samples, and the plot description samples corresponding to the video marketing sample video are covered, which are data that can represent the original video plot content in different dimensions; it can fully grasp the patterns and rules of marketing copy creation in the video marketing sample video, so that according to the minute-level summary and lines information of the video to be processed, a marketing reference copy that is closely related to the plot content of the video to be processed and meets the audience's interests can be generated, thereby realizing automatic provision of marketing reference copy as a reference for creative inspiration and creative direction of marketing copy when the user creates a video marketing short video, thereby improving the efficiency of video marketing short video creation.

[0041] In an optional embodiment of the present invention, the marketing reference copy includes a marketing reference theme and a plot description, and the method further includes:

[0042] Obtain a video clip of the video to be processed; generate an audio commentary corresponding to the video clip based on the plot description, and add relevant content of the marketing reference theme to the video clip; generate a marketing video of the video to be processed based on the video clip and the audio commentary.

[0043] In this embodiment, the marketing reference copy generated by the first language model can be used to provide users with creative inspiration and reference for the creative direction of the marketing video of the video to be processed, and can also be directly used to generate a marketing video for the video to be processed.

[0044] Specifically, the corresponding video clips can be firstly captured from the video to be processed, and then the commentary audio corresponding to the video clips can be generated based on the plot description in the marketing reference copy by using audio generation technology.

[0045] In practical applications, since the marketing copy in a video marketing video generally attracts the audience's viewing interest in the form of a marketing theme, relevant content of the marketing reference theme can be added to the video clip.

[0046] After obtaining the commentary audio and processed video clips, a marketing video of the video to be processed can be generated, which enables the direct generation of marketing videos based on marketing reference copy. Marketing videos can be automatically generated in large quantities without human intervention, further improving the production efficiency of marketing videos.

[0047] In an optional embodiment of the present invention, adding content related to the marketing reference theme to the video clip includes:

[0048] Add text corresponding to the marketing reference theme to the video title of the video clip, and / or add audio corresponding to the marketing reference theme to the commentary audio of the video clip.

[0049] Marketing reference themes can be presented in marketing videos using at least two methods: text and audio. Specifically, the text corresponding to the marketing reference theme can be used as the video title and added to the video clip, or the text corresponding to the marketing reference theme can be converted into audio and added to the audio commentary of the video clip.

[0050] For example, based on the marketing reference theme, you can add a video title that is closely related to the plot of a vertical screen film and television clip above it. Alternatively, you do not need to add a video title directly to the short video, but instead introduce the marketing reference theme at the beginning of the video through dubbing and commentary. Alternatively, you can add a video title that is closely related to the plot of the clip above the video and further add content such as commentary. In this case, the marketing copy is represented as a combination of the marketing theme and the commentary.

[0051] In an optional embodiment of the present invention, the marketing reference copy includes a marketing reference theme and a plot description. The generating of the marketing reference copy for the to-be-processed video based on the minute-level summary and the line information using the pre-trained first language model includes:

[0052] Utilizing the pre-trained first language model, based on the minute-level summary and the line information, a marketing reference theme and plot description of the video to be processed are generated; wherein, the marketing reference theme is used to provide users with reference information on the marketing theme when creating a marketing video for the video to be processed.

[0053] In actual applications, the first language model can be used to generate a marketing reference theme for the video to be processed based on the minute-level summary and dialogue information as a reference for creating the marketing theme in the marketing copy (i.e., as a reference marketing direction for the video to be processed), and a plot description for the video to be processed can be generated simultaneously. The plot description can correspond to the original plot to which the minute-level summary input into the first language model belongs, thereby providing users with a corresponding plot description while providing a marketing reference theme. The plot description can be used as a dubbing or commentary reference copy for the short video so that users can create dubbing or commentary copy based on the plot content, or it can be directly used as the dubbing or commentary copy for the short video.

[0054] On this basis, the marketing reference theme and plot description can be combined to form a marketing reference copy, which can serve as a reference for the creative inspiration and creative direction of the marketing copy, ensuring that users create marketing copy that is closely related to the plot content of the video to be processed and in line with the audience's interests.

[0055] In one example, the format of the marketing reference copy can be a table consisting of a marketing reference theme and a plot description, so that the marketing reference copy can be displayed intuitively to users, thereby improving the efficiency of creating video marketing videos.

[0056] In an optional embodiment of the present invention, the generating of a marketing reference theme and plot description of the video to be processed based on the minute-level summary and the dialogue information using a pre-trained first language model includes:

[0057] Writing the minute-level summary and the dialogue information into a preset first prompt template to obtain a first prompt text; wherein the first prompt template at least includes a prompt theme, and the prompt theme is used to prompt the first language model to output content based on the plot that the audience is interested in;

[0058] The first prompt text is input into the first language model, so that the first language model outputs the marketing reference theme and the plot description according to the first prompt text.

[0059] In the embodiment of the present application, when the marketing reference theme and the plot description of the video to be processed are generated based on the minute-level summary and the dialogue information by using the first large language model, the minute-level summary and the dialogue information can be written into the preset first prompt template as input data, so as to obtain the corresponding first prompt text, and the first prompt text can be input into the first large language model to make the first large language model output the marketing reference theme and the plot description according to the first prompt text; wherein the first prompt template can be a preset text containing a minute-level summary and a dialogue information filling area, and a fixed text can be pre-set as a prompt theme in the preset text, and the prompt theme can prompt the first large language model to output the answer content based on the plot interesting to the audience through the way of questioning or requiring.

[0060] In an example, the setting and application of the first prompt template can refer to the following example:

[0061] {

[0062] "input": "As a film and television drama operator, you need to edit the plot in the drama that has attraction or can arouse the audience's empathy into a short video to attract more people to watch the complete plot.

[0063] The plot material is as follows:

[0064] Plot segment 0:

minute-level summary 0

[0065] Plot segment 1:

minute-level summary 1

[0066] Plot segment 2:

minute-level summary 2

[0067] ====

[0068] Dialogue segment 0:

dialogue 0

[0069] Dialogue segment 1:

dialogue 1

[0070] Dialogue segment 2:

dialogue 2

[0071] ====

[0072] Please analyze and select the above plot material to obtain the short video theme, and give the corresponding plot content of the theme (note that the plot and the theme do not involve actor information, and do not fabricate the drama name), and the answer is given in json format, key is the plot content and the theme.

[0073] "output": "{'plot content', 'theme'}"

[0074] };

[0075] In this example, "As a film and television drama operator, you need to edit the attractive or audience-resonating plots in the series into a short video and release it to attract more people to watch the complete plot" is an example of a prompt topic. It can guide the first language model to generate target content based on the attractive or audience-resonating plots in the series, and further ensure that the first language model can generate marketing reference copy based on full consideration of the content direction that the audience is interested in.

[0076] The embodiment of the present invention provides the first language model with minute-level summaries and dialogue information of the video in a format corresponding to a predefined template, thereby helping the first language model better understand the context and key points of the input content; reducing the first language model's dependence on the quality of the input data, and making the generated results more accurate.

[0077] In an optional embodiment of the present invention, the first language model is trained by the following method:

[0078] Constructing a training dataset based on the marketing theme samples, minute-level summary samples, dialogue information samples, and plot description samples;

[0079] A large language model is trained under the conditions that at least one minute-level summary in the training data set and at least one line information corresponding to the at least one minute-level summary are used as input data, and the marketing reference copy is used as output data to obtain the first large language model.

[0080] In this embodiment, before performing large language model training to obtain the first large language model, a training data set can be constructed based on marketing theme samples, minute-level summary samples, dialogue information samples, and plot description samples; then, using at least one minute-level summary in the training data set and at least one dialogue information corresponding to the at least one minute-level summary as input data, and marketing reference copy as output data, the basic model of the large language model can be trained to generate the first large language model. Specifically, Qwen72b-Chat (Tongyi Qianwen) can be used as the basic model, the training method is Lora (Low-Rank Adaptation), the training set data volume is 6k, and a total of 8 epochs (complete traversal) are trained to obtain the first large language model.

[0081] In an optional embodiment of the present invention, the plot description sample corresponding to the video marketing sample video is determined by the following steps:

[0082] Writing the minute-level summary sample, the line information sample, the marketing theme sample, and the lines in the video marketing sample video into a preset second prompt template to obtain a second prompt text;

[0083] The second prompt text is input into the second largest language model, so that the second largest language model outputs a plot description sample corresponding to the video marketing sample video according to the second prompt text.

[0084] In an embodiment of the present invention, the plot description sample corresponding to the video marketing sample video refers to the plot description corresponding to the film and television content edited by the video marketing sample video, which can be understood as the plot description corresponding to the video marketing sample video itself or the marketing copy of the video marketing sample video, and the plot description is often not provided in the video marketing sample video; therefore, the obtained minute-level summary sample, line information sample, marketing theme sample and the lines edited in the video marketing sample video can be written into a preset second prompt template to obtain the corresponding second prompt text, and the second prompt text is further input into another large language model to output the plot description sample corresponding to the video marketing sample video. Among them, the second prompt template can be a preset text containing a minute-level summary sample filling area, a line information sample filling area, a marketing theme sample filling area, and a line filling area in the video marketing sample video.

[0085] In one example, the setting and application of the second prompt template can refer to the following example:

[0086] ===

[0087] The TV series operators edited some of the plots in the series into short videos for promotion and promotion to attract more people to watch. The edited plots are as follows:

[0088]

Multi-minute summary

[0089] The corresponding lines and dialogues for the plot are as follows:

[0090] [Lines corresponding to multiple minute-level summaries]

[0091] The theme of the short video is: [Short Video Marketing Copy]

[0092] The lines included in the editing section are: [Short video lines]

[0093] Please briefly summarize the plot content of the short video based on the plot and lines, centering on the editing theme, in 50 words or less.

[0094] ===

[0095] The embodiment of the present invention generates plot description samples corresponding to video marketing sample videos based on minute-level summary samples, dialogue information samples, marketing theme samples, and dialogues in video marketing sample videos. It adds plot description samples corresponding to video marketing sample videos as additional training data on the basis of the existing training data of the first language model, constructs a more comprehensive training data set for the first language model, improves the training effect of the first language model, and further improves the generation effect when using the first language model to generate plot descriptions for the video to be processed.

[0096] In an optional embodiment of the present invention, the original video corresponding to the video marketing sample video is determined by the following steps:

[0097] The lines of the video marketing sample video are matched with the lines of the target video, and when the number of consecutive lines of text that match the video marketing sample video and the target video exceeds a preset threshold, the target video is determined to be the original video corresponding to the video marketing sample video.

[0098] In this embodiment, when constructing the training data set of the first largest language model, a video marketing sample video can be obtained and the original video corresponding to the video marketing sample video can be determined to further determine the minute-level summary sample and line information sample in the original video as the training sample data of the first largest language model.

[0099] like Figure 2 As shown, in order to obtain the complete plot segment corresponding to the plot in the video marketing sample video in the original video, the plot can be located in the episode and time period by matching the dialogue text of the video marketing sample video with the dialogue text of the target video. The specific process can be:

[0100] 1. Get the time and text of the lines of the entire album of the target video, match the first sentence of the lines in the video marketing sample video with each sentence of the target video lines. If the current sentence of the target video lines cannot be successfully matched during the matching, the first sentence of the lines in the video marketing sample video can be directly matched with the next sentence of the current sentence of the target video lines. If the match is successful (assuming the matched target video lines are numbered as idx), then continue to match the second sentence of the lines in the video marketing sample video with the idx+1th sentence of the target video lines, and so on until no match is found.

[0101] 2. If the number of consecutive dialogue texts matched in step 1 exceeds 5, the positioning is considered successful. If the number of consecutive dialogue texts matched in step 1 does not exceed 5, the positioning is considered unsuccessful, and the matching process needs to be restarted from the first line of the dialogue in the video marketing sample video. When the positioning is successful, the target video can be determined as the original video corresponding to the video marketing sample video, and the complete plot segment corresponding to the plot in the video marketing sample video in the original video can be synchronously located. For example, the timestamp of the first line of dialogue in the video marketing sample video in the original video can be used as the starting timestamp, and the timestamp of the last line of dialogue in the video marketing sample video in the original video can be used as the ending timestamp, and then the plot between the starting timestamp and the ending timestamp of the original video can be determined as the complete plot segment corresponding to the plot in the video marketing sample video in the original video.

[0102] by Figure 2 As an example, although the first line of the lines in the video marketing sample video (short video) (i.e. "Liu Moumou" with number 1) is successfully matched with the same line in the main line of the second episode of the target video ("Liu Moumou" with number 2, episode number 2), the second line of the lines in the video marketing sample video ("Get in the car" with number 2) is not successfully matched with the next line of the matched lines in the main line of the second episode of the target video ("What's wrong" with number 3, episode number 2). Therefore, the matching process needs to be restarted from the first line of the lines in the video marketing sample video. Figure 2 After successfully matching the same lines in the main lines of the fifth episode of the target video, the subsequent lines numbered 2-6 in the video marketing sample video were successfully matched with the subsequent lines (numbered 201-205) of the matched lines in the fifth episode of the target video. At this time, it can be considered that the positioning is successful.

[0103] The embodiment of the present invention determines the original video corresponding to the video marketing sample video by matching the lines of the video marketing sample video with the lines of the target video. It can accurately determine the original video corresponding to the video marketing sample video and accurately locate the complete plot segment corresponding to the plot in the video marketing sample video in the original video, ensuring the reasonable construction of the training data set of the first language model, thereby improving the training effect of the first language model.

[0104] In an optional embodiment of the present invention, the minute-level summary sample in the original video that matches the video marketing sample video is determined by the following steps:

[0105] Determine the single episode information and / or time information of the stock footage used in the video marketing sample video in the original video;

[0106] According to the single episode information and / or time information, a minute-level summary sample that matches the video marketing sample video is determined in the original video.

[0107] In this embodiment, after locating the complete plot segments corresponding to the plot of the stock footage used in the video marketing sample video in the original video, the single episode information and / or time information of these plot segments in the original video can be determined. For example, when the original video is a movie, the timestamp information of the complete plot segment itself can be directly used as the time information, but when the original video is a film and television series, it is necessary to determine the single episode information and time information of the complete plot segment in the original video. Specifically, the single episode information of the complete plot segment in the original video can be determined by the actual episode number information marked in the original video dialogue file of the original video matched in the aforementioned step. After the single episode information is clarified, the timestamp information of the complete plot segment itself can be directly used as the time information of the complete plot segment in the original video, thereby using this information to determine the minute-level summary sample that matches the video marketing sample video in the original video as the training sample data of the first language model.

[0108] The embodiment of the present invention can efficiently locate the minute-level summary samples in the original video that match the video marketing sample video by analyzing the single episode information and / or time information of the video marketing sample video plot clips in the original video, thereby ensuring a high degree of matching between the minute-level summary sample data and the video marketing sample video content, and improving the training effect of the first language model.

[0109] In an optional embodiment of the present invention, the line information sample in the original video that matches the minute-level summary sample is determined by the following steps:

[0110] The time period information corresponding to the minute-level summary sample in the original video is determined, and based on the time period information, a line information sample in the original video that matches the minute-level summary sample is determined.

[0111] In this embodiment, after determining the minute-level summary sample that matches the video marketing sample video in the original video, the time period information corresponding to the minute-level summary sample in the original video can be determined by referring to the correspondence shown in Table 1, and then the line information sample that matches the minute-level summary sample in the original video can be determined based on the time period information as the training sample data of the first language model. For example, the corresponding line paragraph of each minute-level summary sample can be obtained as the line information sample based on the time period in which it is located.

[0112] In an embodiment of the present invention, based on the time period information corresponding to the minute-level summary sample in the original video, the line information sample that matches the minute-level summary sample in the original video can be accurately determined, further ensuring the reasonable construction of the training data set of the first language model and improving the training effect of the first language model.

[0113] In an optional embodiment of the present invention, determining, based on the time period information, a line information sample in the original video that matches the minute-level summary sample includes:

[0114] According to the start and end timestamp range in the time period information, a speech text having the same start and end timestamp range is determined in the speech file of the original video, and the speech text is used as the speech information sample.

[0115] In this embodiment, the timestamp information in the time period information (e.g., the start and end timestamp range) can be compared with the timestamp information corresponding to the speech data in the original video speech file (e.g., a subtitle file or a speech recognition file) to find speech segments with the same start and end timestamp range in the speech data. Subsequently, all speech texts within the timestamp information range can be automatically extracted from the speech data and can be spliced ​​into paragraph text as speech information samples corresponding to minute-level summary samples.

[0116] The following will be combined Figure 3 The embodiments of the present invention are further described as follows:

[0117] Reference Figure 3 , is a schematic diagram of the overall execution flow of an embodiment of a method for generating video text according to the present invention;

[0118] like Figure 3 As shown, in a specific implementation, a training data construction module, a large language model training module and a forward reasoning module can be set separately to implement the video copy generation method of the present invention.

[0119] Training data construction module: The input is a video marketing short video, and the output is a minute-level summary of one or more plot segments corresponding to the video marketing short video in the main episode (original video), as well as lines, the marketing direction (marketing theme) of the video marketing short video, and the plot description corresponding to the video marketing short video (which can also be considered as the plot description corresponding to the marketing direction). The training dataset is constructed based on the minute-level summary of one or more plot segments corresponding to the video marketing short video in the main episode, as well as lines, the marketing direction of the video marketing short video, and the plot description corresponding to the video marketing short video.

[0120] Large language model training module: This module can train the basic model of the large language model to generate a marketing point mining model (i.e., the first large language model) under the condition that one or more random minute-level summaries and corresponding lines in the training data set are used as input data and the marketing direction and the corresponding plot description of the marketing direction are used as output data. For example, Qwen72b-Chat can be used as the basic model, the training method is Lora, the training set data volume is 6k, and a total of 8 epochs are trained. The training sample examples are as follows (where the corresponding real data is filled in the []):

[0121] {

[0122] "input": "As a TV series operator, you need to edit the plot that is attractive or resonates with the audience into a short video to attract more people to watch the full plot.

[0123] The plot materials are as follows:

[0124] Plot fragment 0: [Minute summary 0]

[0125] Plot Clip 1: [Minute Summary 1]

[0126] Plot Clip 2: [Minute Summary 2]

[0127] ===

[0128] Dialogue fragment 0: [Dialogue 0]

[0129] Dialogue fragment 1: [Dialogue 1]

[0130] Dialogue fragment 2: [Dialogue 2]

[0131] ===

[0132] Please analyze and filter the above plot materials to obtain the short video theme, and provide the plot content corresponding to the theme (*Note that the plot and theme should not include actor information, and do not make up the name of the series*). The answer is given in JSON format, with the key being the plot content and theme.

[0133] “output”: “{\"plot content\",\"theme\"}”

[0134] }.

[0135] Forward reasoning module: Input one or more minute-level summaries and lines from a certain episode into the trained first language model to generate the marketing direction of the plot and the plot description corresponding to the marketing direction (i.e., generate marketing points).

[0136] It should be noted that, for the method embodiments, the series of acts combined is described for simplicity, but those skilled in the art should know that the present application is not limited to the order of the acts described, because according to the present application, certain steps can be performed in other orders or at the same time. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the acts involved are not necessarily the necessary acts of the present application.

[0137] On the basis of the above-mentioned embodiments, the present embodiment further provides a video script generation device, which is applied to terminal equipment, server and other electronic equipment.

[0138] With reference to Figure 4 , a structural block diagram of an embodiment of a video script generation device of the present application is shown, which can specifically include the following modules:

[0139] The to-be-processed video acquisition module 401 is configured to acquire a to-be-processed video.

[0140] The determination module 402 is configured to determine a minute-level abstract of the to-be-processed video and dialogue information corresponding to the minute-level abstract, wherein the minute-level abstract is a plot of each minute of the video described in the form of text.

[0141] The marketing reference script generation module 403 is configured to generate a marketing reference script of the to-be-processed video based on the minute-level abstract and the dialogue information by using a pre-trained first large language model, wherein the marketing reference script is used to provide reference information for users when creating a marketing video of the to-be-processed video.

[0142] The first large language model is obtained by training based on a marketing theme sample in a video marketing sample video, a minute-level abstract sample in an original video corresponding to the video marketing sample video and matching the video marketing sample video, dialogue information samples in the original video corresponding to the minute-level abstract sample and matching the minute-level abstract sample, and a plot description sample corresponding to the video marketing sample video.

[0143] Optionally, the marketing reference script includes a marketing reference theme and a plot description, and the marketing reference script generation module 403 includes:

[0144] The marketing reference theme and plot description generation sub-module is configured to generate a marketing reference theme and plot description of the to-be-processed video based on the minute-level abstract and the dialogue information by using a pre-trained first large language model, wherein the marketing reference theme is used to provide reference information of a marketing theme for users when creating a marketing video of the to-be-processed video.

[0145] Optionally, the marketing reference theme and plot description generation submodule includes:

[0146] a first prompt text generating unit configured to write the minute-level summary and the dialogue information into a preset first prompt template to generate a first prompt text; wherein the first prompt template includes at least a prompt theme, and the prompt theme is used to prompt the first language model to output content based on the plot of interest to the audience;

[0147] The marketing reference theme and plot description generating unit is configured to input the first prompt text into the first language model, so that the first language model outputs the marketing reference theme and the plot description according to the first prompt text.

[0148] Optionally, the plot description sample corresponding to the video marketing sample video is determined by the following modules: a second prompt text determination module is used to write the minute-level summary sample, the line information sample, the marketing theme sample and the lines in the video marketing sample video into a preset second prompt template to obtain a second prompt text;

[0149] The plot description sample determination module is used to input the second prompt text into the second largest language model, so that the second largest language model outputs the plot description sample corresponding to the video marketing sample video according to the second prompt text.

[0150] Optionally, the original video corresponding to the video marketing sample video is determined by the following modules, including:

[0151] A dialogue matching module, for matching the dialogue of the sample video of the video marketing with the dialogue of the target video;

[0152] The original video determination module is used to determine the target video as the original video corresponding to the video marketing sample video when the number of consecutive dialogue text sentences matching the target video exceeds a preset threshold.

[0153] Optionally, the minute-level summary sample in the original video that matches the video marketing sample video is determined by the following modules:

[0154] A module for determining single episode information and / or time information, for determining single episode information and / or time information of the stock footage used in the video marketing sample video in the original video;

[0155] A minute-level summary sample determination module is used to determine a minute-level summary sample that matches the video marketing sample video in the original video based on the single episode information and / or time information.

[0156] Optionally, the line information samples in the original video that match the minute-level summary samples are determined by the following modules:

[0157] A time period information determination module, configured to determine the time period information corresponding to the minute-level summary sample in the original video;

[0158] The dialogue information sample determination module is configured to determine, based on the time period information, dialogue information samples in the original video that match the minute-level summary samples.

[0159] Optionally, the dialogue information sample determination module includes:

[0160] The speech information sample determination submodule is used to determine the speech text with the same start and end timestamp range in the speech file of the original video according to the start and end timestamp range in the time period information, and use the speech text as the speech information sample.

[0161] Optionally, the marketing reference copy includes a marketing reference theme and a plot description, and the device further includes:

[0162] A video clip acquisition module, configured to acquire the video clip of the video to be processed;

[0163] A video clip processing module, configured to generate audio commentary corresponding to the video clip based on the plot description, and to add relevant content of the marketing reference theme to the video clip;

[0164] A marketing video generation module is used to generate a marketing video of the video to be processed based on the video clip and the commentary audio.

[0165] Optionally, the video clip processing module includes:

[0166] A video title processing submodule is used to add text corresponding to the marketing reference theme to the video title of the video clip, and / or a commentary audio processing submodule is used to add audio corresponding to the marketing reference theme to the commentary audio of the video clip.

[0167] In an embodiment of the present invention, a first language model is trained based on samples including marketing theme samples in video marketing sample videos, and the first language model is used to automatically generate marketing reference copy for the video to be processed. Thus, when the user creates a video marketing short video, the marketing reference copy can be automatically provided as a reference for creative inspiration and creative direction of the marketing copy, thereby improving the efficiency of video marketing short video creation.

[0168] The embodiment of the present invention further provides an electronic device, such as Figure 5As shown, it includes a processor 501, a communication interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504.

[0169] Memory 503, used for storing computer programs;

[0170] The processor 501 is configured to execute the program stored in the memory 503 by performing the following steps:

[0171] Get the video to be processed;

[0172] Determine a minute-level summary of the video to be processed and line information corresponding to the minute-level summary; wherein the minute-level summary is a textual description of the plot of each minute of the video;

[0173] Using the pre-trained first language model, based on the minute-level summary and the line information, a marketing reference copy for the video to be processed is generated; wherein the marketing reference copy is used to provide reference information for users when creating a marketing video for the video to be processed;

[0174] Among them, the first language model is trained based on marketing theme samples in video marketing sample videos, minute-level summary samples in the original video corresponding to the video marketing sample videos that match the video marketing sample videos, dialogue information samples in the original video that match the minute-level summary samples, and plot description samples corresponding to the video marketing sample videos.

[0175] The communication bus mentioned in the terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or only one type of bus.

[0176] The communication interface is used for communication between the above terminal and other devices.

[0177] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage. Alternatively, the memory may be at least one storage device located away from the processor.

[0178] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0179] In another embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores instructions that, when executed on a computer, enable the computer to execute the video copy generation method described in any one of the above embodiments.

[0180] In another embodiment of the present invention, a computer program product including instructions is provided. When the computer program product is run on a computer, the computer executes the video copy generation method described in any one of the above embodiments.

[0181] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0182] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0183] Each embodiment in this specification is described in a related manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiment is generally similar to the method embodiment, so the description is relatively simple. For related parts, refer to the description of the method embodiment.

[0184] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention are included in the scope of protection of the present invention.

Claims

1. A method for generating video copy, characterized in that: The method comprises: Get the video to be processed; Determine a minute-level summary of the video to be processed and line information corresponding to the minute-level summary; wherein the minute-level summary is a textual description of the plot of each minute of the video; Using the pre-trained first language model, based on the minute-level summary and the line information, a marketing reference copy for the video to be processed is generated; wherein the marketing reference copy is used to provide reference information for users when creating a marketing video for the video to be processed; Among them, the first language model is trained based on marketing theme samples in video marketing sample videos, minute-level summary samples in the original video corresponding to the video marketing sample videos that match the video marketing sample videos, dialogue information samples in the original video that match the minute-level summary samples, and plot description samples corresponding to the video marketing sample videos.

2. The method according to claim 1, characterized in that The marketing reference copy includes a marketing reference theme and a plot description. The use of the pre-trained first language model to generate the marketing reference copy for the video to be processed based on the minute-level summary and the line information includes: Utilizing the pre-trained first language model, based on the minute-level summary and the line information, a marketing reference theme and plot description of the video to be processed are generated; wherein, the marketing reference theme is used to provide users with reference information on the marketing theme when creating a marketing video for the video to be processed.

3. The method according to claim 2, characterized in that The method utilizes the pre-trained first language model to generate a marketing reference theme and plot description of the video to be processed based on the minute-level summary and the dialogue information, including: Writing the minute-level summary and the dialogue information into a preset first prompt template to obtain a first prompt text; wherein the first prompt template at least includes a prompt theme, and the prompt theme is used to prompt the first language model to output content based on the plot that the audience is interested in; The first prompt text is input into the first language model, so that the first language model outputs the marketing reference theme and the plot description according to the first prompt text.

4. The method according to any one of claims 1 to 3, characterized in that The plot description sample corresponding to the video marketing sample video is determined by the following steps: Writing the minute-level summary sample, the line information sample, the marketing theme sample, and the lines in the video marketing sample video into a preset second prompt template to obtain a second prompt text; The second prompt text is input into the second largest language model, so that the second largest language model outputs a plot description sample corresponding to the video marketing sample video according to the second prompt text.

5. The method according to any one of claims 1 to 3, characterized in that The original video corresponding to the video marketing sample video is determined by the following steps: Matching the lines of the video marketing sample video with the lines of the target video; When the number of consecutive lines of text that match the video marketing sample video and the target video exceeds a preset threshold, the target video is determined to be the original video corresponding to the video marketing sample video.

6. The method according to any one of claims 1 to 3, characterized in that The minute-level summary sample in the original video that matches the video marketing sample video is determined by the following steps: Determine the single episode information and / or time information of the stock footage used in the video marketing sample video in the original video; According to the single episode information and / or time information, a minute-level summary sample that matches the video marketing sample video is determined in the original video.

7. The method according to any one of claims 1 to 3, characterized in that The line information sample in the original video that matches the minute-level summary sample is determined by the following steps: Determine time period information corresponding to the minute-level summary sample in the original video; According to the time period information, a line information sample in the original video that matches the minute-level summary sample is determined.

8. The method according to claim 7, characterized in that The step of determining, based on the time period information, a line information sample in the original video that matches the minute-level summary sample includes: According to the start and end timestamp range in the time period information, a speech text having the same start and end timestamp range is determined in the speech file of the original video, and the speech text is used as the speech information sample.

9. The method according to any one of claims 1 to 3, characterized in that The marketing reference copy includes a marketing reference theme and a plot description, and the method further includes: Obtaining a video clip of the video to be processed; Generate audio commentary corresponding to the video clip based on the plot description, and add relevant content of the marketing reference theme to the video clip; A marketing video of the video to be processed is generated based on the video clip and the commentary audio.

10. The method according to claim 9, characterized in that The content related to the marketing reference theme added to the video clip includes: Add text corresponding to the marketing reference theme to the video title of the video clip, and / or add audio corresponding to the marketing reference theme to the commentary audio of the video clip.

11. The method according to any one of claims 1 to 3, characterized in that The first language model is trained using the following method: Constructing a training dataset based on the marketing theme samples, minute-level summary samples, dialogue information samples, and plot description samples; A large language model is trained under the conditions that at least one minute-level summary in the training data set and at least one line information corresponding to the at least one minute-level summary are used as input data, and the marketing reference copy is used as output data to obtain the first large language model.

12. A video copy generation device, characterized in that: The device comprises: A module for obtaining videos to be processed, used to obtain videos to be processed; A determination module, configured to determine a minute-level summary of the video to be processed and line information corresponding to the minute-level summary; wherein the minute-level summary is a textual description of the plot of each minute of the video; a marketing reference copy generation module, utilizing the pre-trained first language model to generate a marketing reference copy for the video to be processed based on the minute-level summary and the line information; wherein the marketing reference copy is used to provide reference information for users when creating a marketing video for the video to be processed; Among them, the first language model is trained based on marketing theme samples in video marketing sample videos, minute-level summary samples in the original video corresponding to the video marketing sample videos that match the video marketing sample videos, dialogue information samples in the original video that match the minute-level summary samples, and plot description samples corresponding to the video marketing sample videos.

13. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; The processor is configured to implement the steps of the video copy generation method described in any one of claims 1 to 11 when executing the program stored in the memory.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the video copy generation method as described in any one of claims 1 to 11 is implemented.

Images