Method for adjusting cue word and electronic equipment

Abstract

The embodiment of the invention provides a cue word adjusting method and electronic equipment, and the method comprises the steps: firstly, obtaining an original cue word input by a user; then, the original cue word is recognized, structured information is obtained, the structured information comprises a plurality of information items, and the value of a single information item is null or at least part of the original cue word; then, carrying out first adjustment processing on information items, which do not meet the media generation condition, in the structured information; and then, integrating the information items meeting the media content generation condition in the structured information and the information items subjected to the first adjustment processing to obtain a target prompt word, the target prompt word being used for indicating generation of the media content. Therefore, the text semantics of the target cue word can be richer and more complete (namely, the quality of the target cue word is better), so that the quality of the media content generated by the media generation model can be improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of media technology, and more particularly to a method for adjusting a prompt word and an electronic device. Background Art

[0002] AIGC (Artificial Intelligence-Generated Content) technology refers to the process of using artificial intelligence technology to automate or assist information creation to meet the personalized needs of users. For example, AIGC technology can generate copywriting, knowledge graphs, and be used as text customer service in the fields of media, film and television, advertising, games, e-commerce, education, and medical care; in the field of images, AIGC technology can be used for image editing, image generation, and image recognition; in the field of audio, AIGC technology can be used for music production and speech synthesis; in the field of video, AIGC technology can be used for video post-production and virtual human creation.

[0003] Users can interact with AI models through prompts, instructing the AI ​​model to output the required AIGC (such as images, audio, etc.). The quality of prompts affects the quality of AIGC; in actual use, the quality of prompts input by users varies, making it difficult to guarantee the quality of generated AIGC. Of course, users can also look for prompts with better quality to generate AIGC with better quality, but this takes a lot of time and effort and is costly. Summary of the invention

[0004] In view of this, the present application provides a prompt word adjustment method and an electronic device. The text semantics of the target prompt word obtained by using the prompt word adjustment method is richer and more complete, thereby improving the quality of the generated media content.

[0005] In a first aspect, an embodiment of the present application provides a method for adjusting a prompt word, the method comprising: first, obtaining an original prompt word input by a user; then, identifying and processing the original prompt word to obtain structured information, wherein the structured information includes multiple information items, and the value of a single information item is empty or at least part of the original prompt word; then, performing a first adjustment process on the information items in the structured information that do not meet the media generation conditions; thereafter, integrating the information items in the structured information that meet the media content generation conditions and the information items that have been subjected to the first adjustment process to obtain a target prompt word, wherein the target prompt word is used to indicate the generation of media content.

[0006] That is to say, the present application divides the original prompt word into fine-grained information and then adjusts each granularity of information; compared with the prior art that directly adjusts the original prompt word input by the user, the present application adjusts the original prompt word input by the user at a finer and more comprehensive granularity, thereby making the target prompt word text semantics richer and more complete (that is, the quality of the target prompt word is better), thereby improving the quality of the generated media content.

[0007] In addition, the present application converts different original prompt words into preset structured information before making adjustments, which can increase the controllability of the target prompt words, thereby improving the controllability of the quality of the generated media content.

[0008] The quality of media content can be measured by the consistency between the media content and the target prompt words, the aesthetic score of the media content and the safety of the media content.

[0009] Exemplarily, the first adjustment process may include at least one of the following: text enhancement process or keyword extraction process. The text enhancement process may include: adding detailed description of the text, enriching the text, etc.; the keyword extraction process may include extracting keywords from the original prompt words.

[0010] Exemplarily, a single information item in the structured information may include a name of the information item and a value of the information item; the name of the information item is also the type name of the information item.

[0011] For example, the original prompt word is "a cute little girl", and the structured information can be as follows:

[0012] “Subject: A cute little girl.

[0013] Style: None.

[0014] Light and shadow: None.

[0015] Image quality: None. "

[0016] Among them, structured information can include 4 information items:

[0017] Information item 1 is "Subject: a cute little girl"; the name of information item 1 is "Subject", and the value of information item 1 is "a cute little girl", wherein the value of information item 1 is the original prompt word.

[0018] Information item 2 is "Style: None"; the name of information item 2 is "Style", and the value of information item 2 is empty.

[0019] Information item 3 is “Light and Shadow: None”; the name of information item 3 is “Light and Shadow”, and the value of information item 3 is empty.

[0020] Information item 4 is “Image quality: None”; the name of information item 3 is “Image quality”, and the value of information item 4 is empty.

[0021] Correspondingly, the first adjustment process may be text enhancement process, and the target prompt words may be as follows:

[0022] "A cute little girl in a beautiful dress walking on the street, kawaii anime style, sunset, ultra high definition, 8k".

[0023] For example, if the original prompt word is "a cute little girl who is as energetic as the rising sun in the morning, wearing a very beautiful skirt, walking happily on the street, kawaii anime style, sunset, ultra high definition, 8k", the structured information can be as follows:

[0024] “Subject: A lovely little girl, as lively as the rising sun in the morning, wearing a very beautiful skirt, strolling happily on the street.

[0025] Style: Kawaii anime style.

[0026] Light and shadow: sunset.

[0027] Image quality: Ultra HD, 8k.”

[0028] Among them, structured information can include 4 information items:

[0029] Information item 1 is "Subject: A cute little girl who is as energetic as the rising sun in the morning, wearing a very beautiful skirt, strolling happily on the street"; the name of information item 1 is "Subject", and the value of information item 1 is "A cute little girl who is as energetic as the rising sun in the morning, wearing a very beautiful skirt, strolling happily on the street", where the value of information item 1 is part of the original prompt word.

[0030] Information item 2 is "Style: Kawaii anime style"; the name of information item 2 is "Style" and the value of information item 2 is "Kawaii anime style". The value of information item 2 is part of the original prompt word.

[0031] Information item 3 is "light and shadow: sunset"; the name of information item 3 is "light and shadow", and the value of information item 3 is "sunset". The value of information item 3 is part of the original prompt word.

[0032] Information item 4 is “Image quality: ultra high definition, 8k”; the name of information item 3 is “Image quality”, and the value of information item 4 is “Ultra high definition, 8k”. The value of information item 4 is part of the original prompt word.

[0033] Correspondingly, the first adjustment process may be a keyword extraction process, and the target prompt word may be as follows:

[0034] "A cute little girl in a beautiful dress walking on the street, kawaii anime style, sunset, ultra high definition, 8k".

[0035] For example, if the original prompt word is "a lovely little girl who is as energetic as the rising sun in the morning", the structured information can be as follows:

[0036] “Subject: A lovely little girl who is as lively as the rising sun in the morning.

[0037] Style: None.

[0038] Light and shadow: None.

[0039] Image quality: None. "

[0040] Among them, structured information can include 4 information items:

[0041] Information item 1 is “Subject: A lovely little girl who is like the rising sun in the morning and full of vitality”; the name of information item 1 is “Subject”, and the value of information item 1 is “A lovely little girl”, wherein the value of information item 1 is the original prompt word.

[0042] Information item 2 is "Style: None"; the name of information item 2 is "Style", and the value of information item 2 is empty.

[0043] Information item 3 is “Light and Shadow: None”; the name of information item 3 is “Light and Shadow”, and the value of information item 3 is empty.

[0044] Information item 4 is “Image quality: None”; the name of information item 3 is “Image quality”, and the value of information item 4 is empty.

[0045] Correspondingly, the first adjustment process may include text enhancement processing and keyword extraction, and the target prompt words may be as follows:

[0046] "A cute little girl in a beautiful dress walking on the street, kawaii anime style, sunset, ultra high definition, 8k".

[0047] For example, media content can also be called digital content. Digital content is content of different content types such as text, images, and sounds in digital form. It can be stored on digital carriers such as CDs and hard disks, and spread through the Internet and other means. Digital content is the overall product or service that integrates images, text, audio and video through digital technology. It is the product of the combination of digital media technology and cultural creativity.

[0048] The media content may include, but is not limited to, text, images, graphics, audio, and video, etc., which are not limited in this application.

[0049] Exemplarily, the media generation model is an AI model (or a neural network or a neural network model). The media generation model model of the present application can be implemented by at least one of the following neural networks: deep neural networks (DNN), convolutional neural networks (CNN), recurrent neural networks (RNN), residual networks, neural networks using transformer models or other neural networks, etc., and the present application does not limit this.

[0050] Exemplarily, the target cue word may be used for at least a media generation model to generate media content.

[0051] Exemplarily, the media generation model can be used to generate media content (the media content referred to in this application may refer to AIGC). The types of media generation models may include, but are not limited to: image generation models, audio generation models, video generation models, text generation models, graphics generation models, etc., which are not limited in this application. For example, ChatGPT (a natural language processing tool driven by artificial intelligence technology), Midjourney (an AI drawing tool), etc.

[0052] Exemplarily, prompt words (which may include original prompt words and target prompt words) are Prompts, which may also be called media generation prompt words, prompt words of media generation model or AI prompt words, and may be used to guide (instruct) or inspire the media generation model to complete a specific task. Prompts can help the media generation model better understand the input intent and make corresponding responses such as outputting media content (i.e., AIGC). Among them, Prompts can be composed of multiple words, phrases or short sentences.

[0053] Exemplarily, when the prompt word is a prompt word for generating an image, the prompt word can be input into an image generation model, and the image generation model generates an image. Exemplarily, when the prompt word is a prompt word for generating an audio, the prompt word can be input into an audio generation model, and the audio generation model generates an audio. Exemplarily, when the prompt word is a prompt word for generating a video, the prompt word can be input into a video generation model, and the video generation model generates a video. Exemplarily, when the prompt word is a prompt word for generating a text, the prompt word can be input into a text generation model, and the text generation model generates a text. Exemplarily, when the prompt word is a prompt word for generating a graphic, the prompt word can be input into a graphic generation model, and the graphic generation model generates a graphic.

[0054] According to the first aspect, the structured information includes at least the following two information items: a subject information item and a style information item.

[0055] Exemplarily, when the original prompt words are used to generate an image / video, the structured information may further include a light and shadow information item and a picture quality information item.

[0056] Exemplarily, when the original prompt words are used to generate audio, the structured information may further include a timbre information item.

[0057] It should be understood that the present application does not limit the types of information items contained in the structured information. The structure of the reference prompt words can be determined by analyzing the prompt words (hereinafter referred to as reference prompt words) required for the media generation model to generate high-quality media content; specifically, the information items contained in the reference prompt words can be determined. Afterwards, the original prompt words input by the user can be identified and processed according to the types of information items contained in the reference prompt words to obtain structured information containing multiple information items. The types of information items contained in the structured information are the same as the types of information items contained in the reference prompt words.

[0058] According to the first aspect, or any implementation of the first aspect above, the method further includes: performing a first evaluation process on each information item in the structured information to obtain first evaluation information corresponding to each information item in the structured information; and judging whether each information item in the structured information satisfies the media content generation condition according to the first evaluation information corresponding to each information item in the structured information. In this way, it is possible to quickly judge whether each information item satisfies the media content condition.

[0059] Exemplarily, the first evaluation process includes but is not limited to scoring, completeness evaluation / matching evaluation, and grade evaluation. The corresponding first evaluation information may include but is not limited to: score, completeness / matching evaluation result, and grade information. Among them, the matching evaluation may be an evaluation of the matching degree between the original prompt word and the media generation model. This application takes the first evaluation process as scoring and the first evaluation information as score as an example for explanation.

[0060] Exemplarily, the present application may pre-set media content generation conditions; for example, the media content generation conditions are: the score corresponding to the information item in the structured information is higher than or equal to the corresponding threshold. The threshold can be set as required, and the present application does not limit this. A corresponding threshold can be set for each information item, for example, the threshold corresponding to the subject information item is set to 85%, the threshold corresponding to the style information item is set to 1, the threshold corresponding to the light and shadow information is set to 1, and the threshold corresponding to the image quality information item is set to 1.

[0061] Furthermore, for an information item in the structured information, it may be determined whether a score corresponding to the information item is higher than or equal to a corresponding threshold value, so as to determine whether the information item meets the media content generation condition.

[0062] According to the first aspect, or any implementation of the first aspect above, the first evaluation information includes a score, and the media content generation condition includes: a score corresponding to a single information item in the structured information is higher than or equal to a corresponding threshold.

[0063] According to the first aspect, or any implementation of the first aspect above, the method further includes: acquiring media content, the media content is generated according to the target prompt word; when the media content does not meet the media content quality condition, re-performing a first adjustment process on the information item in the structured information that does not meet the media content generation condition. In this way, the quality of the target prompt word finally obtained can be improved, thereby improving the quality of the media content generated based on the target prompt word.

[0064] Exemplarily, the media content may be obtained by inputting the target cue word into the media generation model.

[0065] According to the first aspect, or any implementation of the first aspect above, the method also includes: performing a second evaluation process on the media content to obtain second evaluation information corresponding to the media content; and judging whether the media content meets the media content quality condition based on the second evaluation information corresponding to the media content.

[0066] Exemplarily, the second evaluation process can be performed on the media content from multiple dimensions such as security, aesthetics, and semantic consistency to obtain a security score corresponding to the media content, an aesthetics score corresponding to the media content, and a semantic consistency score corresponding to the media content. In one possible approach, the second evaluation information corresponding to the media content may include: a security score corresponding to the media content, an aesthetics score corresponding to the media content, and a semantic consistency score corresponding to the media content.

[0067] Exemplarily, the media content quality condition may be preset, for example, the media content quality condition is: the security score is greater than the security threshold, the aesthetic score is greater than the aesthetic threshold, and the semantic consistency score is greater than the consistency threshold.

[0068] Specifically, it can be determined whether the security score corresponding to the media content is greater than the security threshold, whether the aesthetic score of the media content is greater than the aesthetic threshold, and whether the semantic consistency score corresponding to the media content is greater than the consistency threshold; when the security score corresponding to the media content is greater than the security threshold, and the aesthetic score of the media content is greater than the aesthetic threshold, and the semantic consistency score corresponding to the media content is greater than the consistency threshold, it can be determined that the media content meets the media content quality conditions; otherwise, it can be determined that the media content does not meet the media content quality conditions.

[0069] According to the first aspect, or any implementation of the first aspect above, the method further includes: performing a safety judgment on the original prompt word; when the original prompt word does not meet the safety condition, performing a second adjustment process on the original prompt word, so that the prompt word after the second adjustment process meets the safety condition. In this way, the appearance of pornographic, violent, and racially discriminatory content can be reduced, and the use occasions of prompt word adjustment and AIGC products can be expanded.

[0070] Exemplarily, the second adjustment process may refer to security rewriting, which may include deleting preset words, replacing preset words with words with similar meanings, etc. Preset words may refer to words with security issues, such as words describing pornography, words with racial discrimination, etc.

[0071] According to the first aspect, or any implementation of the first aspect above, the method further includes: when the original prompt word does not meet the security condition, displaying a security prompt interface; wherein the security prompt interface includes a manual adjustment option and an automatic adjustment option; receiving an operation of selecting the automatic adjustment option by a user; and in response to the operation of selecting the automatic adjustment option, executing a step of performing a second adjustment process on the original prompt word. In this way, it can be determined according to the user's selection whether the system performs a security rewrite or the user customizes the security rewrite of the original prompt word; and multiple security rewrite methods are provided.

[0072] According to the first aspect, or any implementation of the first aspect above, the method further includes: when the original prompt word does not meet the security condition, displaying a security prompt interface; wherein the security prompt interface includes a manual adjustment option and an automatic adjustment option; receiving an operation of the user selecting the manual adjustment option; and in response to the operation of selecting the manual adjustment option, obtaining the original prompt word adjusted by the user. In this way, the user can customize the security rewriting of the original prompt word with security issues.

[0073] According to the first aspect, or any implementation of the first aspect above, the method further includes: displaying structured information.

[0074] In this way, by displaying structured information, users can understand the defects of the original prompt words they input; further, when users need to modify the original prompt words they input this time, the displayed structured information can guide the modification direction of the original prompt words, reducing the cost of manual trial and error. In addition, the original prompt words that users input later can also be prompted to improve the quality of the original prompt words that users input later.

[0075] According to the first aspect, or any implementation of the first aspect above, the method further includes: displaying first evaluation information corresponding to each information item in the structured information.

[0076] In this way, by displaying the first evaluation information corresponding to the original prompt word, the user can understand the more fine-grained defects of the original prompt word he or she inputs; further, when the user needs to modify the original prompt word input this time, the displayed first evaluation information can guide the modification direction of the prompt word, reducing the cost of manual trial and error. In addition, the user can also be prompted for the original prompt word input later to improve the quality of the original prompt word input later by the user.

[0077] According to the first aspect, or any implementation of the first aspect above, the method further includes: displaying the target prompt word. In this way, it is convenient for the user to know the form and content of the prompt word with better quality, so as to improve the quality of the original prompt word input by the user later.

[0078] According to the first aspect, or any implementation of the first aspect above, the method further includes: receiving an adjustment operation of the user on the target information item in the structured information, and obtaining the target information item adjusted by the user. In this way, after the prompt words are subdivided, the user can modify them by himself, which can reduce the difficulty of the user modifying the prompt words.

[0079] According to the first aspect, or any implementation of the first aspect above, the method further includes: receiving an adjustment operation of the user on the target prompt word, and obtaining the target prompt word adjusted by the user. In this way, when the user is not satisfied with the target prompt word obtained by the system performing the first adjustment process on the original prompt word, the user can modify it by customization, so as to facilitate the user to customize the personalized prompt word.

[0080] According to the first aspect, or any implementation of the first aspect, the original prompt word is recognized and processed to obtain structured information, including: inputting the original prompt word into a language model to obtain structured information.

[0081] Exemplarily, the language model may be a Large Language Model (LLM), which may be composed of an artificial neural network with many parameters (typically billions of weights or more) and trained using self-supervised learning or semi-supervised learning.

[0082] In one possible way, the language model can be one; in this case, the context learning (InContext Learning, ICL) technology can be used to issue recognition tasks to the language model so that the language model can recognize the values ​​of each information item in the structured information from the original prompt words entered by the user. For example, ICL issues a subject information recognition task to LLM, and LLM recognizes and processes the original prompt words entered by the user, and outputs the value of the subject information item (such as "a cute little girl"). Next, ICL issues a style information recognition task to LLM, and LLM recognizes and processes the original prompt words entered by the user, and outputs the value of the style information item. Subsequently, ICL issues a light and shadow information recognition task to LLM, and LLM recognizes and processes the original prompt words entered by the user, and outputs the value of the light and shadow information item. After that, ICL issues a picture quality information recognition task to LLM, and LLM recognizes and processes the original prompt words entered by the user, and outputs the value of the picture quality information item.

[0083] In one possible way, there can be multiple language models; in this case, one language model can be used to identify the value of one or more information items in the structured information. For example, there are four language models: language model A1, language model A2, language model A3, and language model A4; the original prompt word input by the user can be input into language model A1 to obtain the value of the main information item; the original prompt word input by the user can be input into language model A2 to obtain the value of the style information item; the original prompt word input by the user can be input into language model A3 to obtain the value of the light and shadow information item; the original prompt word input by the user can be input into language model A4 to obtain the value of the image quality information item.

[0084] According to the first aspect, or any implementation of the first aspect above, the method also includes: displaying multiple adjustment performance options, the multiple adjustment performance options corresponding to multiple groups of adjustment models; receiving a user's selection operation for a target adjustment performance option; performing a first adjustment process on information items in the structured information that do not meet the media generation conditions, including: calling a group of adjustment models corresponding to the target adjustment performance option, and performing a first adjustment process on information items in the structured information that do not meet the media generation conditions.

[0085] In one possible approach, a group of adjustment models may be used to perform a first adjustment process on multiple information items that do not meet media generation conditions; wherein a group of adjustment models may include one or more adjustment models, wherein one adjustment model may be used to perform a first adjustment process on one or more information items that do not meet media generation conditions.

[0086] In one possible manner, multiple groups of adjustment models may be pre-set, and one group of adjustment models may be selected from the multiple groups of adjustment models each time to perform the first adjustment processing on multiple information items that do not meet the media generation condition; wherein one group of adjustment models may include one or more adjustment models, wherein one adjustment model may be used to perform the first adjustment processing on one or more information items that do not meet the media generation condition. wherein, among the multiple groups of adjustment models, the adjustment models used to perform the first adjustment processing on the same information item that does not meet the media generation condition may be adjustment models with different numbers of model parameters, and different adjustment efficiencies and adjustment qualities.

[0087] In this way, a set of matching adjustment models can be adaptively determined to perform the first adjustment process according to the user's requirements on adjustment efficiency, adjustment quality, etc., as well as factors such as the computing power of the device.

[0088] According to the first aspect, or any implementation of the first aspect above, the target prompt word is used to instruct the media generation model to generate media content, and the information items in the structured information that meet the media content generation conditions and the information items that have been processed by the first adjustment are integrated to obtain the target prompt word, including: according to the prompt word structure that matches the media generation model, the information items in the structured information that meet the media content generation conditions and the information items that have been processed by the first adjustment are spliced ​​to obtain the target prompt word. In this way, due to the media generation model, the quality of the media content generated by using the prompt word with the prompt word structure whose information structure matches the media generation model as input is better, and further, the present application can improve the quality of the media content by integrating according to the prompt word structure that matches the media generation model.

[0089] Exemplarily, the prompt word structure that matches the media generation model may refer to the prompt word structure required by the media generation model to generate higher quality media content.

[0090] It should be noted that the first aspect and any implementation method of the first aspect can be executed by the server (except the display steps and the steps of receiving user operations), or by the terminal device, or can be executed collaboratively by the server and the terminal device, and this application does not impose any restrictions on this.

[0091] In a second aspect, an embodiment of the present application provides a device for adjusting a prompt word, the device comprising:

[0092] An acquisition module is used to acquire the original prompt word input by the user;

[0093] A structured recognition module, used to recognize and process the original prompt word to obtain structured information, wherein the structured information includes multiple information items, and the value of a single information item is empty or at least part of the original prompt word;

[0094] An adjustment module, configured to perform a first adjustment process on information items in the structured information that do not meet the media generation condition;

[0095] The integration module is used to integrate the information items in the structured information that meet the media content generation condition and the information items that have undergone the first adjustment process to obtain a target prompt word, wherein the target prompt word is used to indicate the generation of media content.

[0096] It should be understood that the prompt word adjustment device of the second aspect can execute the steps in the first aspect and any implementation method of the first aspect, which will not be described in detail herein.

[0097] The second aspect and any implementation of the second aspect correspond to the first aspect and any implementation of the first aspect respectively. The technical effects corresponding to the second aspect and any implementation of the second aspect can refer to the technical effects corresponding to the above-mentioned first aspect and any implementation of the first aspect, which will not be repeated here.

[0098] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a memory and a processor, wherein the memory is coupled to the processor; the memory stores program instructions, and when the program instructions are executed by the processor, the electronic device executes the method for adjusting the prompt word in the first aspect or any possible implementation of the first aspect.

[0099] The third aspect and any implementation of the third aspect correspond to the first aspect and any implementation of the first aspect, respectively. The technical effects corresponding to the third aspect and any implementation of the third aspect can refer to the technical effects corresponding to the first aspect and any implementation of the first aspect, which will not be repeated here.

[0100] In a fourth aspect, an embodiment of the present application provides a chip comprising one or more interface circuits and one or more processors; the one or more processors receive or send data through the one or more interface circuits, and when the one or more processors execute computer instructions, the steps of the method for adjusting the prompt word in the first aspect or any possible implementation of the first aspect are executed.

[0101] The fourth aspect and any implementation of the fourth aspect correspond to the first aspect and any implementation of the first aspect, respectively. The technical effects corresponding to the fourth aspect and any implementation of the fourth aspect can refer to the technical effects corresponding to the above-mentioned first aspect and any implementation of the first aspect, which will not be repeated here.

[0102] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a computer or a processor, the computer or the processor executes the method for adjusting the prompt word in the first aspect or any possible implementation of the first aspect.

[0103] The fifth aspect and any implementation of the fifth aspect correspond to the first aspect and any implementation of the first aspect, respectively. The technical effects corresponding to the fifth aspect and any implementation of the fifth aspect can refer to the technical effects corresponding to the first aspect and any implementation of the first aspect, which will not be repeated here.

[0104] In a sixth aspect, an embodiment of the present application provides a computer program product, which includes computer instructions. When the computer instructions are executed by a computer or a processor, the computer or the processor executes the method for adjusting the prompt word in the first aspect or any possible implementation of the first aspect.

[0105] The sixth aspect and any implementation of the sixth aspect correspond to the first aspect and any implementation of the first aspect, respectively. The technical effects corresponding to the sixth aspect and any implementation of the sixth aspect can refer to the technical effects corresponding to the first aspect and any implementation of the first aspect, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 is a schematic diagram of an application scenario shown as an example;

[0107] Figure 2 is a schematic diagram of an exemplary adjustment process of a prompt word;

[0108] Figure 3A is a schematic diagram of a mobile phone interface shown as an example;

[0109] Figure 3B is a schematic diagram of a mobile phone interface shown as an example;

[0110] Figure 3C is a schematic diagram of a mobile phone interface shown as an example;

[0111] Figure 3D is a schematic diagram of a mobile phone interface shown as an example;

[0112] Figure 4A is a schematic diagram of a mobile phone interface shown as an example;

[0113] Figure 4B is a schematic diagram of a mobile phone interface shown as an example;

[0114] Figure 4Cis a schematic diagram of a mobile phone interface shown as an example;

[0115] Figure 4D is a schematic diagram of a mobile phone interface shown as an example;

[0116] Figure 4E is a schematic diagram of a mobile phone interface shown as an example;

[0117] Figure 5A is a schematic diagram of a mobile phone interface shown as an example;

[0118] Figure 5B is a schematic diagram of a mobile phone interface shown as an example;

[0119] Figure 5C is a schematic diagram of a mobile phone interface shown as an example;

[0120] Figure 5D is a schematic diagram of a mobile phone interface shown as an example;

[0121] Figure 5E is a schematic diagram of a mobile phone interface shown as an example;

[0122] Fig. 5F is a schematic diagram of a mobile phone interface shown as an example;

[0123] Figure 5G A schematic diagram of an image generated by an exemplary media generation model;

[0124] Figure 5H A schematic diagram of an image generated by an exemplary media generation model;

[0125] Figure 6 is a schematic diagram of an exemplary adjustment process of a prompt word;

[0126] Figure 7 is a schematic diagram of an exemplary adjustment process of a prompt word;

[0127] Fig. 8A is a schematic diagram of a mobile phone interface shown as an example;

[0128] Figure 8B is a schematic diagram of a mobile phone interface shown as an example;

[0129] Figure 8C is a schematic diagram of a mobile phone interface shown as an example;

[0130] Fig. 9 is a schematic diagram of an exemplary adjustment process of a prompt word;

[0131] Fig.10 is a schematic structural diagram of an adjustment device for prompt words shown as an example;

[0132] Fig.11 Schematic diagram of the structure of the device shown as an example. DETAILED DESCRIPTION

[0133] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0134] The term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0135] The terms "first" and "second" in the description and claims of the embodiments of the present application are used to distinguish different objects rather than to describe a specific order of objects. For example, a first target object and a second target object are used to distinguish different target objects rather than to describe a specific order of target objects.

[0136] In the embodiments of the present application, the words "exemplarily" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplarily" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplarily" or "for example" is intended to present related concepts in a specific way.

[0137] In the description of the embodiments of the present application, unless otherwise specified, the meaning of "multiple" refers to two or more than two. For example, multiple processing units refer to two or more processing units; multiple systems refer to two or more systems.

[0138] Exemplarily, when a user needs to use a media generation model to generate media content, the user can input an original prompt word in a terminal device (e.g., a mobile phone, a laptop computer, a tablet, etc.); then, the terminal device sends the original prompt word input by the user to the first server. A prompt word adjustment module is deployed in the first server, and the first server can call the prompt word adjustment module to perform a first adjustment process on the original prompt word input by the user, obtain a target prompt word, and send the target prompt word to the second server. A media generation model is deployed in the second server, and then, after receiving the target prompt word, the second server can input the target prompt word into the media generation model to obtain the media content output by the media generation model; then, the second server can return the media content to the first server, and the first server returns the media content to the terminal device, and the terminal device displays or plays the media content.

[0139] It should be understood that the first server and the second server can be the same server, that is, the prompt word adjustment model and the media generation model are deployed on the same server, and the first adjustment processing and media content generation are completed by the same server, and this application does not impose any restrictions on this.

[0140] It should be understood that the prompt word adjustment module can also be deployed in the terminal device. In this way, after the terminal device receives the original prompt word input by the user, it can call the local prompt word adjustment module to perform a first adjustment process on the original prompt word input by the user, obtain the target prompt word and send the target prompt word to the second server.

[0141] It should be understood that the media generation model can also be deployed in the terminal device. In this way, after the terminal device receives the target prompt word from the first server, it can input the target prompt word into the locally stored media generation model to obtain the media content output by the media generation model; then, the media content can be played or displayed.

[0142] It should be understood that both the prompt word adjustment module and the media generation model can be deployed in the terminal device; in this way, the terminal device performs the first adjustment processing and media content generation locally.

[0143] This application is described by taking the example that the prompt word adjustment module is deployed on the first server and the media generation model is deployed on the second server.

[0144] It should be noted that the prompt word adjustment module can be used to execute the prompt word adjustment method involved in this application; the prompt word adjustment module can be implemented by hardware or software, and this application does not limit this.

[0145] For example, media content can also be called digital content. Digital content is the content of different content types such as text, images, and sounds in digital form. It can be stored on digital carriers such as CDs and hard disks, and spread through the Internet and other means. Digital content is the overall product or service that integrates images, text, audio and video through digital technology. It is the product of the combination of digital media technology and cultural creativity.

[0146] Digital technology is a science and technology that is closely associated with electronic computers. It refers to the technology that uses certain equipment to convert various information, including pictures, text, sound, and images, into binary numbers "0" and "1" that can be recognized by electronic computers for calculation, processing, storage, transmission, dissemination, and restoration. Since computers are used to encode, compress, and decode information in the calculation and storage process, it is also called digital technology, computer digital technology, etc. Digital technology is also called digital control technology.

[0147] The media content may include, but is not limited to, text, images, graphics, audio, and video, etc., which are not limited in this application.

[0148] Exemplarily, the media generation model is an AI model (or a neural network or a neural network model). The media generation model model of the present application can be implemented by at least one of the following neural networks: deep neural networks (DNN), convolutional neural networks (CNN), recurrent neural networks (RNN), residual networks, neural networks using transformer models or other neural networks, etc., and the present application does not limit this.

[0149] Exemplarily, the media generation model can be used to generate media content (the media content referred to in this application may refer to AIGC). The types of media generation models may include, but are not limited to: image generation models, audio generation models, video generation models, text generation models, graphics generation models, etc., which are not limited in this application. For example, ChatGPT (a natural language processing tool driven by artificial intelligence technology), Midjourney (an AI drawing tool), etc.

[0150] Exemplarily, prompt words (which may include original prompt words and target prompt words), i.e., prompts, may also be referred to as media generation prompt words, prompt words of media generation models, or AI prompt words, which may be used to guide (instruct) or inspire media generation models to complete specific tasks. Prompts can help media generation models better understand the intent of the input and make corresponding responses such as outputting media content (i.e., AIGC). Among them, prompts may be composed of multiple words, phrases, or short sentences.

[0151] Exemplarily, when the prompt word is a prompt word for generating an image, the prompt word can be input into an image generation model, and the image generation model generates an image. Exemplarily, when the prompt word is a prompt word for generating an audio, the prompt word can be input into an audio generation model, and the audio generation model generates an audio. Exemplarily, when the prompt word is a prompt word for generating a video, the prompt word can be input into a video generation model, and the video generation model generates a video. Exemplarily, when the prompt word is a prompt word for generating a text, the prompt word can be input into a text generation model, and the text generation model generates a text. Exemplarily, when the prompt word is a prompt word for generating a graphic, the prompt word can be input into a graphic generation model, and the graphic generation model generates a graphic.

[0152] Figure 1 The figure is a schematic diagram of an exemplary application scenario.

[0153] Reference Figure 1 , for example, a user can input the original prompt word "a cute little girl" for generating an image / video in a mobile phone; then, the mobile phone can send the original prompt word "a cute little girl" input by the user to the first server. Then, the first server can perform a first adjustment process on the original prompt word input by the user to obtain the target prompt word "a cute little girl, wearing a beautiful skirt, walking on the street, kawaii anime style, sunset, ultra-high definition, 8k"; then, send the target prompt word to the second server. Subsequently, the second server can input the target prompt word "a cute little girl, wearing a beautiful skirt, walking on the street, kawaii anime style, sunset, ultra-high definition, 8k" into the image generation model to obtain the image / video output by the image generation model; then, the server can return the image / video to the mobile phone, and the mobile phone can play the image / video. In this way, the user can get the desired image / video.

[0154] Continue to refer to Figure 1For example, a user can input the original prompt words "rainy day, soothing and quiet" for generating audio in a laptop computer; then, the laptop computer can send the original prompt words "rainy day, soothing and quiet, male voice" input by the user to the first server. Then, the first server can perform a first adjustment process on the original prompt words input by the user to obtain the target prompt words "soothing and quiet rainy day, folk style, male voice"; then, the target prompt words are sent to the second server. Subsequently, the second server can input the target prompt words "soothing and quiet rainy day, folk style, male voice" into the audio generation model to obtain the audio output by the audio generation model; then, the server can return the audio to the laptop computer, and the laptop computer can play the audio. In this way, the user can get the desired audio.

[0155] Continue to refer to Figure 1 For example, the user can input the original prompt word "moonlight" for generating text in the tablet; then, the tablet can send the original prompt word "moonlight" input by the user to the first server. Then, the first server can perform a first adjustment process on the original prompt word input by the user to obtain the target prompt word "bright moonlight on the sea surface, prose style"; then, the target prompt word is sent to the second server. Subsequently, the second server can input the target prompt word "bright moonlight on the sea surface, prose style" into the text generation model to obtain the audio output by the text generation model; then, the server can return the text to the tablet, and the tablet can display the text. In this way, the user can get the required text.

[0156] It should be understood that the user can also enter the original prompt word in other terminal devices (such as smart TVs, wearable devices, etc.); and the user can also enter the original prompt word for generating other media content, and the corresponding first server can return other media content to the terminal device. This application does not impose any restrictions on this.

[0157] It should be understood that the specific implementation form of the first server and the second server in the present application can be a cloud server, a physical (independent) server, a station cluster server, etc., and the present application does not impose any restrictions on this.

[0158] In this way, by adjusting the original prompt words input by the user, the form and content of prompt words with better quality can be obtained, thereby improving the quality of media content generated by the media generation model according to the target prompt words.

[0159] The following is an explanation of the adjustment process of the prompt words in this application.

[0160] Figure 2 FIG. 1 is a schematic diagram showing an exemplary process of adjusting the prompt words. Figure 2 The embodiment is Figure 1Based on the explanation, that is to say Figure 2 The embodiment is described by taking S201 to S204 being executed by the first server as an example.

[0161] S201, obtaining the original prompt word input by the user.

[0162] Figure 3A The figure is a schematic diagram of a mobile phone interface shown as an example.

[0163] Reference Figure 3A For example, 301 is the main interface of the mobile phone, and the main interface of the mobile phone includes one or more controls, including but not limited to: application icons (for example, application icons of Huawei video applications, application icons of browser applications, application icons 302 of AI-generated applications), network logos, power logos, etc.

[0164] Continue to refer to Figure 3A For example, when a user needs to use a media generation model to generate media content, or needs to adjust a prompt word (also called rewriting a prompt word), the user can click on an application icon 302 of an AI-generated application; the mobile phone displays an application interface 303 of the AI-generated application in response to the user's operation, such as Figure 3B shown.

[0165] Reference Figure 3B Exemplarily, the application interface 303 of the AI ​​generation application may include one or more controls, including but not limited to: AI image generation option 304, AI video generation option, AI audio generation option, AI text generation option, etc., which is not limited in this application.

[0166] Continue to refer to Figure 3B For example, when the original prompt word for generating an image is to be input, the user can click on the AI ​​image generation option 304, and the mobile phone displays the AI ​​image generation interface 305 in response to the user's operation behavior, such as Figure 3C shown.

[0167] Reference Figure 3C , exemplarily, the AI ​​image generation interface 305 includes one or more controls, including but not limited to: a prompt word input box 306, a submit option 307, etc., which are not limited in this application. The user can enter the original prompt word in the prompt word input box 306, such as Figure 3D As shown, the original prompt word input by the user in the prompt word input box 306 is "a cute little girl".

[0168] For example, when a user clicks Figure 3DAfter selecting the submit option 307, the mobile phone can send the original prompt word "a cute little girl" input by the user to the first server in response to the user's operation behavior; in this way, the first server can obtain the original prompt word input by the user, and then execute S202 to S204.

[0169] It should be understood that when the original prompt words for generating a video are to be input, the user can click on the AI ​​video generation option, and the mobile phone will display the AI ​​video generation interface in response to the user's operation; then, the user can enter the original prompt words in the prompt word input box of the AI ​​video generation interface and click the Submit option.

[0170] When the original prompt word for generating audio is to be input, the user can click the AI ​​audio generation option, and the mobile phone will display the AI ​​audio generation interface in response to the user's operation; then, the user can enter the original prompt word in the prompt word input box of the AI ​​audio generation interface and click the Submit option.

[0171] When the original prompt word for generating text is to be input, the user can click the AI ​​text generation option, and the mobile phone will display the AI ​​text generation interface in response to the user's operation; then, the user can enter the original prompt word in the prompt word input box of the AI ​​text generation interface and click the Submit option.

[0172] S202: Recognize and process the original prompt word input by the user to obtain structured information, wherein the structured information includes multiple information items, and the value of a single information item is empty or at least part of the original prompt word.

[0173] For example, the prompt words (hereinafter referred to as reference prompt words) required for the media generation model to generate high-quality media content can be analyzed in advance to determine the structure of the reference prompt words. Specifically, the information items included in the reference prompt words can be determined. Then, according to the types of information items included in the reference prompt words, the original prompt words input by the user can be identified and processed to obtain a structure containing multiple information items (for example, Figure 2 , where N information items are shown, and N is an integer greater than 1); wherein the types of information items included in the structured information are the same as the types of information items included in the reference prompt words.

[0174] Exemplarily, a single information item in the structured information may include a name of the information item and a value of the information item; the name of the information item is also the type name of the information item. The value of a single information item in the structured information is empty or is at least part of the original prompt word.

[0175] For example, if the reference prompt word is a prompt word that indicates that the image generation model generates an image, the types of information items contained in the reference prompt word may include: subject, style, light source, and image quality. Assuming that the original prompt word input by the user is "a cute little girl", the obtained structured information may include subject information items, style information items, light and shadow information items, and image quality information items, which may be as follows:

[0176] “Subject: A cute little girl.

[0177] Style: None.

[0178] Light and shadow: None.

[0179] Image quality: None. "

[0180] Among them, the subject information item is "Subject: A cute little girl"; the name of the subject information item is "Subject", and the value of the subject information item (also called subject information) is "A cute little girl", wherein the value of the subject information item is the original prompt word.

[0181] The style information item is "style: none"; the name of the style information item is "style", and the value of the style information item (also referred to as style information) is empty.

[0182] The light and shadow information item is "light and shadow: none"; the name of the light and shadow information item is "light and shadow", and the value of the light and shadow information item (also called light and shadow information) is empty.

[0183] The image quality information item is "image quality: none"; the name of the image quality information item is "image quality", and the value of the image quality information item (also referred to as image quality information) is empty.

[0184] It should be understood that when the reference prompt word is a prompt word that instructs the image generation model to generate an image, the types of information items contained in the reference prompt word may include more or less than those shown above, and the present application does not limit this.

[0185] For example, if the reference prompt word is a prompt word that instructs the audio generation model to generate audio, the types of information items contained in the reference prompt word may include: subject, style, and timbre. Assuming that the original prompt word input by the user is "rainy day, soothing and quiet", the obtained structured information may include subject information items, style information items, and timbre information items, which may be as follows:

[0186] “Subject: Rainy day, soothing and tranquil.

[0187] Style: None.

[0188] Voice: Male. ”

[0189] Among them, the main information item is "Subject: Rainy Day, Soothing and Quiet"; the name of the main information item is "Subject", and the value of the main information item (also called main information) is "Rainy Day, Soothing and Quiet", wherein the value of the main information item is part of the original prompt word.

[0190] The style information item is "style: none"; the name of the style information item is "style", and the value of the style information item (also referred to as style information) is empty.

[0191] The timbre information item is "timbre: male voice"; the name of the timbre information item is "timbre", the value of the timbre information item (also called timbre information) is "male voice", wherein the value of the main information item is part of the original prompt word.

[0192] It should be understood that when the reference prompt word is a prompt word that instructs the audio generation model to generate audio, the types of information items contained in the reference prompt word may include more or less than those shown above, and the present application does not limit this.

[0193] For example, if the reference prompt word is a prompt word that instructs the text generation model to generate text, the types of information items contained in the reference prompt word may include: subject and style. Assuming that the original prompt word input by the user is "moonlight", the obtained structured information may include subject information items and style information items, which may be as follows:

[0194] “Subject: Moonlight.

[0195] Style: None. ”

[0196] Among them, the subject information item is "subject: moonlight"; the name of the subject information item is "subject", the value of the subject information item (also called subject information) is "moonlight", and the subject information item is the original prompt word input by the user.

[0197] The style information item is "style: none"; the name of the style information item is "style", and the value of the style information item (also referred to as style information) is empty.

[0198] It should be understood that when the reference prompt word is a prompt word that instructs the text generation model to generate text, the types of information items contained in the reference prompt word may include more or less than those shown above, and the present application does not limit this.

[0199] For example, the original prompt word input by the user can be input into the language model to obtain structured information. The language model can be a large language model (LLM), which can be composed of an artificial neural network with many parameters (usually billions of weights or more) and trained using self-supervised learning or semi-supervised learning.

[0200] In one possible way, the language model can be one; in this case, the context learning (InContext Learning, ICL) technology can be used to issue recognition tasks to the language model so that the language model can recognize the values ​​of each information item in the structured information from the original prompt words entered by the user. For example, ICL issues a subject information recognition task to LLM, and LLM recognizes and processes the original prompt words entered by the user, and outputs the value of the subject information item (such as "a cute little girl"). Next, ICL issues a style information recognition task to LLM, and LLM recognizes and processes the original prompt words entered by the user, and outputs the value of the style information item. Subsequently, ICL issues a light and shadow information recognition task to LLM, and LLM recognizes and processes the original prompt words entered by the user, and outputs the value of the light and shadow information item. After that, ICL issues a picture quality information recognition task to LLM, and LLM recognizes and processes the original prompt words entered by the user, and outputs the value of the picture quality information item.

[0201] In one possible way, there can be multiple language models; in this case, one language model can be used to identify the value of one or more information items in the structured information. For example, there are four language models: language model A1, language model A2, language model A3, and language model A4; the original prompt word input by the user can be input into language model A1 to obtain the value of the main information item; the original prompt word input by the user can be input into language model A2 to obtain the value of the style information item; the original prompt word input by the user can be input into language model A3 to obtain the value of the light and shadow information item; the original prompt word input by the user can be input into language model A4 to obtain the value of the image quality information item.

[0202] That is to say, the present application does not limit the number of language models for recognition processing.

[0203] Figure 4A and Figure 4B The figure is a schematic diagram of a mobile phone interface shown as an example.

[0204] Reference Figure 4A In one possible manner, after the first server completes S202, the structured information may be returned to the mobile phone; thereafter, the mobile phone may display a structured information interface 401, such as Figure 4AAs shown. Exemplarily, the structured information interface 401 may include one or more controls, including but not limited to: structured information, edit option 402 and continue option 403, etc., which are not limited in this application. In this way, displaying structured information can enable users to understand the defects of the original prompt words they input; further, when the user needs to modify the original prompt words input this time, the displayed structured information can guide the modification direction for the original prompt words, reducing the cost of manual trial and error. In addition, the user can also be prompted for the original prompt words input later to improve the quality of the original prompt words input later by the user.

[0205] Continue to refer to Figure 4A For example, when the user needs the AI ​​generated application to adjust the original prompt word input by the user, the user can click the continue option 403. The mobile phone responds to the user's operation behavior and sends a first adjustment request to the first server. After receiving the first adjustment request, the first server can execute S203. When the user needs to manually modify the original prompt word input by the user, the user can click the edit option 402. The mobile phone responds to the user's operation behavior and displays the first modification interface 404, such as Figure 4B shown.

[0206] Reference Figure 4B Exemplarily, the first modification interface 404 may include one or more controls, including but not limited to: a subject information edit box, a style information edit box, a light and shadow information edit box, a picture quality information edit box and a confirmation option 405.

[0207] Exemplarily, when the user needs to modify the value of a subject information item, the value of the subject information item can be edited in the subject information edit box. When the user needs to modify the value of a style information item, the value of the style information item can be edited in the style information edit box. When the user needs to modify the value of a light and shadow information item, the value of the light and shadow information item can be edited in the light and shadow information edit box. When the user needs to modify the value of an image quality information item, the value of the image quality information item can be edited in the image quality information edit box.

[0208] For example, after the user completes the modification, the user can click the confirmation option 405, and the mobile phone responds to the user's operation behavior and returns to the structured information interface 401, such as Figure 4A Next, the user may click on the continue option 403, and the mobile phone sends a first adjustment request to the first server in response to the user's operation behavior; after receiving the first adjustment request, the first server may execute S203.

[0209] Exemplarily, the first adjustment request may include a target information item adjusted by the user. If the user edits the value of the subject information item in the subject information editing box, the subject information item is the target information item; if the user edits the value of the style information item in the style information editing box, the style information item is the target information item, and so on.

[0210] S203: Perform a first adjustment process on the information items in the structured information that do not meet the media generation condition.

[0211] Exemplarily, before executing S203, the following S1 and S2 may be executed:

[0212] S1, performing a first evaluation process on each information item in the structured information to obtain first evaluation information corresponding to each information item in the structured information.

[0213] Exemplarily, the first evaluation process may be performed on each information item in the structured information to obtain the first evaluation information corresponding to each information item in the structured information. Specifically, for an information item in the structured information, the first evaluation process may be performed on the value of the information item to obtain the first evaluation information corresponding to the information item.

[0214] Exemplarily, each information item in the structured information can be input into a language model to obtain first evaluation information corresponding to each information item in the structured information; wherein the language model can be an LLM. It should be noted that the language model that outputs the first evaluation information is a different language model from the language model that outputs the structured information.

[0215] In one possible way, the language model can be one; in this case, the ICL technology can be used to issue an evaluation task to the language model so that the language model performs a first evaluation process on each information item in the structured information. For example, ICL issues a subject information evaluation task to LLM, and LLM performs a first evaluation process on the value of the subject information item and outputs the first evaluation information corresponding to the subject information item. Next, ICL issues a style information evaluation task to LLM, and LLM performs a first evaluation process on the value of the style information item and outputs the first evaluation information corresponding to the style information item. Subsequently, ICL issues a light and shadow information evaluation task to LLM, and LLM performs a first evaluation process on the value of the light and shadow information item and outputs the first evaluation information corresponding to the light and shadow information item. Afterwards, ICL issues an image quality information evaluation task to LLM, and LLM performs a first evaluation process on the value of the image quality information item and outputs the first evaluation information corresponding to the image quality information item.

[0216] In one possible way, there may be multiple language models; in this case, one language model can be used to output first evaluation information corresponding to one or more information items in the structured information. For example, there are four language models: language model B1, language model B2, language model B3, and language model B4; the value of the main information item in the structured information can be input into the language model B1 to obtain the first evaluation information corresponding to the main information item; the value of the style information item in the structured information can be input into the language model B2 to obtain the first evaluation information corresponding to the style information item; the value of the light and shadow information item in the structured information can be input into the language model B3 to obtain the first evaluation information corresponding to the light and shadow information item; the value of the image quality information item in the structured information can be input into the language model B4 to obtain the first evaluation information corresponding to the image quality information item.

[0217] Exemplarily, the first evaluation process includes but is not limited to scoring, completeness evaluation / matching evaluation, and grade evaluation. The corresponding first evaluation information may include but is not limited to: score, completeness / matching evaluation result, and grade information. Among them, the matching evaluation may be an evaluation of the matching degree between the original prompt word and the media generation model. This application takes the first evaluation process as scoring and the first evaluation information as score as an example for explanation.

[0218] In one possible manner, the first evaluation process may be a score, and the first evaluation information may include a score. For example, for the main information item, the score is given from 50% of the subject, 25% of the details, and 25% of the background; for the style information item, the light and shadow information item, and the image quality information item, the score is 0 or 1, where "0" indicates that the value of the information item is empty, and "1" indicates that the value of the information item is not empty.

[0219] For example, the structured information is:

[0220] “Subject: A cute little girl.

[0221] Style: None.

[0222] Light and shadow: None.

[0223] Image quality: None. "

[0224] The scores corresponding to each information item in the structured information may be as follows:

[0225] “Main body: 20%.

[0226] Style:0.

[0227] Light and shadow: 0.

[0228] Image quality: 0."

[0229] For example, the structured information is:

[0230] “Subject: Soothing and tranquil.

[0231] Style: None.

[0232] Voice: Male. ”

[0233] The scores corresponding to each information item in the structured information may be as follows:

[0234] “Main body: 80%.

[0235] Style:0.

[0236] Tone: 1. "

[0237] For example, the structured information is:

[0238] “Subject: Moonlight.

[0239] Style: None. ”

[0240] The scores corresponding to each information item in the structured information may be as follows:

[0241] “Main body: 15%.

[0242] Style: 0. ”

[0243] Figure 4C to Figure 4E The figure is a schematic diagram of a mobile phone interface shown as an example.

[0244] In one possible manner, after the first server determines the first evaluation information corresponding to each information item in the structured information, the first evaluation information corresponding to each information item in the structured information may be returned to the mobile phone; thereafter, the mobile phone may display an evaluation interface 406, such as Figure 4C The evaluation interface 406 may include one or more controls, including but not limited to: evaluation information, edit option 407 and continue option 408 .

[0245] In one possible manner, after the first server determines the first evaluation information corresponding to each information item in the structured information, the first evaluation information corresponding to each information item in the structured information may be returned to the mobile phone; thereafter, the mobile phone may display an evaluation interface 409, such as Figure 4C The evaluation interface 409 may include one or more controls, including but not limited to: first evaluation information, an edit option 410 and a continue option 411 .

[0246] In this way, by displaying the first evaluation information corresponding to the original prompt word, the user can understand the more fine-grained defects of the original prompt word he or she inputs; further, when the user needs to modify the original prompt word input this time, the displayed first evaluation information can guide the modification direction of the prompt word, reducing the cost of manual trial and error. In addition, the user can also be prompted for the original prompt word input later to improve the quality of the original prompt word input later by the user.

[0247] Continue to refer to Figure 4C or Figure 4D For example, when the user needs the AI ​​generated application to adjust the original prompt word input by the user, the user can click the continue option 408 or 411. The mobile phone responds to the user's operation behavior and sends a first adjustment request to the first server. After receiving the first adjustment request, the first server can execute S2. When the user needs to manually modify the original prompt word input by the user, the user can click the edit option 407 or 410. The mobile phone responds to the user's operation behavior and displays the second modification interface 412, such as Figure 4E shown.

[0248] Reference Figure 4E Exemplarily, the second modification interface 412 may include one or more controls, including but not limited to: a subject information edit box, a style information edit box, a light and shadow information edit box, a picture quality edit box and a confirmation option 413.

[0249] Exemplarily, when the user needs to modify the value of a subject information item, the value of the subject information item can be edited in the subject information edit box. When the user needs to modify the value of a style information item, the value of the style information item can be edited in the style information edit box. When the user needs to modify the value of a light and shadow information item, the value of the light and shadow information item can be edited in the light and shadow information edit box. When the user needs to modify the value of an image quality information item, the value of the image quality information item can be edited in the image quality information edit box.

[0250] For example, after the user completes the modification, the user can click the confirmation option 413, and the mobile phone responds to the user's operation behavior and returns to the evaluation interface 406 or 409, such as Figure 4C Then, the user may click on the continue option 408 or 411, and the mobile phone sends a second adjustment request to the first server in response to the user's operation behavior; after receiving the second adjustment request, the first server may execute S2.

[0251] S2: judging whether the information items in the structured information meet the media content generation condition according to the first evaluation information corresponding to each information item in the structured information.

[0252] Exemplarily, the present application may pre-set media content generation conditions; for example, the media content generation conditions are: the score corresponding to the information item in the structured information is higher than or equal to the corresponding threshold. The threshold can be set as required, and the present application does not limit this. A corresponding threshold can be set for each information item, for example, the threshold corresponding to the subject information item is set to 85%, the threshold corresponding to the style information item is set to 1, the threshold corresponding to the light and shadow information is set to 1, and the threshold corresponding to the image quality information item is set to 1.

[0253] Furthermore, for an information item in the structured information, it can be determined whether the score corresponding to the information item is higher than or equal to the corresponding threshold value to determine whether the information item meets the media content generation condition. For information items that meet the media content generation condition, the first adjustment process may not be performed; for information items that do not meet the media content generation condition, the first adjustment process may be performed, that is, S203 is executed.

[0254] For example, based on the scores corresponding to each information item in the structured information shown in the above example, it can be determined that the score corresponding to the subject information item is lower than the corresponding threshold, the score corresponding to the style information item is lower than the corresponding threshold, the score corresponding to the light and shadow information item is lower than the corresponding threshold, and the score corresponding to the image quality information item is lower than the corresponding threshold; therefore, the first adjustment processing can be performed on the subject information item, style information item, light and shadow information item, and image quality information item.

[0255] For example, based on the scores corresponding to each information item in the structured information shown in the above example, it can be determined that the score corresponding to the main information item is lower than the corresponding threshold, the score corresponding to the style information item is lower than the corresponding threshold, and the score corresponding to the timbre information item is equal to the corresponding threshold; therefore, the first adjustment processing can be performed on the main information item and the style information item.

[0256] For example, based on the scores corresponding to each information item in the structured information shown in the above example, it can be determined that the score corresponding to the main information item is lower than the corresponding threshold, and the score corresponding to the style information item is lower than the corresponding threshold; therefore, the first adjustment processing can be performed on the main information item and the style information item.

[0257] It should be noted that performing the first adjustment processing on the information item actually refers to performing the first adjustment processing on the value of the information item. For example, performing the first adjustment processing on the subject information item may refer to performing the first adjustment processing on the value of the subject information item, performing the first adjustment processing on the style information item may refer to performing the first adjustment processing on the value of the style information item, performing the first adjustment processing on the light and shadow information item may refer to performing the first adjustment processing on the value of the light and shadow information item, and performing the first adjustment processing on the image quality information item may refer to performing the first adjustment processing on the value of the image quality information item.

[0258] Exemplarily, the first adjustment process may include at least one of the following: text enhancement process or keyword extraction process. The text enhancement process may include: adding detailed description of the text, enriching the text, etc. The keyword extraction process may include extracting keywords from the original prompt words.

[0259] Exemplarily, the value of the subject information item can be analyzed to determine the richness of the value of the subject information item and whether additional details are needed; when it is determined that the richness of the value of the subject information item is lower than a preset value, the value of the subject information item can be enriched; when it is determined that additional details are needed, detailed information of the value of the subject information item can be added.

[0260] Exemplarily, for other information items (except the main information item) whose values ​​are empty, the first adjustment process can be performed on the values ​​of the other information items by semantic analysis of the values ​​of the main information items.

[0261] For example, for the original prompt words used to generate the image, the subject category can be identified based on the semantics of the value of the subject information item; then, the value of the corresponding style information item is matched based on the subject category. For example, if the subject is a person, the high-definition realistic style can be matched; for another example, if the subject is a building, the Gothic style can be matched.

[0262] For example, for the original prompt word used to generate the image, the value of the light and shadow information item can be matched according to the semantics of the value of the main information item; for example, if the value of the main information item is "evening", the value of the matched light and shadow information item is "sunset"; for another example, if the value of the main information item is "outdoor night", the value of the matched light and shadow information item is "sky full of stars".

[0263] For example, for the original prompt word used to generate the image, the value of the image quality information item may be matched according to the semantics of the value of the subject information item.

[0264] Exemplarily, redundancy analysis can be performed on information items whose values ​​are not empty to determine the redundancy of the value of the information item. When the redundancy of the value of the information item is higher than a threshold, keyword extraction processing can be performed on the value of the information item. The information items to be subjected to redundancy analysis include other information items and main information items; when the richness of the main information item is higher than a preset value, redundancy analysis can be performed on the main information item.

[0265] For example, an adjustment model (the adjustment model may be implemented by a neural network) may be used to perform a first adjustment process on the information item.

[0266] In one possible approach, a group of adjustment models may be used to perform a first adjustment process on multiple information items that do not meet media generation conditions; wherein a group of adjustment models may include one or more adjustment models, wherein one adjustment model may be used to perform a first adjustment process on one or more information items that do not meet media generation conditions.

[0267] In one possible manner, multiple groups of adjustment models may be pre-set, and one group of adjustment models may be selected from the multiple groups of adjustment models each time to perform the first adjustment processing on multiple information items that do not meet the media generation condition; wherein one group of adjustment models may include one or more adjustment models, wherein one adjustment model may be used to perform the first adjustment processing on one or more information items that do not meet the media generation condition. wherein, among the multiple groups of adjustment models, the adjustment models used to perform the first adjustment processing on the same information item that does not meet the media generation condition may be adjustment models with different numbers of model parameters, and different adjustment efficiencies and adjustment qualities.

[0268] Exemplarily, a server or a terminal device may select a group of adjustment models from multiple groups of adjustment models according to the device capability.

[0269] Exemplarily, a group of adjustment models may also be selected from multiple groups of adjustment models by a user.

[0270] Figure 5A and Figure 5B The figure is a schematic diagram of a mobile phone interface shown as an example.

[0271] For example, when a user clicks Figure 4C or Figure 4D After selecting the continue option 408 or 411 in the example, the mobile phone may display a performance selection interface 501 (or 502) of the prompt word in response to the user's operation behavior, such as Figure 5A (or Figure 5B ) as shown.

[0272] Reference Figure 5A , exemplarily, the performance selection interface 501 includes multiple performance adjustment options: an extreme speed mode option, a balanced mode option, and a high performance mode option.

[0273] Among them, the number of model parameters of each adjustment model in a group of adjustment models corresponding to the extreme speed mode option is less than the number of model parameters of each adjustment model in a group of adjustment models corresponding to the balanced model option; the number of model parameters of each adjustment model in a group of adjustment models corresponding to the balanced model option is less than the number of model parameters of each adjustment model in a group of adjustment models corresponding to the high performance mode. For example, the number of model parameters of each adjustment model in a group of adjustment models corresponding to the extreme speed mode option is less than 10 billion, the number of model parameters of each adjustment model in a group of adjustment models corresponding to the extreme speed mode option is greater than 10 billion and less than 100 billion, and the number of model parameters of each adjustment model in a group of adjustment models corresponding to the high performance mode option is greater than 100 billion.

[0274] Reference Figure 5B , illustratively, the performance selection interface 502 includes multiple adjustment performance options: model-1b option, model-10b option and model-100b option.

[0275] Among them, the number of model parameters of each adjustment model in the set of adjustment models corresponding to the model-1b option is less than 10 billion, the number of model parameters of each adjustment model in the set of adjustment models corresponding to the model-10b option is greater than 10 billion and less than 100 billion, and the number of model parameters of each adjustment model in the set of adjustment models corresponding to the model-100b option is greater than 100 billion.

[0276] Reference Figure 5A and Figure 5B For example, when the user needs to obtain a large number of target prompt words, the user can click the extreme speed mode option or the model-1b option. When the user needs to obtain a small number of target prompt words with better quality, the user can click the high performance mode option or the model-100b option. When the user wants both higher adjustment efficiency and better quality of the target prompt words, the user can click the balanced mode option or the model-10b option.

[0277] Continue to refer to Figure 5A and Figure 5B For example, after the user clicks on the target adjustment performance option, the mobile phone can send adjustment performance information to the first server in response to the user's operation behavior; wherein the adjustment performance information can include an identifier of a group of adjustment models corresponding to the target adjustment performance option. After receiving the adjustment performance information, the first server can determine a group of adjustment models to be called according to the adjustment performance information; then, the group of adjustment models is called to perform a first adjustment process on the information items in the structured information that do not meet the media content generation condition.

[0278] For example, after the first adjustment processing (text enhancement processing) is performed on the subject information item "a cute little girl" in the above structured information, the subject information item after the first adjustment processing is "subject: a cute little girl, wearing a beautiful skirt, walking on the street". After the first adjustment processing (text enhancement processing) is performed on the style information item "style: none" in the above structured information, the style information item after the first adjustment processing is "style: kawaii anime style". After the first adjustment processing (text enhancement processing) is performed on the light and shadow information item "light and shadow: none" in the above structured information, the light and shadow information item after the first adjustment processing is "light and shadow: sunset". After the first adjustment processing (text enhancement processing) is performed on the image quality information item "image quality: none" in the above structured information, the light and shadow information item after the first adjustment processing is "image quality: ultra high definition, 8k".

[0279] For example, after the first adjustment processing (text enhancement processing) is performed on the subject information item "Rainy day, soothing and quiet" in the above structured information, the subject information item after the first adjustment processing is "Subject: Soothing and quiet rainy day". After the first adjustment processing (text enhancement processing) is performed on the style information item "Style: None" in the above structured information, the style information item after the first adjustment processing is "Style: Folk style".

[0280] For example, after the first adjustment processing (text enhancement processing) is performed on the subject information item "moonlight" in the above structured information, the subject information item after the first adjustment processing is "subject: bright moonlight on the sea surface". After the first adjustment processing (text enhancement processing) is performed on the style information item "style: none" in the above structured information, the style information item after the first adjustment processing is "style: prose style".

[0281] For example, after the first adjustment processing (keyword extraction processing) is performed on the subject information item "a cute little girl who is as energetic as the rising sun in the morning, wearing a very beautiful skirt, strolling happily on the street" in the above-mentioned structured information, the subject information item after the first adjustment processing is "subject: a cute little girl, wearing a beautiful skirt, strolling on the street". After the first adjustment processing (text enhancement processing) is performed on the style information item "style: none" in the above-mentioned structured information, the style information item after the first adjustment processing is "style: kawaii anime style". After the first adjustment processing (text enhancement processing) is performed on the light and shadow information item "light and shadow: none" in the above-mentioned structured information, the light and shadow information item after the first adjustment processing is "light and shadow: sunset". After the first adjustment processing (text enhancement processing) is performed on the image quality information item "image quality: none" in the above-mentioned structured information, the light and shadow information item after the first adjustment processing is "image quality: ultra-high definition, 8k".

[0282] For example, after the first adjustment process (keyword extraction process) is performed on the subject information item "a cute little girl who is as lively as the rising sun in the morning, wearing a very beautiful skirt, strolling happily on the street" in the above structured information, the subject information item after the first adjustment process is "subject: a cute little girl, wearing a beautiful skirt, strolling on the street". The light and shadow information items "light and shadow: sunset", "image quality: ultra high definition, 8k", and "style: kawaii anime style" in the above structured information do not need to be adjusted by the first adjustment process.

[0283] S204: Integrate the information items in the structured information that meet the media content generation condition and the information items that have undergone the first adjustment process to obtain a target prompt word, wherein the target prompt word is used to instruct the media generation model to generate media content.

[0284] Exemplarily, the target prompt word may be obtained by splicing the information items that have undergone the first adjustment process and the information items that meet the media content generation condition in the structured information according to the prompt word structure that matches the media generation model.

[0285] The prompt word structure that matches the media generation model may be the structure of the reference prompt words mentioned above.

[0286] In a possible manner, the value of the information item that has undergone the first adjustment process and the value of the information item that meets the media content generation condition in the structured information may be integrated to obtain the target prompt word.

[0287] For example, by integrating the values ​​of the information items that have been processed by the first adjustment shown in the above example and the values ​​of the information items that meet the media content generation conditions, the target prompt words obtained can be "a cute little girl, wearing a beautiful skirt, walking on the street, kawaii anime style, sunset, ultra-high definition, 8k", "soothing and quiet rainy day, folk style, male voice", "bright moonlight on the sea, prose style".

[0288] In a possible manner, the target prompt word may be obtained by integrating the information item itself that has undergone the first adjustment processing and the information item itself that meets the media content generation condition in the structured information.

[0289] For example, the target cue word could be:

[0290] “Subject: A cute little girl, wearing a beautiful dress, walking down the street.

[0291] Style: Kawaii anime style.

[0292] Light and shadow: sunset.

[0293] Image quality: Ultra HD, 8k.”

[0294] For another example, the target prompt word can be:

[0295] “Subject: A soothing and peaceful rainy day.

[0296] Style: Folk style.

[0297] Voice: Male. ”

[0298] For example, the target prompt word can be:

[0299] “Subject: The bright moonlight on the sea.

[0300] Style: Prose style. ”

[0301] Figure 5C and Figure 5D The figure is a schematic diagram of a mobile phone interface shown as an example.

[0302] Reference Figure 5C For example, after the first server completes S204 and obtains the target prompt word, the target prompt word can be returned to the mobile phone; then, the mobile phone can display the prompt word adjustment result interface 503. The prompt word adjustment result interface 503 may include one or more controls, including but not limited to: target prompt word, confirmation option, etc., which are not limited in this application. Figure 5C One representation of the target cue word “a cute little girl, wearing a beautiful dress, walking on the street, kawaii anime style, sunset, ultra high definition, 8k” is shown in Figure 2.

[0303] Reference Figure 5D For example, after the first server completes S204 and obtains the target prompt word, the target prompt word can be returned to the mobile phone; then, the mobile phone can display the prompt word adjustment result interface 504. The prompt word adjustment result interface 504 may include one or more controls, including but not limited to: target prompt word, confirmation option, etc., which are not limited in this application. Figure 5D A representation of the target cue word is shown in:

[0304] “Subject: A cute little girl, wearing a beautiful dress, walking down the street.

[0305] Style: Kawaii anime style.

[0306] Light and shadow: sunset.

[0307] Image quality: Ultra HD, 8k.”

[0308] Figure 5E and Fig. 5F The figure is a schematic diagram of a mobile phone interface shown as an example.

[0309] Reference Figure 5EFor example, after the first server completes S204 to obtain the target prompt word, the target prompt word can be returned to the mobile phone; then, the mobile phone can display the prompt word adjustment result interface 505. The prompt word adjustment result interface 505 may include one or more controls, including but not limited to: the target prompt word, the edit option 506, the confirmation option, etc., which is not limited in this application.

[0310] Continue to refer to Figure 5E For example, when the user needs to modify the target prompt word, the user can click the edit option 506 in the prompt word adjustment result interface 505, and the mobile phone can display an edit box and a cursor in response to the user's operation behavior. In this way, the user can edit the target prompt word in the edit box.

[0311] Reference Fig. 5F For example, after the first server completes S204 to obtain the target prompt word, the target prompt word can be returned to the mobile phone; then, the mobile phone can display the prompt word adjustment result interface 507. The prompt word adjustment result interface 507 may include one or more controls, including but not limited to: multiple edit boxes, confirmation options, etc., which are not limited in this application.

[0312] Continue to refer to Fig. 5F For example, each edit box displays a portion of the target prompt word, wherein the information displayed in each edit box corresponds to the value of an information item in the structured information (including the information item that has not been processed by the first adjustment and the information item that has been processed by the first adjustment). When the user needs to modify the target prompt word, he can click any edit box to edit the information displayed in the edit box.

[0313] It should be understood that the prompt word adjustment result interface 507 may not display an edit box, but display the target prompt word, and display different parts of the target prompt word in different fonts or colors; wherein different parts of the target prompt word correspond to values ​​of different information items in the structured information (including information items that have not been processed by the first adjustment and information items that have been processed by the first adjustment). When the user needs to modify the target prompt word, he can directly click on the target prompt word to edit it.

[0314] It should be understood that the present application does not limit the manner in which the user modifies the target prompt word.

[0315] Continue to refer to FIG. 5C to FIG. 5FFor example, the user can click on the confirmation option in the prompt word adjustment interface 503 or 504, and the mobile phone sends a first image generation request to the first server in response to the user's operation behavior; after receiving the first image generation request, the first server can generate a second image generation request; wherein the second image generation request includes the target prompt word. Then, the first server can send the second image generation request to the second server; after receiving the second image generation request, the second server can extract the target prompt word from the second image generation request, input the target prompt word into the corresponding image generation model, obtain the image and return it to the first server; the first server sends the image returned by the second server to the mobile phone, and the mobile phone can display the image, such as Figure 5G As shown (where Figure 5G The image is generated by using the prompt word adjustment method of the present application to process the original prompt word "a cute little girl" input by the user to obtain the target prompt word ("a cute little girl, wearing a beautiful skirt, walking on the street, kawaii anime style, sunset, ultra high definition, 8k").

[0316] Figure 5H This is an image generated by using the existing prompt word adjustment method to process the original prompt word "a cute little girl" input by the user to obtain the target prompt word. Figure 5G and Figure 5H , Figure 5G The image quality is better. This is because the prior art directly processes the original prompt words input by the user, while the present application first identifies structured information based on the original prompt words input by the user, and then performs the first adjustment processing on the information items that do not meet the media generation conditions; that is, the present application divides the original prompt words into fine-grained information, and then adjusts the information of each granularity; it can be seen that the adjustment granularity of the original prompt words input by the user in the present application is finer and more comprehensive, which can make the target prompt word text semantics richer and more complete (that is, the quality of the target prompt words is better), thereby improving the quality of the media content generated by the media generation model.

[0317] In addition, the present application converts different original prompt words into preset structured information before making adjustments, which can increase the controllability of the target prompt words. In this way, the controllability of the quality of the media content generated by the media generation model can be improved.

[0318] Figure 6 The figure is a schematic diagram of the adjustment process of the prompt words shown as an example.

[0319] S601, obtaining the original prompt word input by the user;

[0320] S602, identifying and processing the original prompt word input by the user to obtain structured information, wherein the structured information includes multiple information items, and the value of a single information item is empty or at least part of the original prompt word;

[0321] S603, performing a first adjustment process on the information items in the structured information that do not meet the media generation condition;

[0322] S604, integrating the information items satisfying the media content generation condition in the structured information and the information items subjected to the first adjustment process to obtain a target prompt word, wherein the target prompt word is used to instruct the media generation model to generate the media content;

[0323] Exemplarily, S601 to S604 may refer to the description of S201 to S204 above, which will not be repeated here.

[0324] It should be noted that in Figure 6 In the embodiment of the present invention, after the first server executes S604, it may not return the target prompt word to the terminal device, but continue to execute S605 to S607 to generate a prompt word with better quality.

[0325] S605: Input the target prompt word into the media generation model to obtain media content.

[0326] Exemplarily, the first server may send the target prompt word to the second server, and the second server may input the target prompt word into the media generation model to obtain the media content and return the media content to the first server; thereafter, the first server may execute S606.

[0327] S606: Perform a second evaluation process on the media content to obtain second evaluation information corresponding to the media content.

[0328] Exemplarily, the first server may perform a second evaluation process on the media content to obtain second evaluation information corresponding to the media content. Exemplarily, the second evaluation process may be performed on the media content from multiple dimensions such as security, aesthetics, and semantic consistency to obtain a security score corresponding to the media content, an aesthetic score corresponding to the media content, and a semantic consistency score corresponding to the media content. In one possible manner, the second evaluation information corresponding to the media content may include: a security score corresponding to the media content, an aesthetic score corresponding to the media content, and a semantic consistency score corresponding to the media content.

[0329] Exemplarily, the first server may call the evaluation model to perform second evaluation processing on the media content to obtain second evaluation information corresponding to the media content.

[0330] S607: Determine whether the media content meets the media content quality condition according to the second evaluation information corresponding to the media content.

[0331] Exemplarily, the media content quality condition may be preset, for example, the media content quality condition is: the security score is greater than the security threshold, the aesthetic score is greater than the aesthetic threshold, and the semantic consistency score is greater than the consistency threshold.

[0332] Specifically, it can be determined whether the security score corresponding to the media content is greater than the security threshold, whether the aesthetic score of the media content is greater than the aesthetic threshold, and whether the semantic consistency score corresponding to the media content is greater than the consistency threshold; when the security score corresponding to the media content is greater than the security threshold, and the aesthetic score of the media content is greater than the aesthetic threshold, and the semantic consistency score corresponding to the media content is greater than the consistency threshold, it can be determined that the media content meets the media content quality condition, at which time the process can be terminated, and the first server can return the target prompt word to the terminal device for display.

[0333] Otherwise, when it is determined that the media content does not meet the media content quality condition, the process may return to S603. In the process of returning to S603, the information items in the structured information that do not meet the media content generation condition may be re-adjusted according to the second evaluation information corresponding to the media content. For example, the main information items in the structured information may be first adjusted according to the security score and the semantic consistency score, and the style information items, the light and shadow information items, and the image quality information items in the structured information may be first adjusted according to the aesthetic score.

[0334] Exemplarily, the target prompt words used to generate media content that meets the media content quality condition can be used as optimization data for training the adjustment model to optimize the adjustment model.

[0335] Exemplarily, the present application can also make security judgments on the original prompt words input by the user, which can reduce the appearance of pornographic, violent, and racially discriminatory content, and expand the use occasions of prompt word adjustment and AIGC products.

[0336] Figure 7 The following is a schematic diagram of the adjustment process of the prompt words shown as an example.

[0337] S701, obtaining the original prompt word input by the user.

[0338] Exemplarily, S701 may refer to the description of S201 above, which will not be repeated here.

[0339] S702: Perform security assessment on the original prompt word input by the user.

[0340] In one possible manner, the first server may call a classifier to perform a security judgment on the original prompt word input by the user. Then, based on the result output by the classifier, determine whether the original prompt word input by the user meets the security condition. For example, the security condition is that the output result of the classifier is 0; therefore, assuming that the output result of the classifier is 1, it can be determined that the original prompt word input by the user does not meet the security condition; assuming that the output result of the classifier is 0, it can be determined that the original prompt word input by the user meets the security condition.

[0341] In one possible way, the first server can call the large language model (combined with ICL, Chain-of-thought (COT) and other technologies) to make a security judgment on the original prompt word input by the user. Then, based on the result output by the large language model, determine whether the original prompt word input by the user meets the security conditions. For example, the security condition is that the prompt word does not contain preset words (preset words can refer to words with security issues). Assuming that the words output by the large language model are preset words, it is determined that the original prompt word input by the user does not meet the security conditions; if the output of the large language model is empty or the output word is not a preset word, it is determined that the original prompt word input by the user meets the security conditions.

[0342] For example, if the original prompt word input by the user is “a cute little girl”, it can be determined that the original prompt word input by the user meets the safety condition.

[0343] For example, if the original prompt word entered by the user is "an African girl who looks like a chimpanzee", it can be determined that the original prompt word entered by the user does not meet the safety condition.

[0344] S703: When it is determined that the original prompt word input by the user does not meet the safety condition, a second adjustment process is performed on the original prompt word so that the prompt word after the second adjustment process meets the safety condition.

[0345] Exemplarily, the second adjustment process may refer to security rewriting, which may include deleting preset words, replacing preset words with words with similar meanings, and the like.

[0346] In a possible manner, the first server may execute S703 to adjust the original prompt word input by the user to a prompt word that meets the safety condition. For example, the first server may call the large language model to perform a second adjustment process on the original prompt word input by the user. For example, after performing a second adjustment process on the original prompt word "an African girl who looks like a chimpanzee" input by the user, the prompt word "an African girl" after the second adjustment process is obtained.

[0347] In one possible manner, the user may adjust the original prompt word input by the user to a prompt word that meets the safety condition.

[0348] Figure 8A to Figure 8C The figure is a schematic diagram of a mobile phone interface shown as an example.

[0349] Reference Fig. 8A For example, the original prompt word entered by the user in the AI ​​image generation interface 801 is "an African girl who looks like a chimpanzee". When the first server determines that the original prompt word entered by the user does not meet the safety condition, it can send a safety prompt message to the mobile phone. After the mobile phone receives the safety prompt message, it can display the safety prompt interface 802, such as Figure 8B As shown. Exemplarily, the safety prompt interface 801 may include one or more controls, including but not limited to: user editing options, system editing options and safety prompt information, etc., which are not limited in this application.

[0350] Reference Figure 8B For example, when the user needs to modify the original prompt word input by himself, he can click the user editing option, and the mobile phone responds to the user's operation behavior and displays the AI ​​image generation interface 801, such as Fig. 8A Afterwards, the user can edit the prompt word in the prompt word input box in the AI ​​image generation interface 801, so that the mobile phone can obtain the prompt word adjusted by the user.

[0351] Continue to refer to Figure 8B For example, when the user needs the AI ​​generated application to modify the original prompt word input by the user, the user can click the system editing option, and the mobile phone can send a third adjustment request to the first server in response to the user's operation behavior. After the first server receives the third adjustment request, it can execute S703.

[0352] For example, after the first server executes S703, it may send the prompt word processed by the second adjustment to the mobile phone. After the mobile phone receives the prompt word processed by the second adjustment, it may display the security prompt interface 803, such as Figure 8C The security modification interface 803 may include one or more controls, including but not limited to: a confirmation option and a prompt word processed by the second adjustment, which is not limited in the present application.

[0353] For example, when the user determines that an AI-generated application is needed, the user can click Figure 8C In the confirmation option, the mobile phone can send a first adjustment request to the first server in response to the user's operation behavior; after the first server receives the first adjustment request, it can execute S704 to S706.

[0354] For example, when the user wants to re-enter the original prompt word, he can click Figure 8CIn the cancel option, the mobile phone can respond to the user's operation behavior and display the AI ​​image generation interface 801, such as Fig. 8A Afterwards, the user can re-enter the original prompt word in the editing box of the AI ​​image generation interface 801.

[0355] In this way, it can be determined according to the user's choice whether the system performs security rewriting or the user customizes the security rewriting of the original prompt word; multiple security rewriting methods are provided.

[0356] S704: Perform recognition processing on the prompt word after the second adjustment processing to obtain structured information, wherein the structured information includes multiple information items.

[0357] For example, when the user Figure 8B When the user edit option is clicked, S704 may be to identify and process the prompt word adjusted by the user to obtain structured information.

[0358] Exemplarily, when the original prompt word input by the user meets the safety condition, in S704, the original prompt word input by the user is recognized and processed to obtain structured information.

[0359] S705, performing a first adjustment process on the information items in the structured information that do not meet the media generation condition;

[0360] S706: Integrate the information items in the structured information that meet the media content generation condition and the information items that have undergone the first adjustment process to obtain a target prompt word, wherein the target prompt word is used to instruct the media generation model to generate media content.

[0361] Exemplarily, S704 to S706 may refer to the description of S202 to S204 above, which will not be repeated here.

[0362] Fig. 9 Schematic diagram of the adjustment process of the prompt word shown as an example

[0363] S901, obtaining the original prompt word input by the user.

[0364] S902: Perform security assessment on the original prompt word input by the user.

[0365] S903: When it is determined that the original prompt word input by the user does not meet the safety condition, a second adjustment process is performed on the original prompt word so that the prompt word after the second adjustment process meets the safety condition.

[0366] Exemplarily, S901 to S903 may refer to the description of S701 to S703 above, which will not be repeated here.

[0367] S904: Perform a first evaluation process on each information item in the structured information to obtain first evaluation information corresponding to each information item in the structured information.

[0368] S905: Determine whether the information items in the structured information meet the media content generation condition according to the first evaluation information corresponding to each information item in the structured information.

[0369] S906: Perform recognition processing on the prompt word after the second adjustment processing to obtain structured information, wherein the structured information includes multiple information items.

[0370] S907, performing a first adjustment process on the information items in the structured information that do not meet the media generation condition;

[0371] S908, integrating the information items satisfying the media content generation condition in the structured information and the information items subjected to the first adjustment process to obtain a target prompt word, wherein the target prompt word is used to instruct the media generation model to generate the media content;

[0372] Exemplarily, S904 to S908 may refer to the description of S202 to S204 above, which will not be repeated here.

[0373] S909: Input the target prompt word into the media generation model to obtain media content.

[0374] S910: Perform a second evaluation process on the media content to obtain second evaluation information corresponding to the media content.

[0375] S911: Determine whether the media content meets a media content quality condition according to second evaluation information corresponding to the media content.

[0376] Exemplarily, when the media content meets the media content quality condition, the process may be terminated. When the media content does not meet the media content quality condition, the process may return to S907.

[0377] Exemplarily, S909 to S911 may refer to the description of S605 to S607 above, which will not be repeated here.

[0378] It should be noted that the steps (S201-S204, S601-S607, S701-S706 or S901-S911) of the above embodiments may also be executed only by the terminal device; or partially by the terminal device and the other partially by the first server, and this application does not impose any restrictions on this.

[0379] Fig.10The schematic diagram of the structure of the prompt word adjustment device is shown as an example. The prompt word adjustment device can be used to execute the method of the above embodiment, so the beneficial effects that can be achieved can refer to the beneficial effects of the corresponding method provided above, which will not be repeated here.

[0380] Reference Fig.10 , exemplarily, the adjustment device of the prompt word may include:

[0381] The acquisition module 1001 is used to acquire the original prompt word input by the user;

[0382] The structured recognition module 1002 is used to recognize and process the original prompt word to obtain structured information, wherein the structured information includes multiple information items, and the value of a single information item is empty or at least part of the original prompt word;

[0383] An adjustment module 1003, configured to perform a first adjustment process on information items in the structured information that do not meet the media generation condition;

[0384] The integration module 1004 is used to integrate the information items in the structured information that meet the media content generation condition and the information items that have undergone the first adjustment process to obtain a target prompt word, wherein the target prompt word is used to indicate the generation of media content.

[0385] Exemplarily, the structured information includes at least the following two information items: a subject information item and a style information item.

[0386] Exemplarily, the prompt word adjustment device further includes:

[0387] The first evaluation module is used to perform a first evaluation process on each information item in the structured information to obtain first evaluation information corresponding to each information item in the structured information; and judge whether each information item in the structured information meets the media content generation condition according to the first evaluation information corresponding to each information item in the structured information.

[0388] Exemplarily, the first evaluation information includes a score, and the media content generation condition includes: a score corresponding to a single information item in the structured information is higher than or equal to a corresponding threshold.

[0389] Exemplarily, the prompt word adjustment device further includes:

[0390] A media content acquisition module is used to acquire media content, where the media content is generated based on the target prompt word;

[0391] The adjustment module 1003 is further configured to, when the media content does not meet the media content quality condition, re-perform a first adjustment process on the information items in the structured information that do not meet the media content generation condition.

[0392] Exemplarily, the prompt word adjustment device further includes:

[0393] The second evaluation module is used to perform a second evaluation process on the media content to obtain second evaluation information corresponding to the media content; and determine whether the media content meets the media content quality condition according to the second evaluation information corresponding to the media content.

[0394] Exemplarily, the prompt word adjustment device further includes:

[0395] A safety judgment module is used to judge the safety of the original prompt word;

[0396] The adjustment module 1003 is further configured to perform a second adjustment process on the original prompt word when the original prompt word does not meet the safety condition, so that the prompt word after the second adjustment process meets the safety condition.

[0397] Exemplarily, the prompt word adjustment device further includes:

[0398] An interactive module, used for displaying a safety prompt interface when the original prompt word does not meet the safety condition; wherein the safety prompt interface includes a manual adjustment option and an automatic adjustment option; and receiving an operation of selecting the automatic adjustment option by the user;

[0399] The adjustment module 1003 is specifically configured to execute a step of performing a second adjustment process on the original prompt word in response to the operation of selecting the automatic adjustment option.

[0400] Exemplarily, the interactive module is also used to display a safety prompt interface when the original prompt word does not meet the safety conditions; wherein the safety prompt interface includes a manual adjustment option and an automatic adjustment option; receive an operation of the user selecting the manual adjustment option; and in response to the operation of selecting the manual adjustment option, obtain the original prompt word adjusted by the user.

[0401] Exemplarily, the interaction module is also used to display structured information.

[0402] Exemplarily, the interaction module is further used to display the first evaluation information corresponding to each information item in the structured information.

[0403] Exemplarily, the interaction module is also used to display the target prompt word.

[0404] Exemplarily, the interaction module is further used to receive an adjustment operation of a user on a target information item in the structured information, and obtain the target information item adjusted by the user.

[0405] Exemplarily, the interaction module is further configured to receive an adjustment operation of the user on the target prompt word, and obtain the target prompt word adjusted by the user.

[0406] Exemplarily, the structured recognition module 1002 is used to input the original prompt word into the language model to obtain structured information.

[0407] Exemplarily, the interaction module is used to display multiple adjustment performance options, where the multiple adjustment performance options correspond to multiple groups of adjustment models; receive a user's selection operation for a target adjustment performance option;

[0408] The adjustment module 1003 is used to call a group of adjustment models corresponding to the target adjustment performance options to perform a first adjustment process on the information items in the structured information that do not meet the media generation conditions.

[0409] Exemplarily, the target prompt word is used to instruct the media generation model to generate media content. The integration module 1004, according to the prompt word structure matching the media generation model, splices the information items in the structured information that meet the media content generation conditions and the information items that have undergone the first adjustment process to obtain the target prompt word.

[0410] In one example, Fig.11 A schematic block diagram of a device 1100 according to an embodiment of the present application is shown. The device 1100 may include: a processor 1101 and a transceiver / transceiver pin 1102 , and optionally, a memory 1103 .

[0411] The various components of the device 1100 are coupled together via a bus 1104, wherein the bus 1104 includes a power bus, a control bus, and a status signal bus in addition to a data bus. However, for the sake of clarity, various buses are referred to as bus 1104 in the figure.

[0412] Optionally, the memory 1103 may be used to store instructions in the aforementioned method embodiment. The processor 1101 may be used to execute the instructions in the memory 1103, and control the receiving pin to receive a signal, and control the sending pin to send a signal.

[0413] The apparatus 1100 may be the electronic device or a chip of the electronic device in the above method embodiment.

[0414] Exemplarily, the electronic device may include a terminal device, a server, etc., which is not limited in this application.

[0415] Among them, all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module, and will not be repeated here.

[0416] The embodiment of the present application also provides a chip, including one or more interface circuits and one or more processors; the one or more processors receive or send data through the one or more interface circuits, and when the one or more processors execute computer instructions, the above-mentioned related method steps implement the steps of the method in the above-mentioned embodiment. The interface circuit is a transceiver / transceiver pin 1102.

[0417] This embodiment further provides a computer-readable storage medium, in which computer instructions are stored. When the computer instructions are executed on an electronic device, the electronic device executes the above-mentioned related method steps to implement the method in the above-mentioned embodiment.

[0418] This embodiment further provides a computer program product, which includes computer instructions. When the computer instructions are executed by a computer or a processor, the computer executes the above-mentioned related steps to implement the method in the above-mentioned embodiment.

[0419] In addition, an embodiment of the present application also provides a device, which may specifically be a chip, component or module, and the device may include a connected processor and memory; wherein the memory is used to store computer-executable instructions, and when the device is running, the processor may execute the computer-executable instructions stored in the memory so that the chip executes the methods in the above-mentioned method embodiments.

[0420] Among them, the electronic device, computer-readable storage medium, computer program product or chip provided in this embodiment are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above and will not be repeated here.

[0421] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0422] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0423] The units described as separate components may or may not be physically separated, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0424] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0425] Any content of each embodiment of the present application, as well as any content of the same embodiment, can be freely combined. Any combination of the above content is within the scope of the present application.

[0426] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0427] The steps of the method or algorithm described in conjunction with the disclosure of the embodiments of the present application can be implemented in a hardware manner, or can be implemented by a processor executing a software instruction. The software instruction can be composed of corresponding software modules, and the software module can be stored in a random access memory (Random Access Memory, RAM), a flash memory, a read-only memory (Read Only Memory, ROM), an erasable programmable read-only memory (Erasable Programmable ROM, EPROM), an electrically erasable programmable read-only memory (Electrically EPROM, EEPROM), a register, a hard disk, a mobile hard disk, a read-only compact disk (CD-ROM) or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0428] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the embodiments of the present application can be implemented with hardware, software, firmware, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer-readable storage media and communication media, wherein the communication media include any media that facilitates the transmission of a computer program from one place to another. The storage medium can be any available medium that a general or special-purpose computer can access.

[0429] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A method for adjusting a prompt word, characterized in that: The method comprises: Get the original prompt word entered by the user; Performing recognition processing on the original prompt word to obtain structured information, wherein the structured information includes multiple information items, and the value of a single information item is empty or at least part of the original prompt word; Performing a first adjustment process on information items in the structured information that do not meet the media generation condition; The information items in the structured information that meet the media content generation condition and the information items that have undergone the first adjustment process are integrated to obtain a target prompt word, wherein the target prompt word is used to indicate the generation of media content.

2. The method according to claim 1, characterized in that The structured information includes at least the following two information items: a main information item and a style information item.

3. The method according to claim 1 or 2, characterized in that: The method further comprises: Performing a first evaluation process on each information item in the structured information to obtain first evaluation information corresponding to each information item in the structured information; According to the first evaluation information corresponding to each information item in the structured information, it is determined whether each information item in the structured information meets the media content generation condition.

4. The method according to claim 3, wherein the first evaluation information includes a score, and the media content generation condition includes: The score corresponding to a single information item in the structured information is higher than or equal to a corresponding threshold.

5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Acquire the media content, where the media content is generated according to the target prompt word; When the media content does not meet the media content quality condition, the first adjustment process is performed again on the information items in the structured information that do not meet the media content generation condition.

6. The method according to claim 5, characterized in that The method further comprises: Performing a second evaluation process on the media content to obtain second evaluation information corresponding to the media content; It is determined whether the media content meets a media content quality condition according to the second evaluation information corresponding to the media content.

7. The method according to any one of claims 1 to 6, characterized in that: The method further comprises: Performing safety judgment on the original prompt word; When the original prompt word does not meet the safety condition, a second adjustment process is performed on the original prompt word, so that the prompt word after the second adjustment process meets the safety condition.

8. The method according to claim 7, characterized in that The method further comprises: When the original prompt word does not meet the safety condition, a safety prompt interface is displayed; wherein the safety prompt interface includes a manual adjustment option and an automatic adjustment option; Receive an operation of selecting the automatic adjustment option by a user; In response to the operation of selecting the automatic adjustment option, the step of performing a second adjustment process on the original prompt word is executed.

9. The method according to claim 7, characterized in that: The method further comprises: When the original prompt word does not meet the safety condition, a safety prompt interface is displayed; wherein the safety prompt interface includes a manual adjustment option and an automatic adjustment option; Receive an operation of selecting the manual adjustment option by a user; In response to the operation of selecting the manual adjustment option, an original prompt word adjusted by the user is obtained.

10. The method according to any one of claims 1 to 9, characterized in that: The method further comprises: The structured information is displayed.

11. The method according to claim 3 or 4, characterized in that: The method further comprises: The first evaluation information corresponding to each information item in the structured information is displayed.

12. The method according to any one of claims 1 to 11, characterized in that: The method further comprises: The target prompt word is displayed.

13. The method according to any one of claims 10 to 12, characterized in that: The method further comprises: A user adjustment operation on a target information item in the structured information is received, and the target information item adjusted by the user is acquired.

14. The method according to any one of claims 1 to 13, characterized in that: The method further comprises: An adjustment operation of the user on the target prompt word is received, and the target prompt word adjusted by the user is acquired.

15. The method according to any one of claims 1 to 14, characterized in that The identifying process of the original prompt word to obtain structured information includes: The original prompt word is input into a language model to obtain the structured information.

16. The method according to any one of claims 1 to 15, characterized in that The method further comprises: Displaying a plurality of adjustment performance options, wherein the plurality of adjustment performance options correspond to a plurality of groups of adjustment models; receiving a user's selection operation for adjusting a performance option for a target; The performing a first adjustment process on the information items in the structured information that do not meet the media generation condition includes: A group of adjustment models corresponding to the target adjustment performance options are called to perform a first adjustment process on the information items in the structured information that do not meet the media generation condition.

17. The method according to any one of claims 1 to 16, characterized in that: The target prompt word is used to instruct the media generation model to generate media content, and the information items satisfying the media content generation condition in the structured information and the information items subjected to the first adjustment process are integrated to obtain the target prompt word, including: According to the prompt word structure matching the media generation model, the information items in the structured information that meet the media content generation condition and the information items that have undergone the first adjustment process are spliced ​​to obtain the target prompt word.

18. A device for adjusting prompt words, characterized in that: The device comprises: An acquisition module is used to acquire the original prompt word input by the user; a structured recognition module, configured to recognize and process the original prompt word to obtain structured information, wherein the structured information includes a plurality of information items, and the value of a single information item is empty or at least part of the original prompt word; An adjustment module, configured to perform a first adjustment process on information items in the structured information that do not meet the media generation condition; The integration module is used to integrate the information items in the structured information that meet the media content generation condition and the information items that have undergone the first adjustment process to obtain a target prompt word, wherein the target prompt word is used to indicate the generation of media content.

19. An electronic device, characterized in that: include: a memory and a processor, the memory being coupled to the processor; The memory stores program instructions, and when the program instructions are executed by the processor, the electronic device executes the method according to any one of claims 1 to 17.

20. A chip, characterized in that: It comprises one or more interface circuits and one or more processors; the one or more processors receive or send data through the one or more interface circuits, and when the one or more processors execute computer instructions, the steps of the method described in any one of claims 1 to claim 17 are executed.

21. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program runs on a computer or a processor, the computer or the processor executes the method according to any one of claims 1 to 17.

22. A computer program product, characterized in that The computer program product comprises computer instructions, which, when executed by a computer or a processor, cause the steps of the method according to any one of claims 1 to 17 to be performed.

Images