Content generation method, object recommendation method, device, storage medium and program product

Abstract

The embodiment of the invention provides a content generation method, an object recommendation method, equipment, a storage medium and a program product. Comprises: determining a content generation requirement; at least one piece of target attribute description information matched with the content generation requirement and a target effect image corresponding to the at least one piece of target attribute description information are determined, the target effect image comprises at least one target object and a display carrier, and the target effect image is obtained based on fusion of the display image and an object image of the target object; the display image comprises a display carrier and at least one display object conforming to the target attribute description information, and the at least one target object and the at least one display object meet similar requirements; generating a target copywriting based on the content generation requirement and object information corresponding to a target object in the at least one target effect image; and generating target content based on the at least one target effect image and the target copywriting. According to the technical scheme provided by the embodiment of the invention, the content generation efficiency is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of Internet technology, and in particular to a content generation method, an object recommendation method, a device, a storage medium, and a program product. Background Art

[0002] Content generally refers to online information such as text, pictures, audio, video, etc. published on the Internet in any form. It is an emerging model for the creation and organization of online information resources, and is widely used in various fields such as object promotion and information output.

[0003] For example, when content is used for promotion or sharing, the inventors have discovered that content generated from images and text often achieves better recommendations. For example, in e-commerce, when it comes to recommending clothing products, combining images of clothing items displayed by models with content generated from recommended copy can intuitively and accurately present the characteristics of clothing products, making them more appealing to users.

[0004] Currently, content is usually obtained by inviting real models to try on clothing products, taking photos to obtain the effect images, and manually generating corresponding recommendation copy. The cost of content generation is high and the efficiency is low. Summary of the Invention

[0005] The embodiments of the present application provide a content generation method, an object recommendation method, a device, a storage medium, and a program product to improve content generation efficiency and reduce generation costs.

[0006] In a first aspect, the present application provides a content generation method, comprising:

[0007] Determine content generation requirements;

[0008] Determining at least one target attribute description information matching the content generation requirement, and a target effect image corresponding to each of the at least one target attribute description information; the target effect image includes at least one target object and a display carrier, the target effect image is obtained based on a fusion of a display image and an object image of the at least one target object, the display image includes the display carrier and at least one display object matching the target attribute description information, and the at least one target object and the at least one display object meet a similarity requirement;

[0009] generating a target copy based on the content generation requirement and object information corresponding to a target object in at least one target effect image;

[0010] Target content is generated based on the at least one target effect image and the target text; wherein the target content is used to recommend a target object in the at least one target effect image.

[0011] In a second aspect, the present application provides an object recommendation method, comprising:

[0012] In response to the object recommendation request, determining at least one recommended object;

[0013] In a case where the at least one recommended object includes one or more target objects associated with the target content, content prompt information of the target content is used as recommendation prompt information for the one or more target objects; the target content is generated based on at least one target effect image and a target copy, target attribute description information corresponding to each of the at least one target effect image matches the content generation requirements, the target effect image includes at least one target object and a display carrier, the target effect image is obtained based on a fusion of a display image and an object image of the at least one target object, the display image includes the display carrier and at least one display object that meets the target attribute description information, and the at least one target object and the at least one display object meet a similarity requirement; the target copy is generated based on the content generation requirements and object information corresponding to the target object in the at least one target effect image;

[0014] Output the recommendation prompt information corresponding to each of the at least one recommendation object.

[0015] In a third aspect, the present application provides a computing device, including a processing component and a storage component;

[0016] The storage component stores a computer program; the computer program is used to be called and executed by the processing component to implement the content generation method as described in the first aspect, or the object recommendation method as described in the second aspect.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processing component, it implements the content generation method described in the first aspect or the object recommendation method described in the second aspect.

[0018] In a fifth aspect, the present application provides a computer program product, comprising a computer program or instructions, which, when executed by a processing component, implements the content generation method as described in the first aspect, or the object recommendation method as described in the second aspect.

[0019] In an embodiment of the present application, content generation requirements can be determined, and at least one target attribute description information that matches the content generation requirements and a target effect image corresponding to each of the at least one target attribute description information can be determined. The target effect image can include at least one target object and a display carrier. Based on the content generation requirements and the object information corresponding to the target object in the at least one target effect image, a target copy can be generated, and based on the at least one target effect image and the target copy, a target copy can be generated. The target effect image can be obtained by fusing a display image and an object image of at least one target object. The display image can include a display carrier and at least one display object that meets the target attribute description information. The at least one target object and the at least one display object meet similarity requirements. By determining at least one target effect image that matches the content generation requirements, a target copy can be automatically generated based on the content generation requirements and the object information corresponding to the target object in the at least one target effect image. Then, based on the target copy and the at least one target effect image, a target content can be automatically generated, thereby achieving automatic generation of target content that matches the content generation requirements, improving the generation efficiency of the target content, and reducing the generation cost of the target content.

[0020] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0022] Figure 1 A flowchart of an embodiment of a content generation method provided by the present application is shown;

[0023] Figure 2 A flow chart showing an embodiment of a method for generating a target effect image provided by the present application is shown;

[0024] Figure 3 A schematic diagram showing target content in a practical application is shown;

[0025] Figure 4 A flowchart of an embodiment of an object recommendation method provided by the present application is shown;

[0026] Figure 5 A schematic diagram of the system architecture in a practical application is shown;

[0027] Figure 6 A schematic structural diagram of an embodiment of a content generation device provided by the present application is shown;

[0028] Figure 7 A schematic structural diagram of an embodiment of an object recommendation device provided by the present application is shown;

[0029] Figure 8 A schematic diagram of the structure of a computing device in a practical application is shown. DETAILED DESCRIPTION

[0030] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0031] It should be noted that, in the case of user information involved in the embodiments of the present application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse. In addition, the various models involved in this application (including but not limited to language models or large models) are in compliance with relevant laws and standards.

[0032] In addition, it should be noted that when the embodiments of the present application involve user interaction operations or triggering operations, the user interaction operations or triggering operations involved in the embodiments of the present application include but are not limited to: touch operations, gesture operations, voice operations, head movement operations, eye movement operations and other interactive operations in various ways; among which, touch operations include but are not limited to: click operations, double-click operations, long press operations, sliding operations, pinch operations or mouse hover operations, etc. Sliding operations include but are not limited to: straight sliding, curved sliding, etc.

[0033] As described in the background technology section, the current content generation method, taking the clothing product recommendation scenario in e-commerce scenarios as an example, usually involves inviting real-life models to try on clothing products, taking photos to obtain the effect images, and manually generating corresponding recommendation copy to obtain content. The content generation cost is high and the efficiency is low.

[0034] Therefore, in order to improve the content generation efficiency and reduce the generation cost, the inventor proposed the technical solution of the present application, including: determining the content generation requirements; determining at least one target attribute description information that matches the content generation requirements, and the target effect image corresponding to each of the at least one target attribute description information; the target effect image includes at least one target object and a display carrier, the target effect image is obtained based on the fusion of the display image and the object image of the target object, the display image includes the display carrier and at least one display object that meets the target attribute description information, and the at least one target object meets the similarity requirements with the at least one display object; based on the content generation requirements and the object information corresponding to the target object in the at least one target effect image, generating the target copy; based on the at least one target effect image and the target copy, generating the target content; wherein, the target content is used to recommend the at least one target object.

[0035] By determining at least one target effect image that matches the content generation requirements, the target copy can be automatically generated based on the content generation requirements and the object information corresponding to the target object in the at least one target effect image. Then, based on the target copy and the at least one target effect image, the target content can be automatically generated, thereby realizing the automatic generation of target content that matches the content generation requirements, improving the generation efficiency of the target content, and reducing the generation cost of the target content.

[0036] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0037] Figure 1 This is a flowchart of an embodiment of a content generation method provided in this application. The technical solution of this embodiment can be applied to the server.

[0038] In a practical application, a system architecture applicable to the technical solution of the embodiment of the present application may include a server and a client. The client may be for operators to view the generated target content and perform subsequent operations such as object recommendation based on the target content.

[0039] The client and the server may be connected via a network, which may include various connection types, such as wired or wireless communication links or optical fiber cables, etc. The client may interact with the server via the network to send first content or receive recommendation information, etc.

[0040] The client can be a browser, an APP (Application), a web application such as an H5 (HyperText Markup Language 5) application, a light application (also known as a mini-program, a lightweight application), or a cloud application. The client can be deployed in an electronic device and needs to rely on the device or certain apps in the device to run. For example, the electronic device can have a display and support information browsing, such as a personal mobile terminal such as a mobile phone, tablet computer, personal computer, desktop computer, smart speaker, smart watch, etc.

[0041] The server side may include servers that provide various services, such as servers that provide instant messaging, servers that provide model training, etc.

[0042] It should be noted that the server can be implemented as a distributed server cluster consisting of multiple servers or a single server. The server can also be a server in a distributed system or a server integrated with blockchain. The server can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0043] It should be noted that the technical solutions of the embodiments of this application are applicable to virtual network environments. The users described are generally referred to as "virtual users." Real users can register user accounts on the server to obtain user identities in the network environment. Interactions between the server and the user can be implemented based on the user account. The corresponding data received or sent by the server to the user is also implemented based on the user account. In fact, it is the client corresponding to the user account that receives or sends the corresponding data to the generating server.

[0044] Figure 1 The content generation method may include the following steps:

[0045] 101: Determine content generation requirements.

[0046] Content generation requirements can include keywords, for example. For example, in an e-commerce clothing recommendation scenario, keywords can include key information such as style, scene, season, and / or color. In a practical application, keywords may include "spring outfits" or "travel outfits."

[0047] The content generation requirement may be sent by the client. The client may generate a content generation request in response to the content generation requirement input by the user in the user interface, and send the request to the server, so that the server may determine the content generation requirement.

[0048] 102: Determine at least one target attribute description information that matches the content generation requirement, and a target effect image corresponding to each of the at least one target attribute description information.

[0049] Each target attribute description information may correspond to a target effect image, thereby determining at least one target effect image.

[0050] In an embodiment of the present application, each target effect image may include at least one target object and a display carrier. The display carrier may have a target image, and the target effect image may present a visual effect of the at least one target object being displayed by the display carrier. The display carrier may include a human model, an animal model, or any carrier that can display an object. In an e-commerce scenario, the target object may refer to a target commodity, such as clothing, jewelry, or a backpack. In a practical application, the display carrier may include a human model. When the target commodity is a dress, the target effect image may present a human model wearing a dress to show the visual effect of the dress. Taking the target object as a clothing type as an example, different target attribute description information may include different values ​​of style, material, and / or color, etc., which may be determined according to the content generation requirements.

[0051] Attribute description information may include object attributes and / or environmental attributes. For example, in e-commerce contexts, for clothing products, the object attributes in the attribute description information may include, but are not limited to, descriptions of the object's style, material, color, pattern, and / or functional characteristics. Environmental attributes may include, for example, scene style. Scene style describes scene information, such as indoor, outdoor, urban, or grassland, while object style describes object characteristics, such as sportswear or commuter wear.

[0052] The at least one target attribute description information matching the content generation requirement can be obtained, for example, by searching multiple attribute description information through text matching. Of course, the content generation requirement can also be identified to obtain the at least one target attribute description information included in the content generation requirement.

[0053] The target effect image corresponding to any target attribute description information can be obtained by fusing a display image and an object image of at least one target object. The display image can include the display carrier and at least one display object that matches the target attribute description information. The display image can be specifically generated based on the carrier image of the display carrier and the target attribute description information.

[0054] First, a target object that meets similar requirements to each display object can be determined to obtain at least one target object corresponding to at least one display object, and then a target effect image can be obtained by fusing the display image and the object image of at least one target object.

[0055] Alternatively, in actual applications, the target object may refer to a real object, such as a commodity provided by an e-commerce system, while the display object is virtual. In an e-commerce scenario, a target object that meets similar requirements may be understood as the same commodity as the display object.

[0056] When a display image includes a display object, a target object that meets similar requirements to the display object can be determined. The target effect image can be obtained by fusing the display image with the target object's image. For example, if the display object is a dress, the target object can be a dress product with similar color, material, and style.

[0057] When a display image includes multiple display objects, target objects that meet similar requirements as each display object can be determined, thereby determining multiple target objects. A target effect image can be obtained by fusing the display image with the object images of the multiple target objects. For example, if a display image includes two display objects, a crew neck top and a below-the-knee skirt, the target objects can include tops with similar color, material, and style to the crew neck top, and skirts with similar color, material, and style to the below-the-knee skirt.

[0058] It should be noted that the carrier image, display image and display carrier and display object in the target effect image involved in the embodiments of the present application may all be virtual.

[0059] The target effect image may be selected from a plurality of pre-generated effect images, or may be generated in real time according to content generation requirements. The specific generation process will be described in detail in subsequent embodiments.

[0060] 103: Generate a target copy based on the content generation requirement and object information corresponding to a target object in at least one target effect image.

[0061] For example, if the target object is a target product, the object information corresponding to the target object may include product description information of the target product, such as price, material, size, etc. The product description information may be obtained from a product database. The target copy may refer to text information used to describe and recommend the at least one target object.

[0062] The target text may be generated based on the content generation requirements and the object information corresponding to the target object included in each target effect image. The object information may be obtained from an online system that provides the target object.

[0063] Specifically, a text generation model can be used to generate target text based on content generation requirements and object information corresponding to a target object in at least one target effect image. Wherein, the text generation model can be a language model (Language Mode, LM) or a multimodal model (Multimodal Model, MM) based on artificial intelligence, etc. The embodiment of the present application does not limit the number of model parameters supported by the model, with the goal of meeting actual needs. The text generation model can be a deep learning model for processing and generating natural language text or multimodal data, which can be implemented based on a neural network architecture, and can be pre-trained on a large amount of data. In an optional implementation, the text generation model can include an encoder (Encoder), a decoder (Decoder), a self-attention layer (Self-Attention Layer) and a feed-forward neural network (Feed-Forward Neural Network), etc. The encoder is mainly used to convert input data (usually in sequence form) into vector representation. This process can capture the semantic features of the input data. The decoder is responsible for converting the intermediate representation generated by the encoder into output data (usually in sequence form). The self-attention layer is a mechanism that allows the model to pay attention to other positions in the sequence to better encode the current position information. The feedforward neural network can perform nonlinear transformations on the output of the self-attention layer to enhance the model's expressiveness. The various parts work together to enable the models built on them to perform well in various complex processing tasks, such as natural language processing, computer vision, speech recognition, machine translation, text summarization, and intelligent question answering.

[0064] Generating target text using a text generation model can be done based on prompts. Prompts are a form of input used to prompt or guide the model to produce expected outputs, instructing the model on what actions to take or what output to generate when performing a specific task. Prompts are natural language inputs, similar to commands or instructions, that inform the model of what to do.

[0065] Specifically, a first prompt message can be generated based on the content generation requirements and object information corresponding to a target object in at least one target effect image, and the first prompt message can be input into a text generation model to instruct the text generation model to generate a target copy. For example, the first prompt message can read: "You are a copy generation expert who excels at generating copy recommending a target product based on the copy generation requirements and the product description of the target product. Please generate a copy recommending the target product based on the given generation requirements and the product description of the target product provided."

[0066] It should be noted that the first prompt information described here, as well as other prompt information that may be involved below, are merely examples for ease of understanding. This application is not limited thereto and specific settings can be made based on actual conditions.

[0067] 104: Generate target content based on at least one target effect image and target text.

[0068] The target content may be used to recommend a target object in the at least one target effect image. Specifically, the target content may be used to recommend all target objects in each target effect image.

[0069] To improve the convenience and standardization of target content generation, a target content template can optionally be determined, and the target content can be generated based on the target content template, at least one target effect image, and a target copy. The target content template can include a first preset area corresponding to the target effect image and a second preset area corresponding to the target copy. The position, size, style, etc. of the first and second preset areas can be set according to actual needs. Specifically, at least one target effect image can be added to the first preset area, and the target copy can be added to the second preset area to obtain the target content.

[0070] Optionally, when at least one target effect image includes multiple target effect images, the multiple target effect images can be spliced ​​together to obtain a spliced ​​effect image, and the spliced ​​effect image can be added to a third preset area in the target content template, such as the home page or cover area, and the multiple target effect images can be added to the first preset area respectively, and the target copy can be added to the second preset area to obtain the target content.

[0071] Figure 2 A schematic diagram of target content in a practical application is shown. Figure 2 As shown, the display image 201 includes a display carrier and a display object. Figure 2In the example, the display carrier is a human model, the display object is a dress, the target object 202 is a target dress that meets similar requirements to the display object in the display image 201, the target effect image 203 is generated based on the fusion of the display image 201 and the object image of the target object 202, and presents a visual effect of the human model displaying the target dress. The target copy 204 can be generated based on the content generation requirements and the object information corresponding to the target object, for example Figure 2 It can include product information, dressing style and scene style related to the content generation requirements, and can also include information such as recommendation reasons. Target content can be generated based on the target effect image 203 and the target copy 204.

[0072] In this embodiment, content generation requirements can be determined, and at least one target attribute description information matching the content generation requirements and a target effect image corresponding to each of the at least one target attribute description information can be determined. The target effect image can include at least one target object and a display carrier. Based on the content generation requirements and the object information corresponding to the target object in the at least one target effect image, a target copy can be generated. Furthermore, based on the at least one target effect image and the target copy, a target copy can be generated. The target effect image can be obtained by fusing a display image and an object image of the at least one target object. The display image can include a display carrier and at least one display object that meets the target attribute description information. The at least one target object and the at least one display object meet similarity requirements. By determining at least one target effect image matching the content generation requirements, a target copy can be automatically generated based on the content generation requirements and the object information corresponding to the target object in the target effect image. Furthermore, based on the target copy and the at least one target effect image, target content can be automatically generated. This achieves the automatic generation of target content matching the content generation requirements, improves the efficiency of generating the target content, and reduces the cost of generating the target content.

[0073] The generation process of the target effect image is described below. Figure 3 The flowchart of an embodiment of a method for generating a target effect image provided by the present application can be applied to a server. The method may include the following steps:

[0074] 301: Determine at least one target attribute description information that matches the content generation requirement.

[0075] 302: For any target attribute description information, generate a display image using a first image generation model based on the carrier image of the display carrier and the target attribute description information.

[0076] The generated display image may include a display carrier and at least one display object that is integrated with the display carrier and conforms to the target attribute description information.

[0077] There are many ways to implement the determination of target attribute description information and the generation of display images, which will be described in subsequent embodiments and will not be repeated here.

[0078] 303: Search for at least one target object that meets similar requirements to at least one display object in the display image.

[0079] In a practical application, for any displayed object in a displayed image, a target object whose image similarity with the displayed object satisfies a similarity requirement can be searched from an object database. A preset image feature extraction algorithm can be used to extract the image features of the displayed object image and the image features of each object image in the object database. Feature similarities between the image features of the displayed object image and the image features of each object image are calculated, and a target object that satisfies the feature similarity requirement, such as having the greatest feature similarity or having a feature similarity exceeding a similarity threshold, is selected.

[0080] 304: Obtain object images corresponding to at least one target object.

[0081] Optionally, for any target object, it may be checked whether there is a tiled image of the target object;

[0082] If so, the tiled image can be used as the object image;

[0083] If not, a wearing image of the target object may be obtained, and the wearing image may be cropped to obtain an object image of the target object.

[0084] In e-commerce scenarios, for example, a wearable image can refer to a model display image provided by a merchant. For example, if the target object is a clothing item, the object image can be an image of a model trying it on provided by the merchant. In this case, the wearable image can be cropped to obtain an object image that only contains the target object, without other objects such as the model.

[0085] Optionally, for example, when the model's trying-on image includes upper body clothing and lower body clothing, the wearing image is cropped to obtain multiple sub-images, such as the first part of the sub-image corresponding to the upper body clothing and the second part of the sub-image corresponding to the lower body clothing. Based on the object type of the target object, for example, if the target object is upper body clothing, the first part of the sub-image can be used as the object image.

[0086] This can achieve automatic search of the target object and automatic acquisition of the object image of the target object without manual operation, thereby improving the search and acquisition efficiency, and further improving the efficiency of generating the target effect image.

[0087] 305: Fusing the object image of at least one target object into a display carrier in the display image to obtain a target effect image corresponding to the target attribute description information.

[0088] Among them, an image fusion model can be used to fuse the object image of at least one target object into a display carrier of the display image to obtain a target effect image. The image fusion model can be a deep learning model for processing and generating image data, and can be implemented based on a neural network architecture. In a practical application, the image fusion model can be constructed based on a generative adversarial network (GAN), and its model parameters can be pre-trained using training data.

[0089] In this embodiment, by determining at least one target attribute description information that matches the content generation requirements, and combining the carrier image of the display carrier and the target attribute description information to generate a display image, and combining the display image and the object image of at least one target object to generate a target effect image, online and automatic generation of the target effect image is achieved, thereby improving generation efficiency.

[0090] In the embodiment of the present application, there are multiple ways to implement determining at least one target attribute description information that the content generation requirement matches.

[0091] As an optional implementation manner, at least one target attribute description information that matches the content generation requirement may be searched from multiple attribute description information.

[0092] As another optional implementation, target role information and at least one attribute description image set corresponding to the target role information may be determined; and at least one target attribute description information matching the content generation requirement may be extracted from the at least one attribute description image set.

[0093] The target character information may include character settings, animal settings, etc. Optionally, determining the target character information may include: determining the target character information from a plurality of character information.

[0094] For example, if the target character information is character setting information, which can also be called a character portrait, multiple character information can be obtained as follows:

[0095] Obtaining user attribute information corresponding to multiple target users and multiple user-generated contents;

[0096] Performing data analysis on any user attribute information and any user-generated content to determine at least one attribute feature corresponding to the user attribute information and at least one content feature corresponding to the user-generated content;

[0097] A role information is composed of one or more attribute features and one or more content features.

[0098] Among them, the target users may include users who publish content on the target content platform, users whose user interaction data such as the number of users they follow, the number of users they follow, the number of messages they leave, the number of discussions they have had, etc., which meet the interaction requirements. User attribute information may include but is not limited to user gender, age, height, region, season and other information, and user-generated content may include but is not limited to the type of content posted by users, the form of content, etc., which can be determined according to the actual scenario. Taking clothing products in e-commerce scenarios as an example, in an actual application, target users may include, for example, fashion experts, that is, users who love clothing matching and sharing. User-generated content may include, for example, dressing styles, shooting scenes, etc. Dressing styles may include, for example, dresses, skirts, sports suits, etc., and shooting scenes may include, for example, indoor scenes, outdoor scenes, etc.

[0099] Data analysis can be performed on any user attribute information to determine at least one attribute feature corresponding to the user attribute information, and data analysis can be performed on any user-generated content to determine at least one content feature corresponding to the user-generated content. Character information can be constructed by selecting one or more attribute features from multiple attribute features, and one or more content features from multiple content features. In a practical application, the character information of "25-year-old female wearing a dress in an indoor setting" can be constructed based on the two attribute features of "25 years old" and "female", and the two content features of "dress outfit" and "indoor scene".

[0100] By constructing character information based on one or more attribute features and one or more content features, automatic construction of character setting information is achieved. Moreover, since the attribute features and content features are extracted from the user attribute information and user-generated content of the target user, the authenticity of the attribute features and content features is improved, thereby improving the authenticity of the constructed character information. In addition, since one or more attribute features and one or more content features can be selected from the extracted multiple attribute features and content features to construct the character information, the diversity and comprehensiveness of the character information are improved.

[0101] After the target role information is determined, at least one attribute description image set corresponding to the target role information may be determined, and at least one target attribute description information matching the content generation requirement may be extracted from the image set.

[0102] In a practical application, the attribute description information may include object style description information, such as the dressing style description information and scene style description information in the clothing type. Therefore, the dressing style image set and scene style image set corresponding to the target character information can be determined. Among them, the dressing style image set and scene style image set corresponding to the target character information can be pre-set. For example, when the target character information is a 25-year-old woman wearing a dress in an indoor scene, the corresponding dressing style image set can be a dress dressing image set, which can include multiple dress dressing images, and the corresponding scene style image set can be an indoor scene image set, which can include multiple indoor scene images.

[0103] Based on the dressing style image set and the scene style image set, an information extraction model can be used to extract dressing style description information and scene style description information that match the content generation requirements.

[0104] The information extraction model can be an artificial intelligence-based language model (LM) or a multimodal model (MM). The embodiments of the present application do not limit the number of model parameters supported by the model, with the goal of meeting actual needs. The information extraction model can be a deep learning model used to process and generate natural language text or multimodal data, which can be implemented based on a neural network architecture and can be pre-trained on a large amount of data.

[0105] Specifically, a second prompt can be generated based on the outfit style image set and the scene style description set, and the second prompt is input into the information extraction model to instruct the information extraction model to extract outfit style description information and scene style description information that match the content generation requirements. Taking the content generation requirement of spring outfit recommendations as an example, the second prompt can be: You are an information extraction expert who is good at extracting description information that meets the requirements from images. Based on the given outfit style image set and scene style image set, please extract outfit description information and scene style description information that meet the spring outfit recommendations.

[0106] Optionally, an information extraction model can be used to extract at least one attribute description information from at least one attribute description image set, and then the extracted at least one attribute description information can be screened, reorganized, and other operations can be performed to determine at least one target attribute description information that matches the content generation requirements.

[0107] It is understandable that the images in the above-mentioned image set usually contain display carriers such as human models. At this time, the extracted attribute description information may include carrier information, but the display carrier is different from the display carrier involved in the subsequent generation of display images. Therefore, in order to avoid interfering with the display carrier involved in the subsequent generation of display images, the carrier information of the extracted display carrier can be removed here, and the remaining clothing and scene attribute information can be retained.

[0108] By determining the target role information, at least one target attribute description information that matches the content generation requirements is extracted from at least one attribute description image set corresponding to the target role information, thereby achieving automatic extraction and determination of the target attribute description information, and the target attribute description information can be matched with the target role information and the content generation requirements, which can ensure that the subsequent display image that matches the content generation requirements can be generated based on the target attribute description information and the carrier image of the display carrier.

[0109] After the target attribute description information is determined, the first image generation model can be used to generate a display image. In the embodiment of the present application, the first image generation model can be implemented in a variety of ways.

[0110] As an optional implementation, different attribute description information may correspond to different first image generation models. For any first image generation model, the first training data corresponding to the attribute description information may be trained and obtained. The first training data may include sample attribute description information and sample display images corresponding to the sample attribute description information.

[0111] In this implementation, for any target attribute description information, the corresponding first image generation model can be used to generate a display image based on the carrier image of the display carrier and the target attribute description information.

[0112] As another optional implementation, the first image generation model may include a first basic model, and a first adaptation model added to the first basic model and corresponding to at least one target attribute description information.

[0113] Among them, the first basic model can be a language model (Language Mode, LM) or a multimodal model (Multimodal Model, MM) based on artificial intelligence, etc. The embodiment of the present application does not limit the number of model parameters supported by the model, with the goal of meeting actual needs. The first basic model can be a deep learning model for processing natural language text, generating images or multimodal data, which can be implemented based on a neural network architecture and can be pre-trained on a large amount of data. The pre-training process is not described in detail.

[0114] The first adaptation model may refer to a model parameter adjustment model, and the first incremental parameter of the first adaptation model may be used to adjust the first model parameter of the first basic model. In a practical application, the first adaptation model may be implemented as a low-rank adaptation matrix (LoRA for short). LoRA is a parameter adjustment method for deep learning models. It freezes the pre-trained model weights and injects a trainable low-rank matrix into the model to achieve fine-tuning of the model parameters. For example, the matrix dimension of the original weight matrix W0 may be n×m, and the low-dimensional matrices B (dimension may be n×r) and A (dimension may be r×m) are trained. ΔW is constructed by matrix multiplication. The parameter amount of ΔW is lower than that of the original matrix. W0+ΔW is used as the adjusted model parameter to achieve efficient and lightweight model training.

[0115] The first image generation model can be trained as follows:

[0116] Adding at least one first adaptation model corresponding to each attribute description information to the first basic model to form a first image generation model;

[0117] For any attribute description information, by freezing the first model parameters of the first basic model, the corresponding first adaptation model is trained using the first training data corresponding to the attribute description information to obtain the first incremental parameters corresponding to the first adaptation model.

[0118] It should be noted that different first adaptation models are independent of each other, and each first adaptation model has different first incremental parameters and corresponding attribute description information. For any first adaptation model, its first incremental parameters can be trained using its corresponding attribute description information and the corresponding first training data. The first training data can include sample attribute description information and sample display images corresponding to the sample attribute description information.

[0119] Freezing the first model parameters of the first base model may mean maintaining the first model parameters in a fixed state and not adjusting them during the training process. Specifically, the sample attribute description information in the first training data is input into the first image generation model, and after obtaining the output result, the output result is compared with the sample display image in the first training data. Based on the comparison result, the first incremental parameters of the first adaptation model are adjusted until training is complete.

[0120] In this implementation, generating a display image using the first image generation model based on the carrier image of the display carrier and the target attribute description information may include:

[0121] A display image is generated based on a carrier image of a display carrier and the target attribute description information by using a first basic model in the first image generation model and a first adaptation model corresponding to the target attribute description information.

[0122] By training, a first adaptation model corresponding to at least one attribute description information is obtained, and the first image generation model is constructed in combination with the first basic model. It is possible to use one first image generation model to complete the processing of different attribute description information to generate corresponding display images, thereby improving the applicability of the first image generation model. In addition, each first adaptation model is independent of each other, which can avoid interference with the data processing of other attribute description information and improve the accuracy of data processing. In addition, during the model training process, by freezing the first model parameters of the first basic model and only training to obtain the first incremental parameters of the first adaptation model, the complexity of model training is reduced, and the model training efficiency and data processing efficiency are improved.

[0123] To further improve the accuracy of the display image, the carrier image can be generated based on the target character information. Therefore, in some embodiments, before generating the display image, the above method may further include:

[0124] Determine target role information;

[0125] A second image generation model is used to generate a carrier image having a target image corresponding to the target character information.

[0126] There are many ways to implement the second image generation model to generate a carrier image having a target image corresponding to the target character information.

[0127] As an optional implementation method, a plurality of sample carrier images having a target image corresponding to the target character information may be determined; and the plurality of sample carrier images may be fitted using a second image generation model to generate a carrier image.

[0128] In this implementation, a sample carrier image library can be pre-set, which can include sample carrier images corresponding to different target images. Taking the display carrier as a human model as an example, the sample carrier images corresponding to different target images can include sample model images corresponding to 25-year-old women, sample model images corresponding to 30-year-old men, sample model images corresponding to 5-year-old round-faced female children, etc.

[0129] Therefore, a plurality of sample carrier images having a target image corresponding to the target character information can be searched from the sample carrier image library, and the searched plurality of sample carrier images can be fitted to obtain a carrier image.

[0130] As another optional implementation, the second image generation model can be trained as follows:

[0131] adding at least one second adaptation model to the second basic model to form a second image generation model;

[0132] By freezing the second model parameters corresponding to the second base model, at least one second adaptation model is trained using the second training data to obtain second incremental parameters corresponding to each of the at least one second adaptation model. The second training data may include sample role information and a sample carrier image corresponding to the sample role information.

[0133] In this implementation, the second basic model can be a language model (LanguageMode, LM) or a multimodal model (Multimodal Model, MM) based on artificial intelligence, etc. The embodiment of the present application does not limit the number of model parameters supported by the model, with the goal of meeting actual needs. The second basic model can be a deep learning model for processing natural language text, generating images or multimodal data, which can be implemented based on a neural network architecture and can be pre-trained on a large amount of data. The pre-training process is not described in detail.

[0134] The second adaptation model may refer to a model parameter adjustment model, and the second incremental parameter of the second adaptation model may be used to adjust the second model parameter of the second base model. In a practical application, the second adaptation model may be implemented as a low-rank adaptation matrix (LoRA). LoRA is a parameter adjustment method for deep learning models. It freezes the pre-trained model weights and injects a trainable low-rank matrix into the model to achieve fine-tuning of the model parameters. The training process may be consistent with the first image generation model and will not be described in detail here.

[0135] It should be noted that the second basic model can be the same model as the first basic model, or can be a different model. When multiple second adaptation models are added to the second basic model, the multiple second adaptation models are independent of each other.

[0136] At least one second adaptation model is obtained through training, and a second image generation model is constructed in combination with the second basic model. It is possible to use one second image generation model to complete the processing of different character information to generate corresponding carrier images, thereby improving the applicability of the second image generation model. In addition, each second adaptation model is independent of each other, which can avoid interference with the data processing of other character information and improve the accuracy of data processing. In addition, during the model training process, by freezing the second model parameters of the second basic model and only training the second incremental parameters of the second adaptation model, the complexity of model training is reduced, and the model training efficiency and data processing efficiency are improved.

[0137] In some embodiments, using the second image generation model to generate a carrier image having a target image corresponding to the target character information may include:

[0138] Determining a weight coefficient corresponding to each of the at least one second incremental parameters;

[0139] According to the weight coefficient, a second image generation model is used to generate a carrier image having a target image corresponding to the target character information.

[0140] Taking the LoRa model as an example, the strength-model (weight coefficient) corresponding to at least one second incremental parameter can be determined. The weight coefficient can range from (0, 1). Based on the second basic model, the carrier image can be generated using the corresponding second adaptation model according to the weight coefficient.

[0141] In order to improve the authenticity of the carrier image, the above method may further include:

[0142] determining whether the carrier image meets the first image requirement;

[0143] If not, adjusting the weight coefficient corresponding to at least one of the second incremental parameters, and regenerating a carrier image having a target image corresponding to the target character information using the second image generation model according to the adjusted weight coefficient, and continuing to perform the operation of determining whether the carrier image meets the first image requirement;

[0144] If so, an operation of generating a display image based on the carrier image and the target attribute description information by using the first image generation model is executed.

[0145] Among them, the first image requirement can be set according to actual needs, for example, it can include the target carrier image of the generated carrier image and a carrier image that is closer to the basic carrier image, or a carrier image corresponding to a certain role information.

[0146] Taking a model image as an example, it can be understood that when the second image generation model includes a second adaptation model, increasing the weight coefficient of the second incremental parameter corresponding to the second adaptation model can make the model image of the generated model image closer to the model image of the character information corresponding to the second adaptation model, that is, improve the similarity between the two model images. Conversely, decreasing the weight coefficient of the second incremental parameter can make the model image of the generated model image closer to the basic model image corresponding to the second basic model.

[0147] When the second image generation model includes multiple second adaptation models, increasing the weight coefficient of the second incremental parameter corresponding to a second adaptation model can make the model image of the generated model image closer to the model image of the character information corresponding to the second adaptation model. Conversely, decreasing the weight coefficient of a second incremental parameter can reduce the similarity between the model image of the generated model image and the model image of the character information corresponding to the second adaptation model.

[0148] By adjusting the weight coefficient corresponding to at least one second incremental parameter, a carrier image that meets user needs is generated, thereby improving the flexibility and accuracy of carrier image generation and further improving the generation effect of the carrier image.

[0149] In some embodiments, determining whether the carrier image meets the first image requirement may include:

[0150] Sending the carrier image to the client so that the client can display the carrier image;

[0151] In response to a first evaluation request for the carrier image sent by the client, determining a first evaluation result of the carrier image;

[0152] According to the first evaluation result, it is determined whether the carrier image meets the first image requirement.

[0153] The carrier image is sent to the client, and the user can evaluate the generated carrier image. The first evaluation result can include, for example, a first evaluation score. Taking a model image as an example, in an actual application, the model image can be evaluated based on multiple evaluation dimensions, such as authenticity, aesthetics, and consistency with the character information. The scores corresponding to the multiple dimensions are weighted to obtain the first evaluation score corresponding to the model image.

[0154] The first image requirement may include, for example, an evaluation score reaching a first score threshold. The first score threshold may be set according to actual needs, for example, 80 points, 90 points, etc. Based on the first evaluation result, for example, if the first evaluation score reaches the first score threshold, it may be determined that the carrier image meets the first image requirement.

[0155] By sending the carrier image to the client for the user to evaluate the generated carrier image, and determining whether the carrier image meets the first image requirements based on the first evaluation result, the effect of the generated carrier image is judged, and the subsequent display image is generated based on the carrier image that meets the first image requirements, that is, the carrier image with better generation effect, thereby further improving the generation effect of the display image.

[0156] In order to improve the generation effect of the display image, the generated display image can also be evaluated. Therefore, in some embodiments, after generating the display image, the above method can also include:

[0157] Sending the display image to the client so that the client can display the display image;

[0158] In response to a second evaluation request for the display image sent by the client, determining a second evaluation result of the display image;

[0159] According to the second evaluation result, it is determined whether the displayed image meets the second image requirement.

[0160] Based on this, searching for at least one target object that meets similarity requirements with at least one display object in the display image may include:

[0161] When the display image meets the second image requirement, at least one target object that meets similar requirements to at least one display object in the display image is searched for.

[0162] The display image is sent to the client, and the user can evaluate the generated display image. The second evaluation result can include, for example, a second evaluation score. In an actual application, the display image can be evaluated based on multiple evaluation dimensions, such as authenticity, aesthetics, consistency with attribute description information, and the carrier image. The scores corresponding to the multiple dimensions are weighted to obtain the second evaluation score corresponding to the display image.

[0163] The second image requirement may include, for example, an evaluation score reaching a second score threshold. The second score threshold may be set according to actual needs, for example, 90 points, 95 points, etc. Based on the second evaluation result, for example, if the second evaluation score reaches the second score threshold, it may be determined that the displayed image meets the second image requirement.

[0164] By sending the display image to the client for the user to evaluate the generated display image, and determining whether the display image meets the second image requirements based on the second evaluation result, the effect judgment of the generated display image is achieved, and then the subsequent target effect image is generated based on the display image that meets the second image requirements, that is, the display image with better generation effect, thereby further improving the generation effect of the target effect image.

[0165] In order to improve the generation effect of the target effect image, the generated target effect image can also be evaluated. Therefore, in some embodiments, after generating the target effect image and before generating the target content, the above method can also include:

[0166] Sending at least one target effect image to a client, so that the client displays the at least one target effect image;

[0167] In response to a third evaluation request for at least one target effect image sent by the client, determining a third evaluation result respectively corresponding to the at least one target effect image;

[0168] According to the third evaluation result, it is determined whether at least one target effect image meets the third image requirement.

[0169] Based on this, generating target content based on at least one target effect image and target copy includes:

[0170] When the target effect image meets the third image requirement, target content is generated based on at least one target effect image and the target text.

[0171] The target effect image is sent to the client, and the user can evaluate the generated target effect image. The third evaluation result may include, for example, a third evaluation score. In an actual application, the target effect image can be scored based on multiple evaluation dimensions, such as authenticity, aesthetics, and consistency with the displayed image and the object image of at least one target object. The scores corresponding to the multiple dimensions are weighted to obtain the third evaluation score corresponding to the target effect image.

[0172] The third image requirement may include, for example, an evaluation score reaching a third score threshold. The third score threshold may be set according to actual needs, for example, 80 points, 90 points, etc. Based on the third evaluation result, for example, if the third evaluation score reaches the third score threshold, it may be determined that the target effect image meets the third image requirement.

[0173] By sending the target effect image to the client for the user to evaluate the generated target effect image, and determining whether the target effect image meets the third image requirements based on the third evaluation result, the effect judgment of the generated target effect image is achieved, and then the subsequent target content is generated based on the target effect image that meets the third image requirements, that is, the target effect image with better generation effect, thereby further improving the generation effect of the target content.

[0174] One or more of the aforementioned embodiments provide a description of the process of online generation of a target effect image according to content generation requirements. In the embodiment of the present application, the target effect image may also be generated in advance.

[0175] Therefore, in some embodiments, the above method may further include:

[0176] Determine target character information and multiple effect images corresponding to the target character information.

[0177] Based on this, determining at least one target attribute description information matching the content generation requirement and the target effect image corresponding to each of the at least one target attribute description information may include:

[0178] At least one target attribute description information matching the content generation requirement is determined, and at least one target effect image corresponding to each of the at least one target attribute description information is determined from a plurality of effect images corresponding to the target character information.

[0179] The multiple effect images corresponding to the target character information can be pre-generated in the following manner:

[0180] Determine at least one attribute description information corresponding to the target role information;

[0181] For any attribute description information, using the first image generation model, based on the carrier image of the display carrier and the attribute description information, a display image is generated;

[0182] Searching for at least one target object that meets similar requirements to at least one display object in the display image;

[0183] Acquire object images corresponding to at least one target object;

[0184] An object image of at least one target object is fused into a display carrier in a display image to obtain an effect image corresponding to the attribute description information.

[0185] Optionally, the carrier image may be a carrier image generated by using a second image generation model and having a target image corresponding to the target character information.

[0186] Optionally, the generated effect image and its corresponding attribute description information may be stored in correspondence, so that target attribute description information matching the content generation requirement can be generated later, and the target effect image corresponding to the target attribute description information can be searched from the stored data.

[0187] The generation process of the above-mentioned multiple effect images can refer to the generation process of the target effect image described in the aforementioned embodiment part, and will not be repeated here.

[0188] One or more of the above embodiments have described the process of generating the target effect image. Now, the process of generating the target text will be described.

[0189] In some embodiments, the above method may further include:

[0190] Determine target role information and expression description information corresponding to the target role information.

[0191] The expression description information may include expression style, description tone, etc.

[0192] Based on this, generating a target copy based on the content generation requirements and object information corresponding to a target object in at least one target effect image may include:

[0193] A target copy is generated based on the expression description information, content generation requirements, and object information corresponding to a target object in at least one target effect image.

[0194] Optionally, a text generation model may be used to generate a target text based on the expression description information, content generation requirements, and object information corresponding to a target object in at least one target effect image.

[0195] Specifically, first prompt information can be generated based on the expression description information, content generation requirements and object information corresponding to the target object in at least one target effect image, and the first prompt information can be input into the text generation model to instruct the text generation model to generate the target copy.

[0196] By determining the target role information and the corresponding expression description information, combining the expression description information, content generation requirements and object information corresponding to the target object in at least one target effect image, the target copy is generated, thereby improving the consistency between the target copy and the target role information, improving the readability of the target copy, and achieving better generation effect.

[0197] In some embodiments, determining the expression description information corresponding to the target role information may include:

[0198] From the expression description information corresponding to different role information, search for the expression description information corresponding to the target role information.

[0199] The expression description information corresponding to different role information can be pre-generated in the following manner:

[0200] For any role information, the information generation model is used to generate the expression description information corresponding to the role information.

[0201] The information generation model can be an artificial intelligence-based language model (LM) or a multimodal model (MM). The embodiments of the present application do not limit the number of model parameters supported by the model, with the goal of meeting actual needs. The information generation model can be a deep learning model for processing and generating natural language text or multimodal data, which can be implemented based on a neural network architecture and can be pre-trained on a large amount of data.

[0202] It should be noted that the information generation model and the aforementioned text generation model can be the same model or different models.

[0203] By pre-generating the expression description information corresponding to different role information, when generating the target copy, it is only necessary to find the expression description information corresponding to the target role information, thereby improving the operation efficiency.

[0204] Optionally, the above method may further include:

[0205] The expression description information is converted into a standard data structure and stored in the database. This can improve the efficiency of searching the expression description information corresponding to the target role information, thereby further improving the operation efficiency.

[0206] In order to improve the accuracy and comprehensiveness of the target text, the target text can also be generated in combination with the image description information corresponding to the target effect image. Therefore, in some embodiments, the above method can also include:

[0207] Image description information is respectively extracted from at least one target effect image.

[0208] Among them, it can be achieved by using an information extraction model, which will not be described in detail.

[0209] Based on this, generating the target copy based on the expression description information, the content generation requirements, and the object information corresponding to the target object in at least one target effect image may include:

[0210] A target copy is generated based on the image description information, expression description information, content generation requirements corresponding to at least one target effect image and object information corresponding to a target object in at least one target effect image.

[0211] Optionally, a text generation model may be used to generate target text based on image description information, expression description information, content generation requirements corresponding to the at least one target effect image, and object information corresponding to the target object in the at least one target effect image.

[0212] Specifically, a first prompt information can be generated based on the image description information, expression description information, content generation requirements corresponding to the at least one target effect image, and the object information corresponding to the target object in the at least one target effect image, and the first prompt information can be input into the text generation model to instruct the text generation model to generate the target copy.

[0213] By combining the image description information and expression description information corresponding to at least one target effect image, as well as the content generation requirements and the object information corresponding to the target object in at least one target effect image, the target copy is generated, thereby improving the consistency between the target copy and the target effect image, as well as the expression description information, improving the comprehensiveness and accuracy of the target copy, and achieving better generation effect.

[0214] For example, in e-commerce scenarios, target copy can be generated by combining relevant content and selling points with target products. For example, in clothing, hot content could include popular clothing items and trending outfits. This further improves the comprehensiveness, accuracy, and readability of the target copy.

[0215] In order to improve the generation effect of the target copy, the target copy can also be evaluated. Therefore, in some embodiments, after generating the target copy, before generating the target content, the above method can also include:

[0216] Send the target copy to the client so that the client can display the target copy;

[0217] In response to a fourth evaluation request for the target document sent by the client, determining a fourth evaluation result corresponding to the target document;

[0218] Based on the fourth evaluation result, it is determined whether the target copy meets the copy requirements.

[0219] Based on this, generating target content based on at least one target effect image and target copy may include:

[0220] When the target copy meets the copy requirements, target content is generated based on at least one target effect image and the target copy.

[0221] The target document is sent to the client, and the user can evaluate the generated target document. The fourth evaluation result can include, for example, a fourth evaluation score. In an actual application, the target document can be scored based on multiple evaluation dimensions, such as its logic, readability, consistency with content generation requirements, and object information corresponding to a target object in at least one target effect image. The scores corresponding to the multiple dimensions are weighted to obtain the fourth evaluation score corresponding to the target document.

[0222] The copywriting requirements may include, for example, an evaluation score reaching a fourth score threshold. The fourth score threshold may be set based on actual needs, for example, 80 points, 90 points, etc. Based on the fourth evaluation result, for example, if the fourth evaluation score reaches the fourth score threshold, it may be determined that the target copywriting meets the copywriting requirements.

[0223] By sending the target copy to the client for the user to evaluate the generated target copy, and determining whether the target copy meets the copy requirements based on the fourth evaluation result, the effect of the generated target copy is judged, and the subsequent target content is generated based on the target copy that meets the copy requirements, that is, the target copy with better effect, thereby further improving the generation effect of the target content.

[0224] In actual applications, after the target content is generated, the target content may also be published. Therefore, in some embodiments, after the target content is generated, the above method may further include:

[0225] Publish targeted content;

[0226] Alternatively, when at least one recommended object includes one or more target objects associated with the target content, the content prompt information of the target content is used as the recommendation prompt information of the one or more target objects.

[0227] The one or more target objects associated with the target content may refer to one or more target objects among all target objects in all target effect images corresponding to the target content.

[0228] The target content may be published on any or a designated content publishing platform.

[0229] The content prompt information can be generated based on the target content and can be implemented in one or more forms such as text, images, links, etc. This application does not impose any restrictions on this.

[0230] Target object recommendation is achieved by publishing target content or using corresponding content prompt information as recommendation prompt information of its associated target object.

[0231] Figure 4 This is a flowchart of an embodiment of an object recommendation method provided by this application. The technical solution of this embodiment can be applied to the server. The method may include the following steps:

[0232] 401: In response to an object recommendation request, determine at least one recommended object.

[0233] Optionally, the object recommendation request may be an object search request. Therefore, before this, the above method may further include:

[0234] Receive object search requests sent by clients.

[0235] The client may generate an object search request in response to an object search operation in the user interface and send the request to the server.

[0236] 402 : When at least one recommended object includes one or more target objects associated with the target content, use content prompt information of the target content as recommendation prompt information of the one or more target objects.

[0237] In which, the target content can be generated based on at least one target effect image and a target copy, the target attribute description information corresponding to each of the at least one target effect images matches the content generation requirements, the target effect image includes at least one target object and a display carrier, the target effect image is obtained based on the fusion of the display image and the object image of at least one target object, the display image includes the display carrier and at least one display object that meets the target attribute description information, and at least one target object and at least one display object meet similarity requirements; the target copy can be generated based on the content generation requirements and the object information corresponding to the target object in at least one target effect image.

[0238] The generation methods of the display image, target effect image, target text, and target content have been described in detail in the aforementioned embodiments and will not be elaborated on here.

[0239] Optionally, in the case where the object recommendation request is an object search request sent by the client, when at least one recommended object includes one or more target objects associated with the target content and the client meets the content recommendation requirements, the content prompt information of the target content can be used as recommendation prompt information for one or more target objects.

[0240] The client meeting the content recommendation requirements may, for example, refer to the client meeting the content recommendation requirements when searching for a specified content on a platform, and may be set according to actual needs.

[0241] 403: Output recommendation prompt information corresponding to at least one recommendation object.

[0242] Optionally, in the case where the object recommendation request is an object search request sent by the client, recommendation prompt information corresponding to at least one recommended object may also be sent to the client as a search result.

[0243] In this embodiment, by automatically generating target text based on content generation requirements and object information of the target object, and then automatically generating target content based on the target text and at least one target effect image, automatic generation of target content matching the content generation requirements is achieved. Furthermore, when at least one recommended object includes one or more target objects associated with the target content, the content prompt information of the target content is used as recommendation prompt information for one or more target objects, thereby achieving target object recommendation.

[0244] It should be noted that in some of the processes described in the above embodiments and the accompanying drawings, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0245] For ease of understanding, Figure 5 Shows a schematic diagram of the system architecture in a practical application. Figure 5 The schematic diagram shown illustrates the technical solution of this application.

[0246] Taking the clothing product recommendation scenario in e-commerce scenarios as an example, in the existing technology, the target content is usually obtained by inviting real-life models to try on clothing products, taking pictures of the trying-on effects, and manually generating corresponding recommendation copy to recommend clothing products. This is costly and inefficient.

[0247] In the solution of the embodiment of this application, Figure 5 As shown, the system architecture may include a server 501 and a client 502 .

[0248] Among them, the server 501 can obtain user attribute information corresponding to multiple target users and multiple user-generated content. The target users may include, for example, users who publish content on the target content platform, user interaction data such as the number of users they follow, the number of users they follow, the number of messages they leave, the number of discussions they have had, and other users who meet the interaction requirements. The target users may be, for example, fashion experts, that is, users who love to match and share clothes. User attribute information may include, but is not limited to, user gender, age, height, region, season, and other information. User-generated content may include, for example, dressing style, shooting scene, etc. Dressing style may include, for example, dress dressing, skirt dressing, sports suit dressing, etc., and shooting scene may include, for example, indoor scene, outdoor scene, etc. The server 501 may perform data analysis on any user attribute information and any user-generated content to determine at least one attribute feature corresponding to the user attribute information and at least one content feature corresponding to the user-generated content, and form a character information, i.e., character setting information, which may also be called a character portrait, from one or more attribute features and one or more content features. In a practical application, a character portrait may be, for example, a 25-year-old woman wearing a dress in an indoor scene.

[0249] For any character portrait, server 501 can also generate a carrier image with the target image corresponding to the character portrait. For example, if the carrier image is a model image, as an optional implementation, server 501 can determine multiple sample model images with the target image from a sample model image library and, using an image generation model, perform fitting processing on the multiple sample model images to generate the model image.

[0250] As another optional implementation, the server 501 can add at least one adaptation model to the base model to form an image generation model, freeze the model parameters corresponding to the base model, and train at least one adaptation model using training data to obtain the incremental parameters corresponding to at least one adaptation model, thereby using the base model and the adaptation model to generate a model image with the target image. Among them, the base model can be a language model (Language Mode, LM) or a multimodal model (Multimodal Model, MM) based on artificial intelligence, etc. The embodiment of the present application does not limit the number of model parameters supported by the model, with the goal of meeting actual needs. The base model can be a deep learning model used to process natural language text, generate images or multimodal data, and can be implemented based on a neural network architecture, which can be pre-trained on a large amount of data. The adaptation model can refer to a model parameter adjustment model, and the incremental parameters of the adaptation model can be used to adjust the model parameters of the base model. In a practical application, the adaptation model can be implemented as a low-rank adaptation matrix (Low-Rank Adaptation, referred to as LoRA).

[0251] It should be noted that, in the embodiment of the present application, the model object in the model image is a virtual object.

[0252] At least one adaptation model is obtained through training, and an image generation model is constructed in combination with the basic model. It is possible to use one image generation model to process different character portraits to generate corresponding model images, thereby improving the applicability of the image generation model. In addition, each adaptation model is independent of each other, which can avoid interfering with the data processing of other character portraits and improve the accuracy of data processing. In addition, during the model training process, by freezing the model parameters of the basic model and only training the incremental parameters of the adaptation model, the complexity of model training is reduced, and the model training efficiency and data processing efficiency are improved.

[0253] In order to improve the generation effect of the model image, the server 501 can also send the model image to the client for the user to evaluate the model image. For example, the model image can be evaluated from multiple evaluation dimensions such as authenticity, aesthetics, and consistency with the character portrait, and the model image can be judged based on the evaluation results to determine whether it meets the first image requirement. The first image requirement can be set according to actual needs. If the model image meets the first image requirement, subsequent operations such as generating a display image can be performed based on the model image. If the model image does not meet the first image requirement, the weight coefficient corresponding to at least one incremental parameter can be adjusted. For example, increasing the weight coefficient corresponding to a certain incremental parameter can make the model image of the generated model image closer to the model shape of the character portrait corresponding to the adaptation module. According to the adjusted weight coefficient, the model image is regenerated, and the operation of determining whether the model image meets the first image requirement is continued.

[0254] By evaluating the generated model images, subsequent display images are generated based on the model images that meet the first image requirements, that is, the model images with better generation effects, thereby further improving the generation effect of the display images. When the first image requirements are not met, by adjusting the weight coefficient corresponding to at least one incremental parameter, it is possible to generate model images that meet user needs, thereby improving the flexibility and accuracy of model image generation, and thereby improving the generation effect of the model images.

[0255] For any character portrait, the server 501 can also determine at least one attribute description image set corresponding to the character portrait, such as a clothing style image set and a scene style image set. The server 501 can use the information extraction model to extract the clothing style description information and the scene style description information from the image set, and combine it with the above-mentioned model image and use the image generation model to generate a display image, which can be called a pose image. The pose image can include a model object and at least one display garment that matches the clothing style description information and the scene style description information, and the at least one display garment is virtual.

[0256] Specifically, the image generation model may include a base model and an adaptation model added to the base model, corresponding to at least one attribute description information. The server 501 can freeze the model parameters of the base model, use the training data corresponding to the attribute description information, train the corresponding adaptation model, and obtain the incremental parameters corresponding to the adaptation model. In a practical application, the adaptation module can be implemented as a low-rank adaptation matrix (Low-Rank Adaptation, referred to as LoRA), and the training process will not be repeated.

[0257] It should be noted that this base model can be the same as the base model used in the aforementioned image generation model for generating model images. Furthermore, each adaptation module is independent of the others, and each adaptation model has different incremental parameters, corresponding to different attribute descriptions or persona profiles.

[0258] To improve the quality of pose image generation, server 501 can also send the pose image to the client for user evaluation. For example, the pose image can be evaluated based on multiple evaluation dimensions, such as authenticity, aesthetics, and consistency with attribute description information and the model image. Based on the evaluation results, the pose image is judged to determine whether it meets the second image requirement, which can be set according to actual needs. If the pose image meets the second image requirement, subsequent operations such as generating a try-on effect image can be performed based on the pose image.

[0259] By sending the posture image to the client for the user to evaluate the generated posture image, subsequent try-on effect images can be generated based on the posture image that meets the second image requirements, that is, the posture image with better generation effect, thereby further improving the generation effect of the try-on effect image.

[0260] For any display clothing in any posture image, the server 501 can also search the product database for a target clothing product whose clothing image and the image of the display clothing meet the similarity requirements, and obtain the corresponding target clothing image. Specifically, the server 501 can preferentially search whether there is a tiled image of the target clothing image. If so, the tiled image can be directly used as the target clothing image. If not, a wearing image of the target clothing product can be obtained, such as a model trying-on image provided by a merchant, and the wearing image can be cropped to obtain the clothing image of the target clothing product. In a practical application, if the wearing image includes multiple target clothing products, such as an upper body clothing product and a lower body clothing product, the wearing image can be cropped to obtain a first sub-image corresponding to the upper body clothing product and a second sub-image corresponding to the lower body clothing product. The first sub-image is the target clothing image corresponding to the upper body clothing product, and the second sub-image is the target clothing image corresponding to the lower body clothing product.

[0261] Afterwards, the server 501 may fuse the target clothing image of at least one target clothing item into the model image in the pose image to obtain a fitting effect image. This may be achieved using an image fusion model, which may be a deep learning model for processing and generating image data and may be implemented based on a neural network architecture. In a practical application, the image fusion model may be constructed based on a generative adversarial network (GAN), and its model parameters may be pre-trained using training data.

[0262] To improve the generation of the try-on effect image, the server 501 may also send the try-on effect image to the client for the user to evaluate the try-on effect image. For example, the try-on effect image may be evaluated based on multiple evaluation dimensions, such as its aesthetics and consistency with the attribute description information, the model image, and the target clothing image. Based on the evaluation results, it is determined whether the posture image meets the third image requirement, which can be set according to actual needs. If the try-on effect image meets the third image requirement, subsequent operations such as generating target content can be performed based on the try-on effect image.

[0263] By sending the try-on effect image to the client, the user can evaluate the generated try-on effect image, so that subsequent target content can be generated based on the try-on effect image that meets the third image requirements, that is, the try-on effect image with better effect is generated, thereby further improving the generation effect of the target content.

[0264] Optionally, the server 501 may also store the evaluated try-on effect images and their corresponding attribute description information, such as dressing style description information and scene style description information.

[0265] In addition, for any character portrait, the server 501 can also use the information generation model to generate the expression description information corresponding to the character portrait, such as expression style, description tone, etc., and convert it into a standard data structure and store it in the database.

[0266] Therefore, based on a certain target persona portrait, the corresponding target expression description information can be searched to generate recommendation copy according to the target expression description information and the product description information of the target clothing product.

[0267] Optionally, the server 501 can also combine the image description information extracted from the fitting effect image corresponding to the target personality portrait, the target expression description information, the product description information of the target clothing product, the relevant selling point information, the hot wear, etc., to jointly generate the recommendation copy to improve the comprehensiveness, accuracy and readability of the recommendation copy.

[0268] According to the above-generated try-on effect images and recommendation texts, the server 501 can generate target content.

[0269] In actual applications, when there is a demand for content generation, the server 501 can determine the content generation requirements, at least one target attribute description information that matches the content generation requirements, and the target persona portrait, and determine at least one target fitting effect image corresponding to each of the at least one target attribute description information from the multiple fitting effect images corresponding to the target persona portrait, and generate a target copy based on the content generation requirements, the target expression description information corresponding to the target persona portrait, and the product description information corresponding to the target clothing product in the at least one target fitting effect image. The server 501 can add at least one target fitting effect image and the target copy to the template according to the target content template to obtain the target content. Optionally, multiple target fitting effect images can be spliced ​​to obtain a spliced ​​effect image, and the spliced ​​effect image can be added to the template homepage area for user viewing.

[0270] The server 501 can publish the target content, or can receive a clothing product search request sent by the client 502. When at least one recommended clothing product includes one or more target clothing products associated with the target content, the server 501 uses the content prompt information of the target content as the recommendation prompt information of the one or more target clothing products, and sends the recommendation prompt information as the search result to the client 502.

[0271] It can be seen that the embodiment of the present application provides a full-link automatic generation solution including character portraits, model images, posture images, try-on effect images, target copy and target content, which improves operational efficiency and reduces generation costs.

[0272] Figure 6 This is a schematic diagram of the structure of an embodiment of a content generation device provided by this application. The device may include the following modules:

[0273] A first determining module 601 is used to determine content generation requirements;

[0274] The second determining module 602 is configured to determine at least one target attribute description information that matches the content generation requirement, and a target effect image corresponding to each of the at least one target attribute description information.

[0275] The target effect image includes at least one target object and a display carrier, the target effect image is obtained based on the fusion of the display image and the object image of the at least one target object, the display image includes the display carrier and at least one display object that meets the target attribute description information, and the at least one target object and the at least one display object meet the similarity requirement;

[0276] A first generating module 603 is configured to generate a target text based on content generation requirements and object information corresponding to a target object in at least one target effect image;

[0277] The second generating module 604 is configured to generate target content based on at least one target effect image and a target text; wherein the target content is used to recommend a target object in the at least one target effect image.

[0278] In some embodiments, the apparatus may further comprise:

[0279] The third determination module is used to determine the target role information;

[0280] A fourth determining module is used to determine the expression description information corresponding to the target role information;

[0281] The second determining module 602 may be specifically configured to determine at least one target attribute description information matching the content generation requirement, and to determine at least one target effect image corresponding to each of the at least one target attribute description information from the plurality of effect images corresponding to the target character information;

[0282] The first generating module 603 may be specifically configured to generate a target text based on the expression description information, the content generation requirements, and the object information corresponding to the target object in at least one target effect image.

[0283] In some embodiments, the second determining module 602 may include:

[0284] a determining unit, configured to determine at least one target attribute description information matching the content generation requirement;

[0285] a generating unit configured to generate a display image for any target attribute description information by using a first image generation model based on a carrier image of a display carrier and the target attribute description information;

[0286] a search unit, configured to search for at least one target object that meets similarity requirements with at least one display object in the display image;

[0287] an acquiring unit, configured to acquire an object image corresponding to at least one target object;

[0288] The fusion unit is used to fuse the object image of at least one target object into the display carrier in the display image to obtain a target effect image corresponding to the target attribute description information.

[0289] In some embodiments, the apparatus may further comprise:

[0290] The third determination module is used to determine the target role information;

[0291] The third generation module is used to generate a carrier image having a target image corresponding to the target character information by using the second image generation model.

[0292] In some embodiments, the generation unit may be specifically configured to generate a display image based on the model image and the target attribute description information using a first basic model in the first image generation model and a first adaptation model corresponding to the target attribute information.

[0293] In some embodiments, the third generation module can be specifically used to determine the weight coefficient corresponding to at least one second incremental parameter; according to the weight coefficient, using the second image generation model, generate a carrier image with a target image corresponding to the target character information.

[0294] In some embodiments, the apparatus may further comprise:

[0295] A judging module, configured to judge whether the carrier image meets the first image requirement;

[0296] an adjustment module, for adjusting the weight coefficient corresponding to each of the at least one second incremental parameters if the result of the determination module is negative, and regenerating a carrier image having a target image corresponding to the target character information using the second image generation model according to the adjusted weight coefficient, and continuing to perform the operation of determining whether the carrier image meets the first image requirement;

[0297] The generating unit is configured to generate a display image based on the carrier image of the display carrier and the target attribute description information by using the first image generation model when the result of the judging module is yes.

[0298] In some embodiments, the judgment module can be specifically used to send the carrier image to the client so that the client can display the carrier image; determine a first evaluation result of the carrier image in response to a first evaluation request for the carrier image sent by the client; and determine whether the carrier image meets the first image requirement based on the first evaluation result.

[0299] In some embodiments, the third generation module can be specifically used to determine multiple sample carrier images having a target image corresponding to the target character information; and use the second image generation model to fit the multiple sample carrier images to generate a carrier image.

[0300] In some embodiments, the search unit may be specifically configured to search, for any display object, from the object database for a target object whose image similarity with the display object satisfies a similarity requirement;

[0301] The acquisition unit can be specifically used to search for any target object to see whether there is a tiled image of the target object; if so, use the tiled image as the object image; if not, obtain the wearing image of the target object and crop the wearing image to obtain the object image of the target object.

[0302] In some embodiments, the third determination module can be specifically used to determine target role information from multiple role information; multiple role information is obtained in the following manner: user attribute information corresponding to multiple target users and multiple user-generated content are obtained; data analysis is performed on any user attribute information and any user-generated content to determine at least one attribute feature corresponding to the user attribute information and at least one content feature corresponding to the user-generated content; one or more attribute features and one or more content features constitute a role information.

[0303] In some embodiments, the fourth determination module can be specifically used to search for the expression description information corresponding to the target role information from the expression description information corresponding to different role information; the expression description information corresponding to different role information is pre-generated in the following manner: for any role information, the information generation model is used to generate the expression description information corresponding to the role information.

[0304] In some embodiments, the apparatus may further comprise:

[0305] An extraction module, configured to extract image description information from at least one target effect image;

[0306] The first generation module 603 can be specifically used to generate prompt information based on the image description information, expression description information, content generation requirements corresponding to at least one target effect image and the object information corresponding to the target object in at least one target effect image, and input the prompt information into the text generation model to instruct the text generation model to generate the target copy.

[0307] In some embodiments, the apparatus may further comprise:

[0308] Publishing module, used to publish target content;

[0309] Alternatively, the recommendation module is configured to use the content prompt information of the target content as the recommendation prompt information of the one or more target objects when at least one recommended object includes one or more target objects associated with the target content.

[0310] Figure 6 The content generating device may execute Figure 1The implementation principle and technical effects of the content generation method described in the embodiment are not described in detail here. The specific manner in which each module and unit performs operations in the content generation device in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0311] Figure 7 This is a schematic diagram of the structure of an embodiment of an object recommendation device provided by this application. The device may include the following modules:

[0312] A fifth determining module 701 is configured to determine at least one recommended object in response to an object recommendation request;

[0313] A sixth determining module 702 is configured to use content prompt information of the target content as recommendation prompt information of the one or more target objects when at least one recommended object includes one or more target objects associated with the target content;

[0314] The output module 703 is configured to output recommendation prompt information corresponding to at least one recommendation object.

[0315] In some embodiments, the object recommendation request is an object search request, and the apparatus may further include:

[0316] A receiving module, configured to receive an object search request sent by a client;

[0317] The sixth determining module 702 may be specifically configured to use content prompt information of the target content as recommendation prompt information for the one or more target objects if at least one recommended object includes one or more target objects associated with the target content and the client meets the content recommendation requirements;

[0318] The device may also include:

[0319] The sending module is used to send the recommendation prompt information corresponding to at least one recommendation object as a search result to the client.

[0320] Figure 7 The object recommendation device can execute Figure 4 The object recommendation method described in the illustrated embodiment, its implementation principle and technical effects are not repeated here. The specific manner in which each module and unit performs operations in the object recommendation device in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0321] Figure 8 This is a schematic diagram of a computing device according to an embodiment of the present application. Figure 8 As shown, in actual applications, the computing device may include: a storage component 801 and a processing component 802.

[0322] The storage component 801 is used to store computer programs and can be configured to store various other data to support operations on the computing device. Examples of such data include instructions for any application or method operating on the computing device, data structures, contact data, phone book data, messages, images, videos, etc.

[0323] The processing component 802 is coupled to the storage component 801 and is used to execute the computer program in the storage component 801 to implement the following Figure 1 The content generation method shown, or Figure 3 The target effect image method shown, or Figure 4 The object recommendation method shown.

[0324] Further, if Figure 8 As shown, the computing device may further include: a communication component 803, a display component 804, a power component 805, an audio component 806 and other components. Figure 8 Only some components are shown schematically, which does not mean that the equipment only includes Figure 8 In addition, Figure 8 The components in the dotted box are optional components, not mandatory components, and depend on the specific product form of the computing device. The computing device of this embodiment can be implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone or an IOT (Internet of Things) device, or a server device such as a conventional server, a cloud server or a server array. If the computing device of this embodiment is implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone, etc., it can include Figure 8 If the computing device of this embodiment is implemented as a server device such as a conventional server, a cloud server or a server array, it may not include the components in the dotted box; Figure 8 Components within the dotted box.

[0325] The processing component includes one or more processors to execute computer instructions to perform all or part of the steps in the above method. Of course, the processing component can also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0326] The above-mentioned storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0327] The communication component is configured to facilitate wired or wireless communication between the device in which the communication component resides and other devices. The device in which the communication component resides may access a wireless network based on a communication standard, such as a mobile communication network, or a combination thereof. In an exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel.

[0328] The display assembly may include a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touches, slides, and gestures on the touch panel. The touch sensor may not only sense the boundaries of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

[0329] The power supply assembly provides power to various components of the device in which the power supply assembly is located. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply assembly is located.

[0330] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC), and when the device where the audio component is located is in an operating mode, such as call mode, recording mode, and voice recognition mode, the microphone is configured to receive external audio signals. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0331] Accordingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to implement the steps in the above method embodiment. The computer-readable storage medium includes volatile or non-volatile or a combination thereof, and may be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic cassette, tape disk storage or other magnetic storage device or any other non-transmission medium.

[0332] Accordingly, an embodiment of the present application further provides a computer program product, which includes a computer program or instructions, and when the computer program or instructions are executed by a processor, the processor is enabled to implement the steps in the above-mentioned method embodiment. It should be understood that each process or a combination of multiple processes in the above-mentioned method flow can be implemented by a computer program or instruction. In addition, these computer programs or instructions can be applied to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device, so that the processor of the general-purpose computer, the special-purpose computer, the embedded processor or other programmable data processing device can be implemented as a device for implementing the corresponding functions in the above-mentioned method embodiment.

[0333] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0334] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0335] Finally, it should be noted that the above are merely examples of the present application and are not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application are intended to be included within the scope of the claims of the present application.

Claims

1. A content generation method, characterized in that: include: Determine content generation requirements; Determining at least one target attribute description information matching the content generation requirement, and a target effect image corresponding to each of the at least one target attribute description information; the target effect image includes at least one target object and a display carrier, the target effect image is obtained based on a fusion of a display image and an object image of the at least one target object, the display image includes the display carrier and at least one display object matching the target attribute description information, and the at least one target object and the at least one display object meet a similarity requirement; generating a target copy based on the content generation requirement and object information corresponding to a target object in at least one target effect image; Target content is generated based on the at least one target effect image and the target text; wherein the target content is used to recommend a target object in the at least one target effect image.

2. The method according to claim 1, characterized in that Also includes: Determine target role information; Determining the expression description information corresponding to the target role information; The determining of at least one target attribute description information matching the content generation requirement and the target effect image corresponding to each of the at least one target attribute description information includes: determining at least one target attribute description information matching the content generation requirement, and determining, from a plurality of effect images corresponding to the target character information, a target effect image corresponding to each of the at least one target attribute description information; Generating a target copy based on the content generation requirement and object information corresponding to a target object in at least one target effect image includes: A target copy is generated based on the expression description information, the content generation requirements, and object information corresponding to a target object in at least one target effect image.

3. The method according to claim 1, characterized in that The determining of at least one target attribute description information matching the content generation requirement and the target effect image corresponding to each of the at least one target attribute description information includes: Determining at least one target attribute description information matching the content generation requirement; For any target attribute description information, using the first image generation model, based on the carrier image of the display carrier and the target attribute description information, a display image is generated; searching for at least one target object that meets similar requirements to at least one display object in the display image; Acquire object images corresponding to the at least one target object; The object image of the at least one target object is fused into the display carrier in the display image to obtain a target effect image corresponding to the target attribute description information.

4. The method according to claim 2, characterized in that The multiple effect images corresponding to the target character information are pre-generated in the following manner: Determine at least one attribute description information corresponding to the target role information; For any attribute description information, using the first image generation model, based on the carrier image of the display carrier and the attribute description information, a display image is generated; searching for at least one target object that meets similar requirements to at least one display object in the display image; Acquire object images corresponding to the at least one target object; The object image of the at least one target object is fused into the display carrier of the display image to obtain an effect image corresponding to the attribute description information.

5. The method according to claim 3 or 4, characterized in that Also includes: The carrier image having the target image corresponding to the target character information is generated by using the second image generation model.

6. The method according to claim 3, characterized in that The first image generation model is trained in the following manner: Adding at least one first adaptation model corresponding to each attribute description information to the first basic model to form a first image generation model; For any attribute description information, by freezing the first model parameters of the first basic model, the corresponding first adaptation model is trained using first training data corresponding to the attribute description information to obtain first incremental parameters corresponding to the first adaptation model; the first training data includes the sample attribute description information and the sample display image corresponding to the sample attribute description information; The generating of the display image by using the first image generation model based on the carrier image of the display carrier and the target attribute description information includes: A display image is generated based on a carrier image of a display carrier and the target attribute description information by utilizing a first basic model in the first image generation model and a first adaptation model corresponding to the target attribute description information.

7. The method according to claim 5, characterized in that The second image generation model is trained as follows: Adding at least one second adaptation model to the second basic model to form the second image generation model; By freezing the second model parameters corresponding to the second basic model, the at least one second adaptation model is trained using the second training data to obtain the second incremental parameters corresponding to each of the at least one second adaptation models; the second training data includes sample role information and a sample carrier image corresponding to the sample role information.

8. The method according to claim 7, characterized in that The step of generating a carrier image having a target image corresponding to the target character information by using the second image generation model includes: Determining a weight coefficient corresponding to each of the at least one second incremental parameters; According to the weight coefficient, the second image generation model is used to generate the carrier image having the target image corresponding to the target character information.

9. The method according to claim 8, characterized in that Also includes: Determining whether the carrier image meets a first image requirement; If not, adjusting the weight coefficient corresponding to each of the at least one second incremental parameters, and regenerating the carrier image having the target image corresponding to the target character information using the second image generation model according to the adjusted weight coefficient, and continuing to perform the operation of determining whether the carrier image meets the first image requirement; If so, the operation of generating a display image by using the first image generation model based on the carrier image of the display carrier and the target attribute description information is performed.

10. The method according to claim 9, characterized in that Determining whether the carrier image meets the first image requirement includes: Sending the carrier image to a client, so that the client displays the carrier image; determining a first evaluation result of the carrier image in response to a first evaluation request for the carrier image sent by the client; According to the first evaluation result, it is determined whether the carrier image meets the first image requirement.

11. The method according to claim 3, characterized in that The searching for at least one target object that meets similarity requirements with at least one display object in the display image includes: For any display object in the display image, searching from an object database for a target object whose image similarity with the display object meets similarity requirements; The acquiring of the object images corresponding to the at least one target object comprises: For any target object, searching whether there is a tiled image of the target object; If so, taking the tiled image as the object image; If not, obtain a wearing image of the target object, and perform cropping processing on the wearing image to obtain an object image of the target object.

12. The method according to claim 2, characterized in that The determining of target role information includes: Determine target role information from multiple role information; The multiple role information is obtained as follows: Obtaining user attribute information corresponding to multiple target users and multiple user-generated contents; Performing data analysis on any user attribute information and any user-generated content respectively to determine at least one attribute feature corresponding to the user attribute information and at least one content feature corresponding to the user-generated content; A role information is composed of one or more attribute features and one or more content features.

13. The method according to claim 2, characterized in that The step of determining the expression description information corresponding to the target role information includes: Searching for the expression description information corresponding to the target role information from the expression description information corresponding to different role information; The expression description information corresponding to the different role information is pre-generated in the following manner: For any role information, the information generation model is used to generate the expression description information corresponding to the role information.

14. The method according to claim 2, characterized in that Also includes: Extracting image description information from the at least one target effect image respectively; The generating of the target copy based on the expression description information, the content generation requirement, and object information corresponding to the target object in at least one target effect image includes: Based on the image description information corresponding to the at least one target effect image, the expression description information, the content generation requirements, and the object information corresponding to the target object in the at least one target effect image, prompt information is generated, and the prompt information is input into the text generation model to instruct the text generation model to generate the target copy.

15. The method according to claim 1, wherein Also includes: publishing the target content; Alternatively, in a case where at least one recommended object includes one or more target objects associated with the target content, the content prompt information of the target content is used as the recommendation prompt information of the one or more target objects.

16. An object recommendation method, characterized in that: include: In response to the object recommendation request, determining at least one recommended object; In a case where the at least one recommended object includes one or more target objects associated with the target content, using content prompt information of the target content as recommendation prompt information of the one or more target objects; The target content is generated based on at least one target effect image and a target text, the target attribute description information corresponding to each of the at least one target effect image matches the content generation requirements, the target effect image includes at least one target object and a display carrier, the target effect image is obtained based on the fusion of a display image and an object image of the at least one target object, the display image includes the display carrier and at least one display object that meets the target attribute description information, and the at least one target object and the at least one display object meet similarity requirements; the target text is generated based on the content generation requirements and the object information corresponding to the target object in the at least one target effect image; Output the recommendation prompt information corresponding to each of the at least one recommendation object.

17. The method according to claim 16, characterized in that The object recommendation request is an object search request; the method further includes: Receive object search requests sent by clients; In the case where the at least one recommended object includes one or more target objects associated with the target content, using the content prompt information of the target content as the recommendation prompt information of the one or more target objects includes: If the at least one recommended object includes one or more target objects associated with the target content and the client meets the content recommendation requirements, use the content prompt information of the target content as the recommendation prompt information of the one or more target objects; Also includes: The recommendation prompt information corresponding to each of the at least one recommendation object is sent to the client as a search result.

18. A computing device, characterized in that including processing components and storage components; The storage component stores a computer program; the computer program is used to be called and executed by the processing component to implement the content generation method according to any one of claims 1 to 15, or the object recommendation method according to any one of claims 16 to 17.

19. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by the processing component, the content generation method according to any one of claims 1 to 15 or the object recommendation method according to any one of claims 16 to 17 is implemented.

20. A computer program product, characterized in that The method comprises a computer program or an instruction, which, when executed by a processing component, implements the content generation method according to any one of claims 1 to 15, or the object recommendation method according to any one of claims 16 to 17.

Images