Image generation method and apparatus

Abstract

This application provides an image generation method and apparatus. The image generation method includes: acquiring drawing object information, determining a list of intelligent agents based on the drawing object information, and generating drawing theme parameters corresponding to each intelligent agent in the list; processing the corresponding drawing theme parameters using each intelligent agent to obtain element configuration information corresponding to each intelligent agent; processing the element configuration information corresponding to each intelligent agent through a prompt word generation module to obtain global drawing prompt words; and inputting the global drawing prompt words into an image generation model for processing to obtain at least one target object image. This technical solution can be widely applied in the field of next-generation information technology, such as artificial intelligence-based digital cultural and creative product production software and digital culture creation software.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to image generation methods and apparatus. Background Technology

[0002] With the development of artificial intelligence technology, current image generation techniques primarily rely on generative AI drawing models such as Midjourney, Stable Diffusion, and DALL-E. Their core mechanism depends entirely on users manually inputting prompts to complete the image generation process. Users must independently handle the entire process, from character design and creative conception to visual element design, prompt writing, parameter optimization, and result selection. This requires users to be proficient in multiple fields, including character narrative, aesthetic design, original art style, lighting and composition, and prompt engineering. Ordinary users often need a long learning curve and repeated debugging to produce high-quality work. Images generated by novice users generally suffer from severe homogenization, chaotic logical structure, and lack of detail, resulting in exceptionally high overall debugging time costs. Prompt-assisted tools, developed based on large language models, use simple keywords input by the user to output complete drawing prompts from a single large model. While some tools offer basic style and composition options, they still cannot fundamentally overcome the limitation of a single generation logic. Due to the lack of a collaborative mechanism for specialization within specific fields, such tools struggle to simultaneously meet multi-dimensional requirements such as narrative coherence, aesthetic professionalism, differentiated design, and compliance verification. The quality of generated prompts fluctuates significantly, easily leading to highly homogenized content, mediocre creative expression, and non-compliance with the standards of specific competitions or commercial scenarios. The existing AI drawing workflow, from role setting to effect evaluation, is completely fragmented, forcing users to frequently switch between multiple independent tools and interfaces. It lacks both standardized, fully automated closed-loop support and efficient batch generation and iterative optimization capabilities. Therefore, an effective solution is urgently needed to address these issues. Summary of the Invention

[0003] In view of this, embodiments of this application provide an image generation method to address the technical deficiencies existing in the prior art. Embodiments of this application also provide an image generation system, an image generation apparatus, a computing device, a computer-readable storage medium, and a computer program.

[0004] According to a first aspect of the embodiments of this application, an image generation method is provided, comprising: Obtain drawing object information, determine a list of agents based on the drawing object information, and generate drawing topic parameters corresponding to each agent in the agent list; By processing the corresponding drawing topic parameters for each agent, the feature configuration information for each agent can be obtained. The prompt word generation module processes the feature configuration information corresponding to each agent to obtain global drawing prompt words; The global drawing prompts are input into the image generation model for processing to obtain at least one target object image.

[0005] According to a second aspect of the embodiments of this application, an image generation system is provided, including a client and a server, comprising: The client is configured to determine the drawing object information in response to the drawing request submitted by the user, and send the drawing object information to the server. The server is configured to determine a list of agents based on the drawing object information, and generate drawing theme parameters corresponding to each agent in the list; process the corresponding drawing theme parameters using each agent to obtain feature configuration information corresponding to each agent; process the feature configuration information corresponding to each agent through a prompt word generation module to obtain global drawing prompt words; input the global drawing prompt words into an image generation model for processing to obtain at least one target object image, and send the target object image to the client.

[0006] According to a third aspect of the embodiments of this application, an image generation apparatus is provided, comprising: The information acquisition module is configured to acquire drawing object information, determine a list of agents based on the drawing object information, and generate drawing topic parameters corresponding to each agent in the list of agents. The parameter processing module is configured to use each agent to process the corresponding drawing topic parameters and obtain the feature configuration information for each agent. The information processing module is configured to process the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words; The image generation module is configured to input the global drawing prompts into the image generation model for processing to obtain at least one target object image.

[0007] According to a fourth aspect of the embodiments of this application, a computing device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor executes the computer-executable instructions to implement the steps of the image generation method.

[0008] According to a fifth aspect of the present application, a computer-readable storage medium is provided that stores computer-executable instructions that, when executed by a processor, implement the steps of the image generation method.

[0009] According to a sixth aspect of the present application, a chip is provided that stores a computer program, which, when executed by the chip, implements the steps of the image generation method.

[0010] According to a seventh aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the image generation method described above.

[0011] Based on the above technical solutions, this application further elaborates on its advantages. The image generation method provided by this application processes drawing topic parameters and generates global drawing prompts through a list of intelligent agents, realizing multi-dimensional professional division of labor and automated closed loop. It effectively solves the problems of single prompt generation logic, high user operation complexity, and unstable image quality in existing technologies. It has the advantages of improving image generation quality, reducing user operation complexity, and enhancing the diversity of creative expression, thereby further reducing the threshold for use in business scenarios and accelerating the efficiency and effectiveness of image generation in business. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the structure of a computer system for an image generation method provided in an embodiment of this application; Figure 2 This is a flowchart of an image generation method provided in an embodiment of this application; Figure 3 This is a flowchart illustrating an image generation method for a character generation scenario provided in an embodiment of this application. Figure 4 This is a schematic diagram of the structure of an image generation system provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an image generation device provided in an embodiment of this application; Figure 6 This is a structural block diagram of a computing device provided in one embodiment of this application; Figure 7 This is a structural diagram of a terminal provided in one embodiment of this application; Figure 8 This is a structural block diagram of a server provided in one embodiment of this application.

[0013] Obviously, the accompanying drawings are only some illustrative examples of embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort. Detailed Implementation Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0014] The terminology used in one or more embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of one or more embodiments of this application. The singular forms “a,” “the,” and “the” used in one or more embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” used in one or more embodiments of this application refers to and includes any or all possible combinations of one or more associated listed items. For example, “A and / or B” can represent three cases: only A exists, only B exists, and both A and B exist, where A and B can be singular or plural. The character “ / ” generally indicates that the preceding and following objects are in an “or” relationship. “At least one of the following” or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of A, B, or C can represent: A, B, C, “A and B,” “A and C,” “B and C,” or “A and B and C,” where A, B, and C can be single or multiple.

[0015] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those described herein. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this application, and similarly, second may also be referred to as first. Furthermore, the terms "comprising," "having," "including," and variations thereof all indicate non-exclusive inclusion, such as a process, method, system, product, or device that comprises a series of steps or units, not limited to the steps or units explicitly listed, but may also include other steps or units not explicitly listed or inherent in the process itself.

[0016] It should be noted that 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0017] First, the terminology used in one or more embodiments of the present invention will be explained.

[0018] Midjourney Prompts: These are text commands that users input into the AI ​​drawing tool Midjourney to describe the desired image. They can include elements such as subject, style, lighting, composition, and medium, and support parameter control (e.g., `--ar` aspect ratio, `--v` version). The quality of the prompts directly determines the quality of the generated image and is the core language of communication between humans and the AI ​​artist; hence, it's called a Prompt.

[0019] Prompt words are text instructions that users input into AI models, serving as a bridge for communication between humans and models. They typically include three elements: the instruction, the context, and the expected output, directly influencing the quality and direction of the model's response. Prompt engineering has become a core skill in AI application.

[0020] Playwright is an open-source, cross-browser automation framework that supports Chromium, Firefox, and WebKit, using a unified API to control the browser. Compared to Selenium, it requires no additional drivers, starts up faster, and has built-in automatic waiting and network interception capabilities. It supports languages ​​such as Python, JS, and Java, and is commonly used for automated testing and batch operations.

[0021] Large language models are deep learning models based on the Transformer architecture, trained on massive amounts of text, with billions to trillions of parameters. They can understand and generate natural language, performing tasks such as question answering, writing, translation, and programming. Their core capabilities come from the quality of the training data, rather than solely from the algorithm.

[0022] This application provides an image generation method. This application also relates to an image generation system, an image generation apparatus, a computing device, a computer-readable storage medium, and a computer program, which will be described in detail in the following embodiments.

[0023] In practical applications, existing AI drawing technologies require users to manually complete the entire process, from character design and creative conception to visual element design, prompt writing, and parameter optimization. This results in a complex workflow and a high level of expertise required. The mechanism of generating drawing prompts from a single large model cannot effectively coordinate multiple dimensions of needs, such as narrative coherence, aesthetic professionalism, differentiated design, and compliance verification. This leads to significant fluctuations in prompt quality and the generation of images being prone to homogenization, logical inconsistencies, and missing details. Furthermore, the fragmented nature of the image generation process requires users to repeatedly switch between multiple tool interfaces to filter results and evaluate effects, resulting in low automation and limited efficiency for batch generation and iterative optimization. Simultaneously, existing technologies lack specialized optimization capabilities for specific scenarios, making it difficult to flexibly adapt to different industries or personalized drawing needs. Insufficient scalability also affects the overall ease of operation and consistency of work quality.

[0024] In view of this, see Figure 1 The structural diagram shown illustrates the image generation method provided in this embodiment. By collaboratively processing drawing topic parameters and generating global drawing prompts through a list of intelligent agents, it achieves multi-dimensional professional division of labor and automated closed loop. This effectively solves the problems of simple prompt generation logic, high user operation complexity, and unstable image quality in existing technologies. It has the advantages of improving image generation quality, reducing user operation complexity, and enhancing the diversity of creative expression, thereby further lowering the threshold for use in business scenarios and accelerating the efficiency and effectiveness of image generation in business.

[0025] Figure 2 A flowchart of an image generation method according to an embodiment of this application is shown, which specifically includes the following steps: Step S202: Obtain drawing object information, determine a list of intelligent agents based on the drawing object information, and generate drawing topic parameters corresponding to each intelligent agent in the list of intelligent agents.

[0026] Step S204: Use each agent to process the corresponding drawing topic parameters to obtain the feature configuration information for each agent.

[0027] Step S206: The prompt word generation module processes the feature configuration information corresponding to each agent to obtain global drawing prompt words.

[0028] Step S208: Input the global drawing prompt into the image generation model for processing to obtain at least one target object image.

[0029] The image generation method provided in this embodiment can be applied to any image generation scenario, such as games, comics, teaching, social applications, etc., to generate images related to characters, buildings, and objects, reducing the operational threshold for image generation and ensuring that the generated images better meet user needs.

[0030] Specifically, drawing object information refers to comprehensive data describing the theme, style, content, composition, name, background, and core tags of the image the user expects to generate. This information can exist in structured or unstructured form, such as text descriptions, keyword sets, image references, and preset template selections. Its role is to provide initial, high-level guidance for subsequent image generation processes. The agent list refers to a collection of multiple independent functional modules or processing units determined based on the drawing object information, used to collaboratively complete the image generation task. Each agent is designed to handle a specific dimension or professional domain in the image generation process, such as character design, scene construction, lighting effects, style rendering, composition, modeling, lighting, and emotion. The dynamic determination of the agent list allows the system to flexibly adapt to different drawing needs. Drawing theme parameters refer to the specific drawing instructions or configuration items assigned to each agent in the agent list, related to its professional domain. These parameters refine and decompose the drawing object information, guiding each agent to perform specific design and configuration within its responsible domain. For example, a character agent can be assigned theme parameters such as character features and clothing style. For example, the lighting agent can be assigned thematic parameters such as main light, fill light, surface lighting, contour light, special effects light, skin light, and ambient light. The composition agent can be assigned thematic parameters such as perspective, line-of-sight height, subject position, depth, and dynamic direction. The styling agent can be assigned thematic parameters such as hairstyle, lip color, makeup, and upper body styling. The scene agent can be assigned thematic parameters such as main location, dynamic elements, fog, foreground structures, architecture, and decorative elements. The emotion agent can be assigned thematic parameters such as core emotion, surface, middle, and bottom layers, relational boundaries, and audience reaction. In specific implementations, the selection of agents and related drawing thematic parameters can be chosen according to actual needs; this embodiment does not impose any limitations.

[0031] Correspondingly, an agent refers to a software entity that undertakes a specific task and possesses specific processing capabilities in an image generation method. Each agent is configured to receive drawing theme parameters and generate feature configuration information for its domain based on these parameters. Agents can work independently or collaborate by sharing information. Feature configuration information refers to the detailed configuration data output by each agent after processing the received drawing theme parameters, describing the specific image features within its responsible domain. This information can be structured data, such as color codes, shape descriptions, texture types, and position coordinates, or it can be a semi-structured description used to guide subsequent prompt generation. The prompt generation module is a functional unit specifically designed to integrate, transform, and optimize the feature configuration information output by multiple agents into global drawing prompts that meet the input requirements of the image generation model. This module is configured to understand the heterogeneous information output by different agents and unify it into coherent and effective text prompts. Global drawing prompts refer to the complete text instructions generated by the prompt generation module and used to drive the image generation model. The prompt integrates feature configuration information generated by all agents within their respective domains, describing in detail various aspects of the target image in natural language or a specific text format, thereby guiding the image generation model to produce an image that meets expectations. The image generation model is a computational model built based on deep learning or other artificial intelligence techniques, capable of generating corresponding images based on input text prompts. This model is trained to understand the complex mapping relationship between text descriptions and image features, and is able to generate high-quality, diverse visual content. The target object image refers to the final image obtained after processing the global drawing prompts through the image generation model. These images are the output of this method, designed to meet the user's initial drawing needs and reflect the results of the collaborative work of multiple agents.

[0032] Therefore, various methods can be used to obtain information about the drawing object. For example, users can directly type a description of the target image into a text input box, such as "a painting depicting a future city night scene, including aircraft and towering buildings." Alternatively, the system can provide a series of preset templates or keyword options, which users can select or check to combine and form the drawing object information. Furthermore, users can upload reference images, and the system can perform preliminary analysis of the image content to extract key drawing object information. All these methods aim to obtain the user's basic intent and needs regarding image generation.

[0033] Furthermore, based on the acquired drawing object information, a list of intelligent agents is determined, and drawing theme parameters corresponding to each intelligent agent in the list are generated. For example, when the drawing object information is described as "future city night view," the system can determine the required "city building intelligent agent," "transportation intelligent agent," and "lighting effect intelligent agent" based on preset rules or mapping relationships. Subsequently, for each determined intelligent agent, its corresponding drawing theme parameters are extracted or generated from the drawing object information. For example, the "city building intelligent agent" may be assigned theme parameters such as "towering" and "futuristic style," while the "transportation intelligent agent" may be assigned theme parameters such as "aircraft" and "science fiction feel." These parameters can exist in the form of simple text labels, numerical ranges, or structured instructions to guide the subsequent work of the intelligent agents.

[0034] Therefore, by processing the corresponding drawing theme parameters for each agent, the feature configuration information for each agent is obtained. Specifically, each agent is configured to have the ability to process theme parameters for its specific domain. For example, after receiving theme parameters such as "tall" and "futuristic style," the "city building agent" can generate feature configuration information about the shape, material, color, and layout of buildings based on its internal knowledge base or algorithms. This information can be specific numerical values, descriptive text, or references to specific resource files. Similarly, the "transportation agent" will generate feature configuration information about the appearance, power system characteristics, and quantity of aircraft based on parameters such as "aircraft" and "science fiction feel." Each agent independently completes the feature configuration for its domain, ensuring professionalism and parallelism.

[0035] Subsequently, the prompt word generation module processes the feature configuration information corresponding to each agent to obtain global drawing prompt words. For example, when the "city building agent" outputs feature configuration information about buildings, and the "transportation agent" outputs feature configuration information about aircraft, the prompt word generation module receives these scattered and potentially varied pieces of information. This module is configured to integrate, translate, and optimize this information to form a coherent text prompt word that conforms to the input specifications of the image generation model. For example, descriptions such as "towering futuristic buildings" and "sci-fi-looking aircraft" can be combined into a complete prompt word, such as "an image depicting a future city nightscape, including towering futuristic buildings and sci-fi-looking aircraft soaring through the air." This process aims to transform specialized feature configurations into unified instructions that can be understood by the image generation model.

[0036] Finally, the global drawing prompts are input into the image generation model for processing to obtain at least one image of the target object. The image generation model is configured to receive the generated global drawing prompts and, based on the detailed descriptions of the prompts, synthesize the corresponding visual images using its internal generation algorithm and training data. For example, when receiving the prompt "an image depicting a futuristic city nightscape, including towering futuristic buildings and sci-fi-style aircraft soaring through the air," the image generation model will generate one or more images that match these descriptions based on its understanding of concepts such as "futuristic city," "nightscape," "towering buildings," "aircraft," and "sci-fi." These images are the target object images that this method aims to output.

[0037] For example, suppose user A needs to generate an image of a game character. First, the system obtains the drawing object information submitted by user A, such as the character name A, the background being in Dynasty B, and the character's identity being a general. Based on this drawing object information, the system determines a list of intelligent agents that need to work together, including, for example, "composition agent," "shape agent," "scene agent," "lighting agent," and "emotion agent." Simultaneously, it generates corresponding drawing theme parameters for each agent. Specifically, the lighting agent is associated with theme parameters such as key light, fill light, surface lighting, contour light, special effects light, skin light, and ambient light; the composition agent is associated with theme parameters such as perspective, eye level, subject position, depth, and dynamic direction; the shape agent is associated with theme parameters such as hairstyle, lip color, makeup, and upper body attire; the scene agent is associated with theme parameters such as main location, dynamic elements, fog, foreground structures, architecture, and decorative elements; and the emotion agent is associated with subject parameters such as core emotion, surface, middle, and bottom layers, relational boundaries, and audience reaction.

[0038] Subsequently, each agent processes its corresponding drawing theme parameters to obtain its own element configuration information. For example, in the composition agent, the perspective is: low angle, about 15 degrees; the line of sight is: high downward; the main subject's position is: at the lower left corner of the picture, the golden ratio point; another character is positioned slightly to the upper right of the center…; another example is the lighting agent: the main light is: side backlighting from the upper right of the picture, simulating sunset, illuminating the protagonist's profile; the fill light is: warm light reflected from the ground, supplementing the character's chest; the partial lighting is: the left side of the face is illuminated with a warm yellow hue, while the right side is introduced into shadow, enhancing the sense of tension… Furthermore, after receiving and processing the element configuration information from different agents, the prompt word generation module integrates this scattered and specialized information into a unified global drawing prompt word, such as: "A is an ancient general, wearing general's armor…" Finally, this global drawing prompt word is input into the image generation model for processing. Based on this detailed and professional prompt word, the image generation model generates one or more images of the target object that meet user A's initial needs, allowing the user to choose.

[0039] In summary, by using a list of intelligent agents to collaboratively process drawing topic parameters and generate global drawing prompts, multi-dimensional professional division of labor and automated closed loop are achieved. This effectively solves the problems of single prompt generation logic, high user operation complexity, and unstable image quality in existing technologies. It has the advantages of improving image generation quality, reducing user operation complexity, and enhancing the diversity of creative expression, thereby further lowering the threshold for use in business scenarios and accelerating the efficiency and effectiveness of image generation in business.

[0040] Furthermore, the lack of a clear implementation method for obtaining drawing object information makes it unsuitable for different users' input habits and the needs of different scenarios. Users are still required to submit drawing requests in a fixed format, placing high demands on their ability to organize these requests. This fails to meet the needs of ordinary users who can freely describe their needs in natural language, nor does it satisfy the needs of users with specific requirements who require fine-grained configuration of drawing information. The problem of high usage threshold and poor adaptability persists, making it impossible to flexibly and accurately obtain users' drawing requirements. Therefore, in this embodiment, obtaining drawing object information includes: In response to a natural language editing request, determine the drawing object information; or, in response to a trigger operation on the information configuration interface in the information configuration interface, determine the object editing information and generate drawing object information based on the object editing information.

[0041] Specifically, responding to a natural language editing request refers to the system receiving and understanding the user's drawing needs expressed in unstructured, colloquial, or written natural language. This can be achieved by parsing the user's input text using natural language processing (NLP) models, such as large language models (LLMs) or specially trained intent recognition models, to extract key entities, attributes, and relationships. Alternatively, speech recognition technology can be used to convert the user's voice input into text before performing natural language processing, or a combination of keyword matching and semantic analysis can be used to identify the core drawing intent and related descriptions in the user's request.

[0042] Correspondingly, determining the drawing object information refers to extracting, constructing, or generating structured data from user input for subsequent image generation processes. This data describes the user's desired image content, style, composition, and other elements. When responding to natural language editing requests, pre-defined ontology or knowledge graphs can be used to map parsed natural language elements to predefined drawing parameter fields, thus forming structured drawing object information. When responding to information configuration interface operations, the data items selected or entered by the user in the interface can be directly aggregated to form drawing object information containing all configuration parameters. The information configuration interface refers to a visual interactive interface provided to users for structured, parameterized input. This can be a graphical user interface (GUI) containing various input boxes, drop-down menus, sliders, checkboxes, and other controls for users to select or input specific drawing parameters. It can also be a web-based form interface where users configure various details of the drawing by filling out a form. The information configuration interface refers to the interactive elements in the information configuration interface used to receive user input or operations. This can be any control on the interface that users can click, input, or select, such as buttons, text input boxes, drop-down lists, and sliders. It can also be an interactive area that supports gestures, touch, or voice commands, where users trigger information input through specific actions. A trigger action refers to the user's interaction with the information configuration interface, resulting in the system receiving configuration information. This could be the user clicking the "Confirm" button on the interface, selecting an option from a drop-down menu, dragging a slider, entering content in a text box and pressing Enter, etc. Alternatively, the user could activate a configuration item through keyboard shortcuts, voice commands, or gestures.

[0043] Correspondingly, determining object editing information refers to the raw data collected by the system after the user operates through the information configuration interface, reflecting the user's editing or configuration of the drawing object. This can be directly read from the current values ​​of various controls in the information configuration interface, such as text in text boxes, selected items in drop-down menus, and values ​​of sliders. It can also be a log record or state snapshot of a series of operations performed by the user on the interface. Generating drawing object information based on the object editing information means processing, verifying, transforming, or integrating the raw object editing information collected from the information configuration interface to form the final structured drawing object information used in the image generation process. This can be done by converting the format of the raw object editing information to conform to the requirements of the internal data model, such as converting the color name entered by the user to RGB values. It can also be done by verifying the raw information to ensure that it conforms to business rules and data specifications, and performing necessary default value filling or missing information completion.

[0044] Based on this, the first path is "responding to natural language editing requests and determining drawing object information." When users express their drawing intentions in natural language, the system can recognize and parse this unstructured input. Through advanced natural language processing technology, the system extracts key drawing elements from the user's description and transforms this information into structured drawing object information. This approach greatly lowers the barrier to entry for ordinary users; they do not need to learn complex prompt vocabulary or professional terminology, but can initiate drawing requests simply by using everyday language. The second path is "responding to triggering operations on information configuration interfaces in the information configuration interface, determining object editing information, and generating drawing object information based on the object editing information." For users with specific and refined needs for drawing details, the system provides a structured information configuration interface. This interface contains a series of interactive information configuration interfaces, which users can trigger to precisely set various drawing parameters item by item. The system collects all user operations and inputs on the interface to form raw object editing information, then processes, verifies, and integrates this raw information to finally generate well-organized and accurate drawing object information. This approach meets the needs of professional users for precise control and refined configuration. These two paths are not mutually exclusive, but rather work together to ensure the comprehensiveness and accuracy of the drawing object information. Regardless of which method the user chooses, they will ultimately obtain drawing object information in a unified format. This drawing object information is then used in subsequent image generation processes. For example, in the image generation method described above, this information will be used to determine the list of agents and generate the drawing topic parameters corresponding to each agent. In this way, the solution of this application can flexibly obtain the user's drawing requirements, laying a solid foundation for subsequent agent processing and image generation.

[0045] For example, suppose a user needs to generate an image of "a cat in a spacesuit playing on the moon." As a specific implementation, the user can input their drawing requirements in two ways. Method 1: Natural Language Input. The user can directly enter in a text input box: "Please draw me a cat in a spacesuit playing on the moon's surface, with Earth and stars in the background, and a science fiction style." After receiving this natural language editing request, the system will send it to a natural language processing module based on a Large Language Model (LLM). This module performs semantic analysis on the text, identifying the core entities "cat," "spacesuit," "moon," "Earth," and "stars," as well as the action "playing" and the style "science fiction." Subsequently, the module structures this information into drawing object information. Method 2: Information Configuration Interface Input. Users can also operate through a graphical information configuration interface. The interface may contain the following configuration options: a "Subject" dropdown menu where the user selects "Animal" and then "Cat"; a "Clothing" input box where the user enters "Spacesuit"; a "Scene" dropdown menu where the user selects "Space" and then "Lunar Surface"; a "Background Elements" checkbox where the user selects "Earth" and "Starry Sky"; an "Action" input box where the user enters "Play"; and a "Style" slider where the user adjusts it to "Science Fiction". When the user completes all configurations and clicks the "Generate" button on the interface (i.e., triggers the operation), the system collects the current values ​​of these interface controls to form object editing information. For example, the system will read `{"Subject Type": "Animal", "Subject Specific": "Cat", "Clothing": "Spacesuit", "Scene": "Lunar Surface", "Background Elements": ["Earth", "Starry Sky"], "Action": "Play", "Style": "Science Fiction"}`. Subsequently, an information processing module will convert this object editing information into unified drawing object information so that it can be used for subsequent image generation and processing operations.

[0046] In summary, the above processing effectively addresses the issues of high barriers to entry and poor adaptability in traditional drawing tools when acquiring user drawing requirements. This allows the system to broadly adapt to the input habits of different users and the needs of different scenarios, significantly improving the usability, flexibility, and user experience of the entire image generation process, and providing high-quality input data for subsequent agent processing and image generation. Furthermore, relying solely on automatically matched agents cannot adapt to users' personalized customization needs. Users cannot flexibly adjust the agents involved in drawing according to actual drawing requirements. Automatic matching may miss necessary specialized agents, while unnecessary agents may be retained, slowing down the generation process. This fails to balance the dual needs of automated matching and personalized customization, and still suffers from poor scalability and an inability to flexibly adapt to different drawing requirements. Therefore, in this embodiment, determining the agent list based on the drawing object information includes: Based on the drawing object information, at least one first agent is selected from the candidate agent set; based on the drawing object information, an agent editing page is displayed, and in response to an agent editing request submitted through the agent editing page, at least one second agent is determined; an agent list is generated based on the at least one first agent and / or the at least one second agent.

[0047] Specifically, drawing object information refers to structured or unstructured data describing the user's drawing intentions or needs. This can be a natural language description, such as the user-inputted text "a fox running in the forest, watercolor style," which details the image's theme, subject, scene, and artistic style. Alternatively, it can be structured data, such as a JSON object containing multiple fields (e.g., theme, style, composition, color preference), each corresponding to a specific drawing element. The candidate agent set refers to a pre-built database or knowledge base containing various specialized agents. Each agent in this set possesses specific professional capabilities or domain knowledge; for example, it could include an "animal agent" specializing in animal images, a "watercolor style agent" proficient in a specific artistic style, or a "composition agent" responsible for scene composition. These agents can be tagged or categorized to facilitate system retrieval and matching. The first agent refers to the agent automatically selected by the system from the candidate agent set based on the drawing object information. For example, the system can extract keywords or perform semantic analysis on the information of the drawing objects and match them with the professional tags or descriptions of the agents in the candidate agent set, thereby automatically identifying and selecting agents that are initially related to the user's needs.

[0048] Correspondingly, the agent editing page refers to a graphical user interface (GUI) that provides users with the ability to manage and adjust agents. This page can display a list of currently selected agents and offer interactive functions such as searching, adding, deleting, and replacing agents. For example, this page could be a webpage where users can adjust the agent list by clicking buttons or dragging; or it could be a desktop application interface providing richer interactive controls. An agent editing request refers to an operation command issued by the user through the agent editing page to modify the current agent list. For example, when a user clicks the "Add" button on the agent editing page and selects an agent that has not been automatically matched, an edit request to add an agent is generated. Similarly, a user clicking the "Delete" button to remove an agent or modifying the configuration parameters of an agent also constitutes an agent editing request. Second agents refer to agents that the user manually determines or adjusts through agent editing requests. These agents may be manually selected and added to the list by the user from the candidate agent set, or they may be formed after the user modifies or replaces the first agent automatically selected by the system. The introduction of second agents aims to meet users' needs for personalized and refined control over the agent list. The agent list refers to the final set of agents determined and used in the subsequent image generation process. This list is the result of integrating the first and / or second agents, and it contains the configuration information of all agents that will participate in the image generation task and their possible associations. For example, this list could be a data structure containing unique identifiers for each agent and their corresponding functional descriptions, for use by subsequent modules.

[0049] Based on this, the system will perform preliminary screening from a pre-set set of candidate agents according to the drawing object information provided by the user, automatically selecting at least one primary agent. This step leverages the advantages of automated matching, quickly and efficiently providing the user with a basic agent combination, thereby reducing the time cost and operational burden for the user to select agents from scratch. Building on this, to cater to the user's personalized needs, the system will display an agent editing page based on the drawing object information. This page allows users to intuitively view the currently selected agents and adjust the agent list by submitting agent editing requests according to their specific drawing intentions or expectations for the image effect. For example, users can add specialized agents that the system did not automatically match but that they deem necessary, or delete agents that the system automatically matched but that they deem unnecessary, or even replace certain agents for more precise control.

[0050] Furthermore, upon responding to the user's editing request, the system identifies at least one second agent, representing the user's personalized intervention and adjustments to the agent list. Finally, the system generates a final agent list based on at least one automatically selected first agent and / or at least one manually determined second agent. This final agent list includes both the core agents automatically identified by the system based on drawing object information and personalized adjustments made by the user based on their experience and needs. This combination of automation and manual intervention ensures the rationality, efficiency, and adaptability of the agent list to the user's personalized needs. Combined with basic image generation methods, this approach significantly improves the efficiency and quality of the overall image generation process. A list of agents that has undergone automated initial matching and fine-tuning by the user more accurately reflects the user's drawing intent. This means that each agent can obtain more precise element configuration information when processing corresponding drawing theme parameters, enabling the prompt generation module to construct more accurate global drawing prompts. Ultimately, after receiving these high-quality prompts, the image generation model is able to generate target object images that are more in line with user needs, more creative, and more professional, effectively avoiding problems such as unstable image quality and content homogenization caused by improper agent selection.

[0051] In summary, the above processing effectively solves the problems of poor scalability and inflexibility in adapting to different drawing needs in traditional image generation methods. This mechanism gives users a high degree of customization capability, ensuring that the agent list accurately reflects the user's personalized needs and avoiding omissions or inaccuracies that may occur with automatic matching.

[0052] In practical implementation, existing image generation methods, when determining the list of agents and generating drawing theme parameters based on the drawing object information, are prone to low processing efficiency, mismatch between agent selection and drawing requirements, and mismatch between parameters and agent responsibilities if manual selection and non-standardized methods are used. This affects the rationality and automation of multi-agent collaboration, ultimately reducing the quality of the generated image. Therefore, in this embodiment, determining the list of agents based on the drawing object information and generating the drawing theme parameters corresponding to each agent in the list includes: The drawing object information is input to the agent scheduling module for processing to obtain an agent list and drawing topic parameters corresponding to each agent in the agent list; wherein, the drawing topic parameters corresponding to each agent include at least one sub-topic parameter, and the drawing topic parameters corresponding to each agent are associated with the feature configuration dimension corresponding to the agent.

[0053] Specifically, the agent scheduling module is a component dedicated to automating agent selection and drawing topic parameter generation. This module can be built based on a pre-defined rule engine, selecting a suitable agent by matching keywords or features in the drawing object information with the agent's capability labels; alternatively, it can be implemented based on a machine learning model, such as training a large language model to understand drawing object information and automatically identifying the required agent and its corresponding drawing topic parameters based on learned knowledge. Sub-topic parameters are a finer-grained division of drawing topic parameters, designed to guide the agent in processing specific aspects of the drawing content. For example, a "character topic parameter" can include "pose sub-topic parameters" and "clothing sub-topic parameters," thereby achieving more refined control over the character's appearance. These sub-topic parameters can be nested within the drawing topic parameters or exist as independent parameter items logically belonging to a main topic. Element configuration dimensions refer to the specific aspects or attributes that the agent is responsible for processing during image generation. These dimensions can be predefined categories, such as "character pose," "background environment," "lighting effects," and "art style," or they can be the processing range determined by the agent's own capability set.

[0054] Based on this, by introducing an agent scheduling module, the process of generating drawing object information, agent lists, and drawing theme parameters is automated and standardized. When the system receives drawing object information, this information is uniformly input into the agent scheduling module. The agent scheduling module no longer relies on manual intervention but uses its internal intelligent logic (e.g., based on preset rules or machine learning models) to deeply analyze the drawing object information, thereby accurately identifying the various agents needed to complete the drawing task and constructing an agent list. Simultaneously, the agent scheduling module generates exclusive drawing theme parameters for each agent in the agent list based on the specific content of the drawing object information. These drawing theme parameters not only include overall guidance but are further refined into at least one sub-theme parameter, allowing each agent to focus on processing more specific and refined drawing content. More importantly, the drawing theme parameters corresponding to each agent are explicitly associated with the element configuration dimension that the agent is responsible for, ensuring that the parameters obtained by the agent precisely match its processing capabilities and scope of work. This mechanism avoids problems of parameter misalignment or chaotic division of labor, ensuring that the element configuration information output by each agent meets its responsibilities. In this way, the solution transforms the selection of agents and parameter generation process from non-standardized and inefficient manual operation to automated and standardized intelligent scheduling, which greatly improves the efficiency and accuracy of the entire image generation process and lays a solid foundation for efficient collaboration of agents and generation of high-quality images.

[0055] For example, suppose a user wants to generate an image of "a cat in a spacesuit playing guitar on the moon in a cyberpunk style." The agent scheduling module performs semantic analysis on the corresponding drawing object information, identifying core concepts such as "cat," "spacesuit," "moon," "guitar," and "cyberpunk." Based on these concepts, the agent scheduling module automatically selects from a preset set of agents: a "character agent" (responsible for the cat and spacesuit design), a "prop agent" (responsible for the detailed depiction of the guitar), a "background agent" (responsible for constructing the lunar environment), and a "style agent" (responsible for rendering the overall cyberpunk style), thus forming an agent list. Furthermore, the agent scheduling module generates corresponding drawing theme parameters for each agent in the agent list. For example, for the "character agent," its drawing theme parameters might include sub-theme parameters such as "cat breed: Ragdoll," "spacesuit color: silver," and "posture: sitting." The element configuration dimensions associated with these parameters are "character appearance" and "character action." For the "prop agent," its drawing theme parameters might include sub-theme parameters such as "guitar type: electric guitar" and "guitar color: neon blue," with the associated element configuration dimension being "prop details." For the "background agent," its drawing theme parameters might include sub-theme parameters such as "lunar surface features: craters, thin atmosphere" and "lighting: Earth reflection," with the associated element configuration dimension being "environment scene." For the "style agent," its drawing theme parameters might include sub-theme parameters such as "overall color tone: dark tone, neon lights" and "material: metal, plastic," with the associated element configuration dimension being "art style." In this way, each agent receives detailed drawing guidance precisely matched to its role, laying the foundation for subsequent element configuration.

[0056] In summary, the above-described approach effectively solves the problems of inefficiency, mismatch between agent selection and drawing requirements, and mismatch between parameters and agent responsibilities caused by traditional manual selection and allocation of agents or non-standardized generation of drawing topic parameters. The introduction of the agent scheduling module automates and standardizes the process of determining the agent list and generating drawing topic parameters, significantly improving processing efficiency and reducing the user's workload.

[0057] In practical applications, the execution progress of multiple agents generating feature configuration information is inconsistent. This can easily lead to situations where some agents prematurely read incomplete feature information and generate prompts before completing feature configuration generation. Furthermore, the lack of a unified and standardized storage management method makes it easy for feature information generated by different agents to be mistakenly modified or overwritten. This compromises the completeness and accuracy of the feature configuration information input to the prompt generation module, resulting in errors in the generated global drawing prompts and affecting the quality of the final image. Therefore, in this embodiment, before the step of processing the feature configuration information corresponding to each agent through the prompt generation module to obtain the global drawing prompts, the following steps are also included: The feature configuration information corresponding to each agent is written into a shared field, wherein the shared field is managed by a field-level protection strategy; wherein the step of processing the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words includes: when the task orchestrator determines that each agent is in the execution completion state, reading the feature configuration information corresponding to each agent from the shared field; and inputting the feature configuration information corresponding to each agent into the prompt word generation module for processing to obtain global drawing prompt words.

[0058] Specifically, a shared field refers to a data storage area accessible and writable by multiple agents. This can be a memory area, a database table, a distributed caching system, or a specific file or directory in a file system, used for temporary storage or persistent storage of feature configuration information generated by agents. This shared field provides a unified and standardized storage location for feature information generated by all agents, facilitating subsequent unified retrieval and avoiding read chaos caused by scattered storage. Field-level protection strategies are data access control mechanisms used for fine-grained management of read and write permissions for specific data items in shared fields. They can be implemented based on security models such as access control lists, role-based access control, or attribute-based access control. This strategy restricts the modification permissions of different agents on feature information already written to other agents, preventing accidental modification or overwriting of already written feature information by different agents, ensuring the integrity and correctness of feature configuration information generated by each agent. A task orchestrator is a component responsible for coordinating and managing the execution flow of multiple agents or tasks. It can be a scheduler based on a workflow engine, a message queue consumer, or a centralized control service. This task orchestrator monitors the execution status of each agent, ensuring that all agents have completed the generation and writing of their corresponding feature configuration information. This guarantees that all necessary data is ready when reading feature configuration information. Determining each agent is in a completed state ensures that all agents have completed their respective tasks before reading feature configuration information from the shared field, avoiding subsequent processing errors due to incomplete data reading. The task orchestrator can receive completion signals or status updates from each agent. Once all expected agents have sent completion signals, all agents are considered to be in a completed state. Reading the feature configuration information corresponding to each agent from the shared field involves retrieving complete feature configuration information from a unified and protected shared storage after confirming that all agents have completed their tasks, providing accurate input for subsequent prompt generation. The task orchestrator or prompt generation module can directly access the shared field through a preset interface or protocol and extract the feature configuration information corresponding to each agent based on the agent identifier or other indexing mechanisms.

[0059] Based on this, by uniformly managing and verifying the storage and execution status of feature configuration information generated by multiple agents, the problems of incomplete and error-prone feature information under parallel processing of multiple agents are solved. This ensures the completeness and accuracy of feature configuration information in the input prompt generation module, providing a foundation for generating high-quality global drawing prompts. Before executing the step of obtaining global drawing prompts, the feature configuration information of each agent is written into a shared field, providing a unified and standardized storage location for the feature information generated by all agents. This facilitates subsequent unified reading and avoids reading chaos caused by scattered storage. Setting a field-level protection strategy for the shared field restricts the modification permissions of different agents on the feature information of other agents that have already been written, preventing different agents from accidentally modifying or overwriting the feature information that has already been written during the writing process, and ensuring the integrity and correctness of the feature configuration information generated by each agent. The feature configuration information in the shared field is only read after the task orchestrator determines that all agents are in the execution completion state. This can adapt to the different execution progress of multiple agents, ensuring that all agents have completed the generation and writing of their own feature configuration information, avoiding premature reading of incomplete feature information, and ensuring that the feature information used in subsequent processing covers the processing results of all agents. After confirming that all agents have completed their execution, the feature configuration information of each agent is read from the shared field and then input into the prompt word generation module to obtain global drawing prompt words. This allows the prompt word generation module to generate global drawing prompt words that meet the requirements based on the complete and accurate multi-agent division of labor processing results, thereby improving the quality of global drawing prompt words from the source and ensuring that the final generated target object image meets the requirements.

[0060] For example, after each agent generates its own feature configuration information, it can write it to a shared field. This shared field can be a JSON document in a distributed key-value store, where each agent's output corresponds to a specific key. To prevent data conflicts, the system configures a field-level protection strategy for this shared field. For example, agent A can only write to the "dragon_config" key, agent B can only write to the "knight_config" key, and agent C can only write to the "environment_config" key. Once written, other agents cannot modify it. The task orchestrator continuously monitors the execution status of agents A, B, and C. Only when all agents send a "complete" signal to the task orchestrator does the task orchestrator determine that all agents are in the execution completion state. At this time, the prompt word generation module is triggered, reading the complete "dragon_config," "knight_config," and "environment_config" information from the distributed key-value store. Based on this complete and accurate feature configuration information, the prompt word generation module can construct a high-quality global drawing prompt word for subsequent image generation.

[0061] In summary, the above-described approach effectively addresses the incompleteness of feature configuration information caused by inconsistent execution progress in multi-agent parallel processing, as well as the issues of error susceptibility and overwriting due to the lack of a unified storage and protection mechanism. By introducing shared fields and field-level protection strategies, the integrity and accuracy of feature configuration information generated by each agent are ensured, providing a reliable data foundation for subsequent prompt generation. Simultaneously, the task orchestrator's rigorous verification of agent execution status guarantees that the prompt generation module always processes the information only after all feature configuration information is ready, thereby significantly improving the quality and stability of global drawing prompts. Furthermore, the generated multiple target object images may contain content that does not meet compliance requirements. Existing technologies require users to manually filter results and make compliance judgments, which is not only fragmented and requires users to repeatedly operate at multiple stages, increasing user time costs and operational burdens, but also prone to the leakage of non-compliant images due to oversights in manual screening. This fails to meet the image compliance requirements of scenarios such as competitions and commercial applications, and also fails to form an automated, complete image generation loop. Therefore, in this embodiment, after the step of inputting the global drawing prompts into the image generation model for processing to obtain at least one target object image is executed, the method further includes: The at least one target object image is input into the compliance detection module for processing to obtain detection information corresponding to each target object image; the at least one target object image is filtered according to the detection information, and failed object images that do not meet the image usage conditions are deleted; the target image is determined based on the filtering results and deletion results.

[0062] Specifically, the at least one target object image is input into a compliance detection module for processing. This module is a software or hardware component specifically designed to identify whether image content complies with preset specifications or laws and regulations. This module can be built based on a deep learning model, for example, by training a convolutional neural network (CNN) or a Transformer model to identify sensitive content, copyright infringement, violence, pornography, and other illegal elements in the image. Furthermore, the compliance detection module can also combine a rule engine and image feature extraction technology, using preset rules such as keywords, image texture, and color distribution for matching and judgment to ensure that the image content complies with specific compliance standards.

[0063] Accordingly, after processing by the compliance detection module, detection information is obtained for each target object image. This detection information is data on the compliance status of the image output by the compliance detection module after analyzing the image. This detection information may include a compliance score (e.g., 0 to 100 points) to quantify the compliance level of the image, as well as the specific type of violation (e.g., political, violent, pornographic, copyright infringement, etc.) and the corresponding confidence level. Alternatively, the detection information may also contain only a Boolean value (compliant / non-compliant), along with detailed markings of the violation area or a description of the violation content, to facilitate subsequent review and processing.

[0064] Accordingly, the at least one target object image is filtered based on the detection information, and failed object images that do not meet the image usage conditions are deleted. This step aims to automatically filter out images that do not meet the requirements based on the compliance detection results, ensuring the quality and compliance of the final output image. For example, a compliance threshold can be set, and all images with a compliance score below the threshold will be marked as failed object images and automatically deleted. Alternatively, if the detection information contains a specific type of serious violation (such as explicit violence or pornography), the image will be directly deleted regardless of its score.

[0065] Accordingly, the target image is determined based on the filtering and deletion results. This step selects the final, compliant image as the output from the images that have passed the compliance screening. One approach is to directly use all images that were not deleted as the final target image. Another approach is to further filter the remaining images based on other preset criteria (e.g., aesthetic scores, user preferences, diversity, etc.) to determine the final target image presented to the user.

[0066] Based on this, after the image generation model initially obtains at least one target object image, an automated compliance detection and screening process is introduced, solving the problems of low efficiency and poor reliability of manual compliance screening. Specifically, after the image generation model generates preliminary images based on global drawing prompts, these images are not directly output but are sent to the compliance detection module for review one by one. The compliance detection module performs in-depth analysis on each image and outputs detailed detection information, which accurately reflects the compliance status of each image. Based on this detection information, the system can automatically identify and filter out failed object images that do not meet the preset image usage conditions, such as deleting images containing sensitive content or copyright issues. In this way, this solution ensures that all final output images meet compliance requirements, thus forming an automated closed loop from image generation to compliance verification. This mechanism not only avoids the omissions and high time costs that may result from manual screening but also ensures the quality and reliability of image output, making it particularly suitable for competitions or commercial applications with extremely high compliance requirements.

[0067] For example, suppose a user requests the generation of an image on a specific topic using the method described above. First, the image generation model generates, for example, five candidate target object images based on global drawing prompts. These five images are then automatically input into a pre-trained compliance detection module. This module can be a deep learning model trained on a large-scale image dataset, capable of identifying sensitive elements, violent scenes, inappropriate symbols, or potential copyright infringement content in the images. For each input image, the compliance detection module outputs a report containing a compliance score (e.g., between 0 and 1, with 1 indicating full compliance) and any detected violation types (e.g., "no violation," "slightly sensitive," "severely violent"). The system presets a compliance threshold, such as 0.7. If an image's compliance score is below 0.7, or its violation type is marked as "copyright infringement," the image is automatically identified as a failed object image that does not meet the image usage requirements and is deleted. For example, if two of the five images are deleted due to low scores or containing severely violating content, the remaining three images are determined as the final target images, which can be further displayed to the user or used for subsequent processing.

[0068] In summary, the above processing achieves automated compliance detection and screening in the image generation process, significantly reducing the time cost and operational burden for users to manually screen and judge images after generation. This solution effectively avoids the risk of non-compliant images being leaked due to human oversight, thus enabling the stable output of images that meet high compliance requirements in scenarios such as competitions and commercial applications. Furthermore, by integrating compliance detection into the later stages of the image generation process, this solution forms a complete automated closed loop, improving overall work efficiency and the reliability of output images. Furthermore, the generated initial target object image may not meet the usage requirements. If an image that does not meet the requirements is directly output, the quality of the final image cannot be guaranteed. Moreover, existing solutions do not combine the configuration dimensions of the elements each agent is responsible for for targeted evaluation, nor do they form an automated closed loop for generation, evaluation, and iteration. Users still need to manually screen and adjust, which is inefficient and cannot guarantee that the generated image meets the corresponding usage requirements, easily resulting in substandard image quality that does not meet the needs of the usage scenario. In this embodiment, after the step of inputting the global drawing prompt words into the image generation model for processing to obtain at least one target object image is executed, the method further includes: The at least one target object image is input into the image evaluation module for processing to obtain an evaluation report for each target object image. The evaluation dimension of the image evaluation module is determined based on the feature configuration dimension corresponding to each agent. If the evaluation report determines that there is an optimized object image among the at least one target object images, the process returns to the step of processing the corresponding drawing theme parameters for each agent to obtain the feature configuration information for each agent; or, the process returns to the step of processing the feature configuration information for each agent through the prompt word generation module to obtain global drawing prompt words; until a target object image that meets the image usage conditions is obtained.

[0069] Specifically, the image evaluation module is a component responsible for analyzing and evaluating generated images. This module can be implemented based on machine learning models, such as training a deep learning model to identify specific elements, styles, compositions, and color harmony in an image, and output quantitative evaluation scores or classification results. Alternatively, the image evaluation module can be implemented based on a pre-defined rule engine or expert system, defining a series of evaluation rules and thresholds, such as checking whether the image contains specific prohibited content, conforms to specific aesthetic standards (e.g., the golden ratio composition), or meets specific scenario requirements (e.g., the purity of the background in a product image). The evaluation report is the structured result output by the image evaluation module, detailing the image's evaluation status. This report can use structured data formats such as JSON or XML, including fields such as evaluation score, list of non-compliance items, suggested optimization directions, and confidence level. It can also be a natural language text description, summarizing the image's strengths and weaknesses and providing modification suggestions, or including visual information such as image heatmaps and defect area annotations. The evaluation dimensions of the image evaluation module are determined based on the element configuration dimensions corresponding to each agent, meaning that the evaluation criteria are dynamic and highly correlated. This can be achieved through a dimension mapping mechanism, whereby a mapping relationship is established between the dimensions of the agent and its responsible element configuration and the evaluation dimensions of the image evaluation module during system initialization or agent configuration. For example, if an agent is responsible for the element configuration of "human pose," the evaluation module will include evaluation dimensions such as "naturalness of human pose" and "reasonableness of human action." When the evaluation report determines that there is an image to be optimized, the system can return to the steps of processing the corresponding drawing topic parameters for each agent to obtain the element configuration information for each agent. This is usually achieved through state machine backtracking or parameter adjustment and retry mechanisms. Based on the problems pointed out in the evaluation report, the system automatically or semi-automatically adjusts the drawing topic parameters of the relevant agents and then restarts the generation process of element configuration information.

[0070] Furthermore, the system can also return to the step of processing the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words. This can be achieved through a prompt word optimization algorithm. Based on the evaluation report, the prompt word generation module can use different strategies to recombine, refine, or expand the feature configuration information to generate more accurate and guiding global drawing prompt words. The entire process will continue until a target object image that meets the image usage conditions is obtained. This is usually achieved through a loop judgment mechanism. After each iteration, the evaluation report is checked. If all evaluation dimensions meet the preset qualification standards or thresholds, the loop terminates and the image is output.

[0071] Based on this, the aforementioned problems are effectively solved by introducing an intelligent evaluation and iterative optimization closed loop after the image generation model outputs at least one target object image. Specifically, after generating the initial target object image, the system does not output it directly but inputs it into the image evaluation module for processing. The unique feature of this image evaluation module is that its evaluation dimensions are determined based on the feature configuration dimensions corresponding to each agent. This means that the evaluation process can accurately check specific aspects of the image that are the responsibility of each agent. For example, if an agent is responsible for the color style of the image, the evaluation module will focus on checking the harmony and saturation of colors. This targeted evaluation can avoid the lack of evaluation dimensions and ensure the accuracy of the evaluation results, thereby accurately locating the parts of the image that do not meet the requirements. When the evaluation report indicates that there is an image that needs optimization, the system will intelligently choose to return to different stages in the generation process for iteration based on the nature of the problem. If the problem lies in the basic drawing feature configuration, for example, if the pose of a character generated by a certain agent is unnatural or the scene layout is unreasonable, the system will return to the step of "using each agent to process the corresponding drawing theme parameters to obtain the feature configuration information corresponding to each agent" to adjust the lower-level parameters. If the problem primarily lies in the inaccurate conversion of feature configuration information to global drawing prompts—for example, if the prompts fail to adequately express the details of the agent's configuration—the system will return to the step of "processing the feature configuration information corresponding to each agent through the prompt generation module to obtain global drawing prompts," optimizing the arrangement and expression of the prompts. This hierarchical iterative mechanism balances the flexibility and efficiency of optimization, avoiding unnecessary repetitive calculations. The entire process continues until a target object image that meets the image usage conditions is obtained, thus forming a complete automated closed loop from generation to evaluation to optimization.

[0072] For example, suppose a user generates an image of a "cyberpunk-style futuristic city night scene, including flying cars and a bustling market." This initial target image can then be input into an image evaluation module. The evaluation dimensions of this module are determined based on the feature configuration dimensions previously handled by the agents. For instance, for the feature configuration dimension of the "object agent," the module checks the "consistency of flying car design" and "quantity"; for the feature configuration dimension of the "scene agent," it checks "market activity" and "neon light layout." Suppose the evaluation report indicates that the flying car design in the image is too ordinary, lacking cyberpunk characteristics, and the market doesn't appear bustling enough. Since flying car design is a core feature configuration issue, the system intelligently chooses to return to the step of "processing the corresponding drawing theme parameters for each agent to obtain the feature configuration information for each agent." At this point, the "object agent" might be instructed to reprocess its theme parameters, emphasizing a "unique, detailed flying car design," or the "scene agent" might be instructed to increase the "crowd density" parameter. The system will regenerate the element configuration information based on these adjustments, then generate new global drawing prompts, and input them into the image generation model again. This evaluation-iteration process will continue until the generated image meets the preset image usage conditions in all evaluation dimensions such as "flying car design" and "market activity," for example, meeting specific aesthetic standards or conforming to the requirements of game concept art.

[0073] In summary, the above processing establishes an automated closed-loop evaluation process for the generated image. This process can specifically evaluate image quality based on the element configuration dimensions corresponding to each agent and automatically trigger iterative optimization at the corresponding level, ultimately obtaining the target image that meets the requirements. This not only improves the quality of the final output image but also achieves full-process automation, significantly reducing the user's operational costs.

[0074] In practical implementation, existing image generation methods do not perform compliance and rationality checks on global drawing prompts before generating images. If the global drawing prompts themselves contain illegal content or content that does not conform to usage specifications, directly inputting them into the image generation model will not only generate images that do not meet the requirements, causing unnecessary waste of computing power, but also bring compliance risks. This fails to meet the requirements for image content compliance in scenarios such as competitions and commercial applications, and also reduces the generation efficiency of compliant target images, failing to solve the drawbacks of non-compliant content and unstable quality in existing technologies. Therefore, in this embodiment, the step of inputting the global drawing prompts into the image generation model for processing to obtain at least one target object image includes: The global drawing prompts are reviewed; if the review is passed, the global drawing prompts are input into an image generation model for processing to obtain at least one target object image; if the review is not passed, the global drawing prompts are optimized to obtain target drawing prompts, which are then input into an image generation model for processing to obtain at least one target object image.

[0075] Specifically, reviewing the global drawing prompts involves evaluating their content to ensure compliance with predefined standards, policies, or ethical guidelines. This review aims to identify and filter out any inappropriate, sensitive, or non-compliant elements to prevent their use in image generation. This review can be implemented in various ways. For example, it can be based on a predefined rule base for keyword matching and pattern recognition; when a prompt contains sensitive words or inappropriate descriptions, the system automatically marks it as failing. Alternatively, machine learning models can be used to train and identify the semantic content and potential risks of prompts. For instance, natural language processing (NLP) models can be used to classify prompts and determine whether they fall into a violation category. Human review can also be combined; prompts that are difficult for the intelligent system to judge or are marked as high-risk are submitted to human reviewers for final confirmation to ensure the comprehensiveness and accuracy of the review.

[0076] If the global drawing prompt is approved, it is input into the image generation model for processing to obtain at least one target object image. This means that when the global drawing prompt is determined to fully comply with all compliance standards, the system directly passes the prompt to the image generation model. This step ensures that only compliant input enters the image generation process, thereby avoiding unnecessary resource consumption and potential compliance risks. After receiving the approved global drawing prompt, the image generation model generates one or more target object images that match the prompt description based on its internal algorithm and training data.

[0077] If the global drawing prompt fails the review, it is optimized to obtain a target drawing prompt. This target prompt is then input into an image generation model for processing to generate at least one target object image. This means that when a global drawing prompt fails the review, the system does not directly reject it but rather corrects and improves it. The optimization process aims to remove or modify non-compliant parts of the prompt while preserving the user's original creative intent as much as possible. Optimization methods may include rule-based replacement or deletion, such as replacing sensitive words with neutral words or directly deleting descriptions that do not conform to the specifications; or rewriting the prompt using a Large Language Model (LLM) to maintain semantic consistency while meeting compliance requirements. The optimized prompt is called the target drawing prompt, which is then input into the image generation model to generate a compliant target object image.

[0078] Based on this, a pre-approval mechanism is introduced before inputting global drawing prompts into the image generation model to ensure their compliance. First, the system rigorously reviews the generated global drawing prompts. This review process aims to identify and filter out any content that may contain violations, inappropriate content, or content that does not conform to preset specifications. This pre-approval effectively prevents non-compliant prompts from directly entering the image generation stage, thereby reducing the risk of generating non-compliant images from the source and avoiding unnecessary waste of computational resources. During the review process, if a global drawing prompt is deemed acceptable, meaning its content fully complies with all preset compliance standards, it will be directly sent to the image generation model for processing. This direct processing method ensures efficient generation of compliant prompts and avoids unnecessary delays or additional processing steps. After receiving a globally approved global drawing prompt, the image generation model generates at least one target object image based on its content. However, if a global drawing prompt fails the review process, indicating that it contains some non-compliant content, the proposed solution does not simply reject the user's request but further optimizes it. This optimization process aims to correct non-compliant parts of the prompts while preserving the user's original creative intent as much as possible, thereby obtaining a compliant target drawing prompt. The optimized target drawing prompt is then input into an image generation model to generate the corresponding target object image. This differentiated processing mechanism not only ensures the compliance of the final generated image but also improves the user experience and drawing success rate by optimizing rather than directly rejecting it, making it particularly suitable for competitions or commercial scenarios with strict content compliance requirements.

[0079] For example, when reviewing global drawing prompts, a preliminary screening can be performed using a rule-based text analyzer. This analyzer has a built-in database containing sensitive words, prohibited phrases, and specific grammatical patterns. For instance, if a global drawing prompt contains explicitly prohibited words such as "A" or "B," or describes a scenario that does not conform to the guidelines, the prompt will be immediately marked as failing the review. If the preliminary screening does not find any obvious violations, a pre-trained deep learning model can be used for semantic analysis. This model can identify the potential meaning and context of the prompt and determine whether there is any implicit risk of violation. For example, even if there are no directly sensitive words, if the word combination and context indicate that it implies inappropriate content, it will also be marked as failing. If the review is passed, for example, a global drawing prompt describing "a cute cat playing in the sun," because its content is positive and there are no violations, will be directly passed to an image generation model. This image generation model can be an AI model based on a diffusion model or a generative adversarial network, which will generate one or more images of a cat that match the description based on the prompt. If the review fails, for example, if a global drawing prompt contains the phrase "depicting a fight scene," the system will initiate an optimization process. The optimization module can be a rewriting tool based on a large language model. Upon receiving the non-compliant prompt, the large language model analyzes its core intent and rewrites it into a compliant expression, thereby generating a target drawing prompt for use in image generation.

[0080] In summary, the above-described process introduces a pre-emptive global drawing prompt review mechanism into the image generation workflow, effectively solving the problems of non-compliant prompts leading to unacceptable images, wasted computing power, and compliance risks inherent in traditional methods. This solution prevents non-compliant content from entering the generation stage at the source, significantly reducing the likelihood of generating illegal images.

[0081] In practical implementation, existing image generation methods, when selecting at least one first agent from a set of candidate agents based on the drawing object information to generate an agent list, rely solely on the original drawing object information to select agents, failing to extract the core drawing theme information. This easily leads to mismatched agent selections that do not meet the core drawing requirements. Therefore, in this embodiment, the step of selecting at least one first agent from the set of candidate agents based on the drawing object information includes: Determine drawing theme information based on the drawing object information; select at least one first agent from the candidate agent set based on the drawing theme information; wherein, inputting the global drawing prompt words into the image generation model for processing to obtain at least one target object image includes: updating the global drawing prompt words using the drawing theme information, inputting the updated global drawing prompt words into the image generation model for processing to obtain at least one target object image.

[0082] Specifically, determining the drawing theme information based on the drawing object information refers to identifying the core and refined drawing theme from the original drawing object information input by the user through analysis, refinement, and integration. For example, Natural Language Processing (NLP) techniques can be used to perform text analysis, keyword extraction, and theme modeling on the drawing object information to identify the main concepts and intentions. Alternatively, a pre-set rule engine or knowledge graph can be used to match the drawing object information with predefined topic tags to determine the drawing theme information. Another option is to input the drawing object information into a pre-trained large language model, which will then understand and summarize the information to output refined drawing theme information.

[0083] Accordingly, selecting at least one first agent from the candidate agent set based on the drawing topic information means, based on the extracted drawing topic information, selecting the most suitable agent from a pre-defined agent library containing multiple functions and professional fields to handle the drawing topic. For example, each candidate agent can be pre-defined with its preferred field or topic tag, and the system can match the drawing topic information with the agent's tag, selecting the agent with the highest matching degree. Alternatively, an agent capability matrix can be constructed, comparing the drawing topic information with the agent's capability requirements, and selecting the agent that can meet the topic requirements. Furthermore, a recommendation algorithm can be used to recommend the most suitable agent based on the drawing topic information and historical agent performance data.

[0084] Correspondingly, updating global drawing prompts using drawing theme information refers to introducing core drawing theme information again after the prompt generation module has already generated global drawing prompts. This allows for the correction and optimization of the prompts to ensure a high degree of consistency with the user's core intent. For example, drawing theme information can be used as additional input, and the generated global drawing prompts can be reconstructed, supplemented, or have their weights adjusted by the prompt generation module or another dedicated optimization module. Alternatively, template matching or keyword insertion can be used to embed core keywords or phrases from the drawing theme information into specific positions within the global drawing prompts to strengthen the theme expression. Furthermore, a large language model can be used to jointly analyze the global drawing prompts and drawing theme information to generate a more accurate and updated global drawing prompt that incorporates theme information.

[0085] Based on this, by introducing drawing theme information, the system enhances both agent selection and the generation and optimization of global drawing prompts in the image generation process. First, the system intelligently extracts core drawing theme information based on the drawing object information provided by the user. This theme information, as a refined expression of the user's intent, can more accurately guide the subsequent agent selection process. In the candidate agent set, the system no longer relies solely on raw, potentially ambiguous drawing object information, but rather matches and selects at least one primary agent based on clear drawing theme information. This ensures that the selected agent possesses professional capabilities highly relevant to the user's core needs, thereby enabling it to more effectively process the corresponding drawing theme parameters and generate high-quality feature configuration information. Second, after the initial generation of global drawing prompts, this solution updates the prompts again using the drawing theme information. This update process corrects potential deviations from core needs during prompt generation, strengthens the expression of core drawing requirements, and ensures that the final global drawing prompts more accurately and clearly convey the user's drawing intent. Finally, the globally updated drawing prompts, updated with thematic information, are input into the image generation model for processing. This allows the model to more accurately understand user needs, thereby generating higher-quality images of the target object that better meet user expectations. In this way, the proposed solution achieves precise transmission and reinforcement of user intent throughout the entire image generation chain, improving the accuracy and quality of generated images from the source and key stages.

[0086] In summary, the above processing can accurately extract the core drawing theme from the original drawing object information, effectively avoiding the blind selection of the agent and ensuring a high degree of matching between the selected agent and the user's core needs, thereby improving the efficiency and accuracy of the agent in processing drawing theme parameters.

[0087] The following is in conjunction with the appendix Figure 3Taking the application of the image generation method provided in this application to character generation as an example, the image generation method will be further explained. Among other things, Figure 3 This application provides an embodiment of an image generation method for character generation scenarios, which includes the following steps: Step S302: In response to the natural language editing request, determine the drawing object information.

[0088] The system receives drawing object information submitted by the user through the character drawing page, including the character name A, background dynasty B, and identity as a general. During this process, a unique task ID can be generated, and the shared context, event bus, and task status can be initialized to complete the preparation of the process execution environment before the character to be drawn is drawn.

[0089] Furthermore, a large language model associated with a specific domain can be pre-trained. During the usage phase, users only need to provide the role name, and the large language model can automatically generate information such as representative works and key events associated with the role name, thereby improving image generation speed. After determining the relevant information associated with the specified role, a theme suitable for representing the role image can be generated based on this information, so that subsequent role images that meet business usage requirements can be generated around this theme.

[0090] Step S304: Input the drawing object information into the agent scheduling module for processing to obtain an agent list and the drawing topic parameters corresponding to each agent in the agent list.

[0091] After obtaining the drawing object information, this information can be input into the agent scheduling module (narrative expert) for processing. At this point, the module outputs a list of agents and the corresponding drawing theme parameters for each agent. For example, the agent list includes "Composition Agent," "Styling Agent," "Scene Agent," "Light and Shadow Agent," and "Emotion Agent." The Light and Shadow Agent is associated with theme parameters such as key light, fill light, partial surface lighting, contour light, special effects light, skin light, and ambient light. The Composition Agent is associated with theme parameters such as perspective, eye level, subject position, depth, and dynamic direction. The Styling Agent is associated with theme parameters such as hairstyle, lip color, makeup, and upper body styling. The Scene Agent is associated with theme parameters such as main location, dynamic elements, fog, foreground structures, architecture, and decorative elements. The Emotion Agent is associated with main emotions, surface, middle, and bottom layers, relational boundaries, and audience reactions.

[0092] In other words, when the narrative expert processes the user-input drawing object information, it is actually determining the intermediate agents to be invoked, clarifying the execution scope of each agent, generating unique input parameters for each agent, and defining the design goals, core requirements, and constraints for each agent. After the narrative expert completes the agent and parameter determination, it writes the output to a dedicated field in the shared context and publishes a narrative completion event via the event bus. This enables subsequent generation of element configuration information through each agent to complete automated operations.

[0093] Step S306: Use each agent to process the corresponding drawing topic parameters to obtain the feature configuration information corresponding to each agent, and write the feature configuration information corresponding to each agent into the shared field.

[0094] Step S308: If the task orchestrator determines that each agent is in the execution completed state, read the feature configuration information corresponding to each agent in the shared field.

[0095] After determining the drawing theme parameters for each agent, these parameters can be processed by the corresponding agent to obtain the element configuration information output by each agent. This element configuration information can also be written into a shared field for subsequent prompt generation. For example, in the composition agent, the perspective is: low-angle upward shot, approximately 15 degrees; the line of sight is: high downward view; the main subject's position is: located at the golden ratio point in the lower left corner of the image, with another character positioned slightly to the upper right of the center…; another example is the lighting agent: main light: side-backlighting from the upper right of the image, simulating sunset and illuminating the protagonist's profile; fill light: warm, diffused light reflected from the ground, supplementing the character's chest; partial lighting: the left side of the face is illuminated with a warm yellow hue, while the right side is introduced into shadow to enhance tension… By writing the above element configuration information into the shared field, it is easy for the subsequent prompt generation module to read and use it.

[0096] In other words, when the task orchestrator detects a narrative completion event, it can dynamically load the corresponding agents from the expert registration pool based on the agent list output by the narrative experts, and execute each loaded agent. During this process, each intermediate agent only needs to read its own parameters, requiring no other data source. Furthermore, after each agent completes its execution, it only needs to write its own output element configuration information to its own dedicated field, and cannot modify the output results of other agents. This ensures the safety of the written coroutines through a shared context, preventing data conflicts.

[0097] Step S310: Input the feature configuration information corresponding to each agent into the prompt word generation module for processing to obtain global drawing prompt words.

[0098] Step S312: Input the global drawing prompts into the image generation model for processing to obtain at least one target object image.

[0099] After reading the feature configuration information for each agent, this information can be input into the prompt word generation module for processing, thereby obtaining a global drawing prompt word covering all drawing intentions. For example, after receiving and processing feature configuration information from different agents, the prompt word generation module can integrate this scattered and specialized information into a unified global drawing prompt word, such as: "A An ancient general, wearing general's armor..." Finally, this global drawing prompt word is input into the image generation model for processing. Based on this detailed and specialized prompt word, the image generation model generates one or more images of the target object that meet user A's initial needs, allowing the user to choose.

[0100] Furthermore, to prevent the global drawing prompts from outputting character images that do not meet business scenarios or review rules after being input into the image generation model, the global drawing prompts can be optimized through review. For example, if the global drawing prompts contain keywords that provide negative guidance, these can be optimized by replacing or deleting them after review rules are established, resulting in global drawing prompts that meet the review rules for subsequent image generation. In practice, the final generated global drawing prompts can also be dynamically generated based on a prompt tiering strategy. For instance, prompts suitable for images suitable for adults and prompts suitable for images suitable for children can be categorized using a prompt tiering strategy, and corresponding global drawing prompts can be generated according to requirements to ensure that the final generated images are both compliant and match the business scenario, thereby improving the user viewing experience.

[0101] In other words, after verifying the completion of all intermediate agents' execution, the task orchestrator can publish all parallel execution completion events via the event bus, thereby triggering the execution of the tail node prompt word expert. When processing, the prompt word expert's sole input source is the sum of all agent outputs; it can disregard any output from the narrative expert, achieving complete decoupling from the head node. By dynamically traversing the outputs of all intermediate agents, the prompt word expert can automatically synthesize complete, standardized, and highly available drawing prompt words according to the Midjourney prompt word engineering specifications. After execution, the prompt word expert can write the final prompt word into a dedicated field in the shared context for subsequent image generation. During image generation, a large language model for image generation can be used to complete the drawing operation. By inputting the prompt words into the large model, the Playwright browser's automated tools can complete the entire process of prompt word submission, 4-grid image generation, and original image download for platforms such as Midjourney / DELL·E / NonoBanana / Qwen / Jimeng, resulting in character images that meet user needs.

[0102] Step S314: Input at least one target object image into the compliance detection module for processing to obtain detection information corresponding to each target object image.

[0103] Step S316: Filter at least one target object image based on the detection information, and delete failed object images that do not meet the image usage conditions, and determine the target image based on the filtering results and deletion results.

[0104] Furthermore, after obtaining the character images, considering that some images may be non-compliant or not meet the generation requirements, a compliance detection module can be used to perform compliance checks on each image to avoid inconvenience for downstream business operations. This allows for filtering and deleting unnecessary images based on the compliance detection results, thereby determining the target character image. For example, the image generation model might generate five candidate target object images based on global drawing prompts. Subsequently, these five images are automatically input into a pre-trained compliance detection module. This compliance detection module can be a deep learning model trained on a large-scale image dataset, capable of identifying sensitive elements, violent scenes, inappropriate symbols, or potential copyright infringement content in the images. For each input image, the compliance detection module outputs a detection report, which includes a compliance score (e.g., between 0 and 1, with 1 indicating full compliance) and any detected violation types (e.g., "no violation," "slightly sensitive," "severely violent"). The system presets a compliance threshold, such as 0.7. If an image's compliance score is below 0.7, or its violation type is marked as "copyright infringement," the image is automatically identified as a failed object image that does not meet the image usage conditions and is deleted. For example, if two out of five images are deleted due to low ratings or containing seriously prohibited content, the remaining three images are identified as the final target images, which can be further displayed to users or used for subsequent processing.

[0105] In other words, compliance experts can perform compliance verification and filter out failed images, outputting a list of valid images and the filtering results, which are then written into the shared context. Furthermore, if the final generated image does not meet the requirements, it can be scored across multiple dimensions, generating a scoring report and optimization suggestions. This allows for repeated image generation based on the optimization suggestions until a satisfactory target image is obtained for downstream business applications.

[0106] In summary, by using a list of intelligent agents to collaboratively process drawing topic parameters and generate global drawing prompts, multi-dimensional professional division of labor and automated closed loop are achieved. This effectively solves the problems of single prompt generation logic, high user operation complexity, and unstable image quality in existing technologies. It has the advantages of improving image generation quality, reducing user operation complexity, and enhancing the diversity of creative expression, thereby further lowering the threshold for use in business scenarios and accelerating the efficiency and effectiveness of image generation in business.

[0107] It should be noted that the image generation method provided in this manual can be applied to multiple industries or scenarios, such as virtual reality processing software, home entertainment product software, digital cultural product production software, digital cultural creative software, digital cultural creative design, education, news, cultural content industry software, digital publishing software, digital music development and production, and digital mobile multimedia development and production. In some cases, it can also be applied to fields such as animation and game production engine software and development systems, game and animation software, animation and game digital content services, digital film and television development and production, and digital performance development and production.

[0108] Corresponding to the above method embodiments, this application also provides an image generation system embodiment. Figure 4 A schematic diagram of the structure of an image generation system according to an embodiment of this application is shown. For example... Figure 4 As shown, the image generation system 400 includes a client 410 and a server 420, comprising: The client 410 is configured to determine drawing object information in response to a drawing request submitted by a user, and send the drawing object information to the server. The server 420 is configured to determine a list of agents based on the drawing object information, and generate drawing theme parameters corresponding to each agent in the list; process the corresponding drawing theme parameters using each agent to obtain feature configuration information corresponding to each agent; process the feature configuration information corresponding to each agent through a prompt word generation module to obtain global drawing prompt words; input the global drawing prompt words into an image generation model for processing to obtain at least one target object image, and send the target object image to the client.

[0109] In one optional embodiment, obtaining the drawing object information includes: determining the drawing object information in response to a natural language editing request; or, determining the object editing information in response to a trigger operation on the information configuration interface in the information configuration interface, and generating the drawing object information based on the object editing information.

[0110] In one optional embodiment, determining the agent list based on the drawing object information includes: selecting at least one first agent from a set of candidate agents based on the drawing object information; displaying an agent editing page based on the drawing object information; determining at least one second agent in response to an agent editing request submitted through the agent editing page; and generating an agent list based on the at least one first agent and / or the at least one second agent.

[0111] In an optional embodiment, determining the list of agents based on the drawing object information and generating drawing topic parameters corresponding to each agent in the list of agents includes: inputting the drawing object information into an agent scheduling module for processing to obtain the list of agents and the drawing topic parameters corresponding to each agent in the list of agents; wherein, the drawing topic parameters corresponding to each agent include at least one sub-topic parameter, and the drawing topic parameters corresponding to each agent are associated with the feature configuration dimension corresponding to the agent.

[0112] In an optional embodiment, before the step of processing the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words, the method further includes: writing the feature configuration information corresponding to each agent into a shared field, wherein the shared field is managed by a field-level protection strategy; wherein, the step of processing the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words includes: when the task orchestrator determines that each agent is in the execution completion state, reading the feature configuration information corresponding to each agent from the shared field; and inputting the feature configuration information corresponding to each agent into the prompt word generation module for processing to obtain global drawing prompt words.

[0113] In an optional embodiment, after the step of inputting the global drawing prompt to the image generation model for processing to obtain at least one target object image is executed, the method further includes: inputting the at least one target object image to the compliance detection module for processing to obtain detection information corresponding to each target object image; filtering the at least one target object image according to the detection information and deleting failed object images that do not meet the image usage conditions; and determining the target image based on the filtering results and deletion results.

[0114] In an optional embodiment, after the step of inputting the global drawing prompts into the image generation model for processing to obtain at least one target object image is executed, the method further includes: inputting the at least one target object image into an image evaluation module for processing to obtain an evaluation report corresponding to each target object image, wherein the evaluation dimension of the image evaluation module is determined according to the feature configuration dimension corresponding to each agent; if it is determined from the evaluation report that there is an optimized object image among the at least one target object image, the method returns to the step of processing the corresponding drawing theme parameters of each agent to obtain the feature configuration information corresponding to each agent; or, the method returns to the step of processing the feature configuration information corresponding to each agent through the prompt generation module to obtain global drawing prompts; until a target object image that meets the image usage conditions is obtained.

[0115] In one optional embodiment, the step of inputting the global drawing prompt into an image generation model for processing to obtain at least one target object image includes: reviewing the global drawing prompt; if the review is passed, inputting the global drawing prompt into an image generation model for processing to obtain at least one target object image; if the review fails, optimizing the global drawing prompt to obtain a target drawing prompt, and inputting the target drawing prompt into an image generation model for processing to obtain at least one target object image.

[0116] In an optional embodiment, selecting at least one first agent from the candidate agent set based on the drawing object information includes: determining drawing theme information based on the drawing object information; selecting at least one first agent from the candidate agent set based on the drawing theme information; wherein, inputting the global drawing prompts into the image generation model for processing to obtain at least one target object image includes: updating the global drawing prompts using the drawing theme information; inputting the updated global drawing prompts into the image generation model for processing to obtain at least one target object image.

[0117] The above is an illustrative scheme of an image generation system according to this embodiment. It should be noted that the technical solution of this image generation system and the technical solution of the image generation method described above belong to the same concept. For details not described in detail in the technical solution of the image generation system, please refer to the description of the technical solution of the image generation method described above.

[0118] Corresponding to the above method embodiments, this application also provides an image generation apparatus embodiment. Figure 5 A schematic diagram of the structure of an image generation apparatus according to an embodiment of this application is shown. Figure 5 As shown, the device includes: The information acquisition module 502 is configured to acquire drawing object information, determine a list of intelligent agents based on the drawing object information, and generate drawing topic parameters corresponding to each intelligent agent in the list of intelligent agents; The parameter processing module 504 is configured to use each agent to process the corresponding drawing topic parameters and obtain the feature configuration information corresponding to each agent. The information processing module 506 is configured to process the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words; The image generation module 508 is configured to input the global drawing prompts into the image generation model for processing to obtain at least one target object image.

[0119] In one optional embodiment, obtaining the drawing object information includes: determining the drawing object information in response to a natural language editing request; or, determining the object editing information in response to a trigger operation on the information configuration interface in the information configuration interface, and generating the drawing object information based on the object editing information.

[0120] In one optional embodiment, determining the agent list based on the drawing object information includes: selecting at least one first agent from a set of candidate agents based on the drawing object information; displaying an agent editing page based on the drawing object information; determining at least one second agent in response to an agent editing request submitted through the agent editing page; and generating an agent list based on the at least one first agent and / or the at least one second agent.

[0121] In an optional embodiment, determining the list of agents based on the drawing object information and generating drawing topic parameters corresponding to each agent in the list of agents includes: inputting the drawing object information into an agent scheduling module for processing to obtain the list of agents and the drawing topic parameters corresponding to each agent in the list of agents; wherein, the drawing topic parameters corresponding to each agent include at least one sub-topic parameter, and the drawing topic parameters corresponding to each agent are associated with the feature configuration dimension corresponding to the agent.

[0122] In an optional embodiment, before the step of processing the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words, the method further includes: writing the feature configuration information corresponding to each agent into a shared field, wherein the shared field is managed by a field-level protection strategy; wherein, the step of processing the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words includes: when the task orchestrator determines that each agent is in the execution completion state, reading the feature configuration information corresponding to each agent from the shared field; and inputting the feature configuration information corresponding to each agent into the prompt word generation module for processing to obtain global drawing prompt words.

[0123] In an optional embodiment, after the step of inputting the global drawing prompt to the image generation model for processing to obtain at least one target object image is executed, the method further includes: inputting the at least one target object image to the compliance detection module for processing to obtain detection information corresponding to each target object image; filtering the at least one target object image according to the detection information and deleting failed object images that do not meet the image usage conditions; and determining the target image based on the filtering results and deletion results.

[0124] In an optional embodiment, after the step of inputting the global drawing prompts into the image generation model for processing to obtain at least one target object image is executed, the method further includes: inputting the at least one target object image into an image evaluation module for processing to obtain an evaluation report corresponding to each target object image, wherein the evaluation dimension of the image evaluation module is determined according to the feature configuration dimension corresponding to each agent; if it is determined from the evaluation report that there is an optimized object image among the at least one target object image, the method returns to the step of processing the corresponding drawing theme parameters of each agent to obtain the feature configuration information corresponding to each agent; or, the method returns to the step of processing the feature configuration information corresponding to each agent through the prompt generation module to obtain global drawing prompts; until a target object image that meets the image usage conditions is obtained.

[0125] In one optional embodiment, the step of inputting the global drawing prompt into an image generation model for processing to obtain at least one target object image includes: reviewing the global drawing prompt; if the review is passed, inputting the global drawing prompt into an image generation model for processing to obtain at least one target object image; if the review fails, optimizing the global drawing prompt to obtain a target drawing prompt, and inputting the target drawing prompt into an image generation model for processing to obtain at least one target object image.

[0126] In an optional embodiment, selecting at least one first agent from the candidate agent set based on the drawing object information includes: determining drawing theme information based on the drawing object information; selecting at least one first agent from the candidate agent set based on the drawing theme information; wherein, inputting the global drawing prompts into the image generation model for processing to obtain at least one target object image includes: updating the global drawing prompts using the drawing theme information; inputting the updated global drawing prompts into the image generation model for processing to obtain at least one target object image.

[0127] The above is a schematic scheme of an image generation apparatus according to this embodiment. It should be noted that the technical solution of this image generation apparatus and the technical solution of the image generation method described above belong to the same concept. Details not described in detail in the technical solution of the image generation apparatus can be found in the description of the technical solution of the image generation method described above. Furthermore, the components in the apparatus embodiment should be understood as functional modules necessary to implement each step of the program flow or each step of the method; these functional modules are not actual functional divisions or separations. An apparatus claim defined by such a set of functional modules should be understood as a functional module architecture that primarily implements the solution through the computer program described in the specification, and not as a physical device that primarily implements the solution through hardware.

[0128] Figure 6 A structural block diagram of a computing device 600 according to an embodiment of this application is shown. The components of the computing device 600 include, but are not limited to, a memory 610 and a processor 620. The processor 620 is connected to the memory 610 via a bus 630, and a database 650 is used to store data.

[0129] The computing device 600 also includes an access device 640, which enables the computing device 600 to communicate via one or more networks 660. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), a 5G communication network, or a combination of communication networks such as the Internet. The access device 640 may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0130] In one embodiment of this application, the aforementioned components of the computing device 600 and Figure 6 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 6 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this application. Those skilled in the art can add or replace other components as needed.

[0131] The computing device 600 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones, in-vehicle computers, POS machines, game consoles, etc.), wearable computing devices (e.g., smartwatches, smart glasses, etc.), smart home appliances, multimedia playback devices, smart voice interaction devices, or other types of mobile devices, or stationary computing devices such as desktop computers or PCs. The computing device 600 can also be a mobile or stationary server. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0132] The processor 620 is used to execute computer-executable instructions for the image generation method.

[0133] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the image generation method described above belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the image generation method described above.

[0134] Reference Figure 7 , Figure 7 To implement the structural block diagram of the image generation method of this embodiment in a terminal application, the terminal includes: a radio frequency (RF) circuit 710, a memory 720, an input unit 730, a display unit 740, a sensor 750, an audio circuit 760, a wireless fidelity (WiFi) module 770, a processor 780, and a power supply 790, etc. Those skilled in the art will understand that... Figure 7 The terminal structure shown does not constitute a limitation on mobile phones or computers and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0135] The RF circuit 710 can be used to receive and transmit signals during information transmission or calls. In particular, it receives downlink information from the base station and processes it with the processor 780; in addition, it transmits uplink data to the base station.

[0136] The memory 720 can be used to store software programs and modules, and the processor 780 executes various functional applications of the terminal by running the software programs and modules stored in the memory 720.

[0137] The input unit 730 can be used to receive input numeric or character information, and to generate key signal inputs related to the terminal's settings and function control. Specifically, the input unit 730 may include a touch panel 731 and other input devices 732.

[0138] The display unit 740 can be used to display input or provided information, as well as various menus of the terminal. The display unit 740 may include a display panel 741.

[0139] Audio circuitry 760, speaker 761, and microphone 762 provide an audio interface.

[0140] In this embodiment, the processor 780 included in the terminal can execute the image generation method of the previous embodiment.

[0141] The terminals disclosed in this embodiment include, but are not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, and aircraft. The embodiments of this invention can be applied to various scenarios.

[0142] Figure 8 This diagram illustrates the structural block diagram of the server portion corresponding to the implementation of the image generation method according to embodiments of the present disclosure. The server 800 can vary significantly due to different configurations or performance characteristics, and may include one or more Central Processing Units (CPUs) 822 (e.g., one or more processors) and a memory 832, and one or more storage media 830 (e.g., one or more mass storage devices) for storing application programs 842 or data 844. The memory 832 and storage media 830 may be temporary or persistent storage. The program stored in the storage media 830 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server. Furthermore, the CPU 822 may be configured to communicate with the storage media 830 and execute the series of instruction operations in the storage media 830 on the server.

[0143] The server may also include one or more power supplies 826, one or more wired or wireless network interfaces 850, one or more input / output interfaces 858, and / or one or more operating systems 841, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0144] The central processing unit 822 in the server can be used to execute the image generation method of the present disclosure embodiments.

[0145] An embodiment of this application also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, are used for an image generation method.

[0146] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the image generation method described above belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the image generation method described above.

[0147] An embodiment of this application also provides a chip that stores a computer program, which, when executed by the chip, implements the steps of the image generation method.

[0148] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the image generation method described above.

[0149] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the image generation method described above belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the image generation method described above.

[0150] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0151] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0152] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0153] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0154] The preferred embodiments disclosed above are merely illustrative of this application. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this application. These embodiments are selected and specifically described in this application to better explain the principles and practical applications of this application, thereby enabling those skilled in the art to better understand and utilize this application. This application is limited only by the claims and their full scope and equivalents.

Claims

1. An image generation method, characterized in that, include: Obtain drawing object information, determine a list of agents based on the drawing object information, and generate drawing topic parameters corresponding to each agent in the agent list; By processing the corresponding drawing topic parameters for each agent, the feature configuration information for each agent can be obtained. The prompt word generation module processes the feature configuration information corresponding to each agent to obtain global drawing prompt words; The global drawing prompts are input into the image generation model for processing to obtain at least one target object image.

2. The image generation method according to claim 1, characterized in that, The process of obtaining drawing object information includes: In response to a natural language editing request, determine the drawing object information; or, In response to a trigger operation on the information configuration interface in the information configuration interface, the object editing information is determined, and drawing object information is generated based on the object editing information.

3. The image generation method according to claim 1, characterized in that, Determining the list of intelligent agents based on the drawing object information includes: Based on the drawing object information, at least one first agent is selected from the candidate agent set; Display an agent editing page based on the drawing object information, and determine at least one second agent in response to an agent editing request submitted through the agent editing page; Generate a list of agents based on the at least one first agent and / or the at least one second agent.

4. The image generation method according to claim 1, characterized in that, The step of determining the list of agents based on the drawing object information and generating drawing topic parameters corresponding to each agent in the list includes: The drawing object information is input into the agent scheduling module for processing to obtain an agent list and drawing topic parameters corresponding to each agent in the agent list; The drawing topic parameters for each agent include at least one sub-topic parameter, and the drawing topic parameters for each agent are associated with the feature configuration dimension of the agent.

5. The image generation method according to claim 1, characterized in that, Before the step of processing the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words is executed, the following steps are also included: The feature configuration information corresponding to each intelligent agent is written into a shared field, wherein the shared field is managed by a field-level protection strategy; The step of processing the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words includes: When the task orchestrator determines that each agent is in the execution completed state, the element configuration information corresponding to each agent is read from the shared field. The feature configuration information corresponding to each agent is input into the prompt word generation module for processing to obtain global drawing prompt words.

6. The image generation method according to any one of claims 1 to 5, characterized in that, After the step of inputting the global drawing prompts into the image generation model for processing to obtain at least one target object image is executed, the method further includes: The at least one target object image is input into the compliance detection module for processing to obtain detection information corresponding to each target object image; The at least one target object image is filtered based on the detection information, and failed object images that do not meet the image usage conditions are deleted. The target image is determined based on the filtering results and deletion results.

7. The image generation method according to any one of claims 1 to 5, characterized in that, After the step of inputting the global drawing prompts into the image generation model for processing to obtain at least one target object image is executed, the method further includes: The at least one target object image is input into the image evaluation module for processing to obtain an evaluation report corresponding to each target object image. The evaluation dimension of the image evaluation module is determined according to the element configuration dimension corresponding to each agent. If the evaluation report determines that there is an optimized object image in at least one target object image, the process returns to the step of processing the corresponding drawing topic parameters for each agent to obtain the feature configuration information for each agent; or, the process returns to the step of processing the feature configuration information for each agent through the prompt word generation module to obtain global drawing prompt words; until a target object image that meets the image usage conditions is obtained.

8. The image generation method according to claim 1, characterized in that, The step of inputting the global drawing prompts into an image generation model for processing to obtain at least one target object image includes: The global drawing prompts will be reviewed. If the review is approved, the global drawing prompts are input into the image generation model for processing to obtain at least one target object image; If the review fails, the global drawing prompts are optimized to obtain target drawing prompts. The target drawing prompts are then input into an image generation model for processing to obtain at least one target object image.

9. The image generation method according to claim 3, characterized in that, The step of selecting at least one first agent from the candidate agent set based on the drawing object information includes: Determine the drawing theme information based on the drawing object information; Based on the drawing topic information, at least one first agent is selected from the candidate agent set; The step of inputting the global drawing prompts into an image generation model for processing to obtain at least one target object image includes: The global drawing prompts are updated using the drawing theme information, and the updated global drawing prompts are input into the image generation model for processing to obtain at least one target object image.

10. An image generation system, characterized in that, Including both client and server sides, including: The client is configured to determine the drawing object information in response to the drawing request submitted by the user, and send the drawing object information to the server. The server is configured to determine a list of agents based on the drawing object information, and generate drawing theme parameters corresponding to each agent in the list; process the corresponding drawing theme parameters using each agent to obtain feature configuration information corresponding to each agent; process the feature configuration information corresponding to each agent through a prompt word generation module to obtain global drawing prompt words; input the global drawing prompt words into an image generation model for processing to obtain at least one target object image, and send the target object image to the client.

11. An image generation apparatus, characterized in that, include: The information acquisition module is configured to acquire drawing object information, determine a list of agents based on the drawing object information, and generate drawing topic parameters corresponding to each agent in the list of agents. The parameter processing module is configured to use each agent to process the corresponding drawing topic parameters and obtain the feature configuration information for each agent. The information processing module is configured to process the feature configuration information corresponding to each agent through the prompt word generation module to obtain global drawing prompt words; The image generation module is configured to input the global drawing prompts into the image generation model for processing to obtain at least one target object image.

12. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the steps of the method according to any one of claims 1 to 9.

13. A computer-readable storage medium storing computer instructions, characterized in that, When executed by the processor, this instruction implements the steps of the method according to any one of claims 1 to 9.

14. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 9.

Images