Game generation method and device, storage medium and computer equipment

Abstract

The invention discloses a game generation method and device, a storage medium and computer equipment. The game generation method comprises the steps of obtaining a game development demand based on natural language interaction; generating game content data based on the game development demand; and driving a game engine based on the game content data to obtain a target game. By means of the method, the game content data can be automatically generated according to the game development requirements, the target game capable of driving the game engine to run is finally generated, efficient and high-quality game generation is achieved, the game development efficiency is improved, the game development threshold is lowered, the game development requirements of more developers are met, and the game development efficiency is improved. The illusion problem occurring in the game code generation task is relieved, the game code capable of well running is generated, and the adverse effect of redundant codes existing in the generated game on the game performance is avoided.

Description

Technical Field

[0001] The present application relates to the field of game development technology, and in particular to a game generation method, device, storage medium and computer equipment. Background Art

[0002] With the rapid development of artificial intelligence technology, large language models (LLMs) have demonstrated their powerful text generation capabilities and have been applied in many fields. However, in the field of game development, the application of existing large language models and related architectures still faces many challenges and limitations.

[0003] First, due to the lack of high-quality corpus for the game industry during model training, existing large language models generally have serious "illusion problems" when performing game code generation tasks, that is, they cannot accurately, stably, and effectively generate game code that can run well, which affects the effectiveness of the code and increases the difficulty of subsequent development and debugging. Secondly, existing models often cannot avoid generating redundant code, which not only wastes computing resources, but also has a negative impact on game performance. The generated games usually lack fluency and often have functional crosstalk, which is not conducive to the user's gaming experience. In addition, most existing technical architectures are limited to the code generation level and cannot directly output complete games that can run in the game engine according to developer needs. As a core component of the game industry, the game engine is crucial to the development and operation of games. The commercialization and industrialization prospects of games that cannot be directly run in the game engine will also be greatly reduced. Summary of the invention

[0004] The present application mainly provides a game generation method, device, storage medium and computer equipment, aiming to solve the technical problem of poor effectiveness of existing game generation methods.

[0005] In order to solve the above technical problems, the technical solution adopted by the present application is: to provide a game generation method. The game generation method includes: obtaining game development requirements based on natural language interaction; generating game content data based on the game development requirements; and driving a game engine based on the game content data to obtain a target game.

[0006] In some embodiments, the obtaining of game development requirements based on natural language interaction includes: obtaining original development requirements based on natural language, and performing semantic analysis on the original development requirements to obtain requirement semantic information; in response to determining that the requirement semantic information is unclear, initiating a first data completion dialogue, and receiving a first response supplementary data to the first data completion dialogue to complete the requirement semantic information; in response to determining that the requirement semantic information is clear, generating the game development requirements based on the requirement semantic information.

[0007] In some embodiments, generating game content data based on the game development requirements includes: parsing the game development requirements and generating a development plan; breaking down the development plan into at least one development subtask; generating a pending game code corresponding to the development subtask, and obtaining a pending game asset corresponding to the pending game code; in response to determining that the parameters in the pending game code are complete and the pending game assets are complete, integrating the pending game code and the pending game assets to obtain pending game content; in response to determining that the pending game content meets the game running conditions, determining the pending game content as the game content data.

[0008] In some embodiments, after generating the pending game code corresponding to the development subtask and obtaining the pending game assets corresponding to the pending game code, it also includes: in response to determining that the pending game assets are incomplete, initiating an asset request, and receiving asset completion data for the asset request to complete the pending game assets.

[0009] In some embodiments, the asset completion data is generated based on the asset request by at least one of an image generation model, a video generation model, a 3D generation model, a music generation model, a special effect generation model, and a multimodal generation model, and / or the asset completion data is generated using retrieval enhancement.

[0010] In some embodiments, in response to determining that the parameters in the pending game code are complete, and the pending game assets are complete, integrating the pending game code and the pending game assets to obtain the pending game content, it also includes: in response to determining that the pending game content does not meet the game running conditions, obtaining running error information; based on the running error information, re-disassembling the development plan to determine the new development subtasks.

[0011] In some embodiments, driving the game engine based on the game content data to obtain the target game includes: sending an application request, and in response to receiving an application permission response to the application request, performing a custom parameter detection on the game content data; in response to determining that there are uncompleted custom parameters in the game content data, initiating a second data completion dialogue, and receiving a second response supplementary data to the second data completion dialogue to complete the custom parameters and update the game content data; in response to determining that there are no uncompleted custom parameters in the game content, driving the game engine based on the game content data to obtain the target game.

[0012] To solve the above technical problems, another technical solution adopted in the present application is: to provide a game generating device, which includes: an interaction module, used to obtain game development requirements based on natural language interaction; a generation module, used to generate game content data based on the game development requirements; a driving module, used to drive the game engine based on the game content data to obtain a target game.

[0013] In some embodiments, the generation module includes: a development architecture agent, which is used to parse the game development requirements and generate a development plan; a development engineer agent, which is used to decompose the development plan into at least one development sub-task; an engine engineer agent, which is used to generate a pending game code corresponding to the development sub-task and obtain the pending game assets corresponding to the pending game code; a review agent, which is used to determine whether the parameters in the pending game code are complete and whether the pending game assets are complete, and in response to determining that the parameters in the pending game code are complete and the pending game assets are complete, integrate the pending game code and the pending game assets to obtain pending game content; a test engineer agent, which is used to test the pending game content and in response to determining that the pending game content meets the game running conditions, determine the pending game content as the game content data.

[0014] In some embodiments, the engine engineer agent is obtained by fine-tuning instructions on a preset large language model using preset code corpus, and the preset code corpus satisfies preset high decoupling conditions and preset high annotation conditions, and corresponds to the game engine.

[0015] In order to solve the above technical problems, another technical solution adopted by the present application is: providing a storage medium, on which program data is stored, characterized in that when the program data is executed by a processor, the steps of the game generation method as described above are implemented.

[0016] In order to solve the above technical problems, another technical solution adopted in the present application is: to provide a computer device, which includes a processor and a memory connected to each other, the memory stores a computer program, and when the processor executes the computer program, the steps of the game generation method as mentioned above are implemented.

[0017] The beneficial effects of the present application are as follows: Different from the prior art, the present application discloses a game generation method, device, storage medium and computer equipment. After obtaining the game development requirements based on natural language interaction, the present application generates game content data based on the game development requirements, and drives the game engine based on the game content data to obtain the target game, effectively realizing the automatic development of games based on the game development requirements, being able to meet the game development requirements of more different developers, and realizing efficient and high-quality game generation. At the same time, by combining the generated game content data with the game engine, the game content can be run more effectively and reliably in the game engine, and the required target game is obtained, thereby improving the game development efficiency, and its intelligence and automation level, lowering the game development threshold, meeting the game development requirements of more developers, and being able to reduce the illusion problem in the game code generation task, generate game code that can run well, and avoid the adverse effects of redundant code in the generation model on game performance, which is conducive to the commercialization and industrialization of the game generation method based on the large language model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0019] Figure 1 It is a flowchart of an embodiment of a game generation method provided by the present application;

[0020] Figure 2 yes Figure 1 A schematic flow chart of an embodiment of step 10 in the embodiment;

[0021] Figure 3 yes Figure 1 A schematic flow chart of an embodiment of step 20 in the embodiment;

[0022] Figure 4 yes Figure 1 A schematic flow chart of an embodiment of step 30 in the embodiment;

[0023] Figure 5 It is a structural schematic diagram of an embodiment of a game generating device provided by the present application;

[0024] Figure 6 yes Figure 5 A schematic diagram of the structure of an embodiment of the generation module 42 in the embodiment;

[0025] Figure 7It is a structural schematic diagram of an embodiment of a storage medium provided by the present application;

[0026] Figure 8 It is a structural diagram of an embodiment of a computer device provided by the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] The terms "first", "second", "third" in the embodiments of the present application are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the feature defined as "first", "second", "third" can expressly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically limited. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0029] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0030] This application provides a game generation method, see Figure 1 , Figure 1 1 is a flow chart of an embodiment of a game generation method provided by the present application, the game generation method comprising:

[0031] Step 10: Obtain game development requirements based on natural language interaction.

[0032] In this embodiment, the game development requirements are specific requirements for game functions, scenes, characters, mechanisms, gameplay and storylines that the game developer hopes to achieve. The acquisition of the game development requirements can be achieved through various channels such as game development documents, market research, user feedback or hot topics in the community. The game development requirements of this solution are preferably obtained through natural language interaction (Natural Language Interaction, NLI), and the development requirements expressed by the developer in natural language are obtained through dialogue. The developer responds and asks questions in natural language with the intention of improving the development requirements, so as to continuously parse and supplement the development requirements, and finally obtain the game development requirements for subsequent generation processing.

[0033] For example, the natural language expression of the game development requirements for developing a Tetris game can be roughly: "Different shapes of combined blocks keep falling from the top of the screen, and the player needs to rotate and move the blocks to arrange them in a row. When a row is completely filled, the row disappears and the player gets points..." and other information. Through natural language interaction, the imperfect semantics of the development requirements expressed in the above natural language can be further improved. For example, for "different shapes", you can ask "which shapes are included"; for "players need to rotate and move blocks", you can ask "controlled by keyboard, touch or mouse", and you can continue to optimize and ask detailed questions in the future, such as "which keyboard key is used for control", "what are the functions of the up, down, left and right keys respectively", etc., and obtain more complete and reliable game development requirements through the developer's answers or information supplements.

[0034] In addition, the game development requirements may also be a statement of improvements in existing games, which may specifically be improvements or adjustments in the appearance, actions, sound effects, and plots of objects in the game. For example, for an existing Tetris game, the game development requirements may specifically be: "adjust the background to gray", "adjust the falling speed from increasing by score to increasing by game time", etc.

[0035] The natural language information on the developer side can be sent in a variety of ways, such as text input in the input box, document import, audio and video recognition, etc., and the product side corresponding to the embodiment of the method can send corresponding natural language information through a large language model, a dialogue system, and a chat robot, etc., and display it through a display interface or window, or broadcast it through an audio and video system. In this way, developers can convey their creativity and needs more intuitively and efficiently, and products that apply the game generation method can effectively generate games according to these needs, reducing the format and standardization requirements for game development needs, and improving the efficiency of game planning and development.

[0036] Optionally, see Figure 2, to obtain game development requirements, you can follow the following steps:

[0037] Step 11: Obtain original development requirements based on natural language, and perform semantic analysis on the original development requirements to obtain semantic information of the requirements.

[0038] Step 12: In response to determining that the demand semantic information is unclear, a first data completion dialogue is initiated, and a first response supplementary data to the first data completion dialogue is received to complete the demand semantic information.

[0039] Step 13: In response to determining that the requirement semantic information is clear, generate game development requirements based on the requirement semantic information.

[0040] Semantic analysis is the process of processing natural language text to extract its meaning and context. In this process, by performing semantic analysis on the original development requirements, keywords, phrases, sentence structures and the relationships between them can be identified, thereby generating demand semantic information so that subsequent steps can more accurately understand the developer's intentions. Semantic analysis can be implemented based on a pre-trained semantic analysis model, or by matching and parsing with a predefined set of rules and a vocabulary, or by using a statistical model in natural language processing to semantically annotate and parse the original development requirements. The demand semantic information obtained through semantic analysis can relatively more clearly indicate the relationship between semantic roles, the emotional tendency expressed by the text, and the purpose and intention of the requirements, which is convenient for subsequent processing.

[0041] However, since the amount of information in the original development requirements is usually limited, it is often impossible to indicate the necessary elements of some objects in the game, and the semantic information obtained is usually only a clarification of the original development requirements within the scope of the original development requirements, and there will still be deviations or omissions in the elements. For example, in the Tetris game above, the falling speed of the blocks, the starting point of the falling, the conditions for stopping the falling, the rotation direction, and the moving touch conditions are all factors that should be considered, but the original development requirements are usually difficult to be detailed. For example, the configuration of the falling speed is ignored, resulting in the blocks only staying in place, and the distinction between the controlled object blocks and other object blocks is ignored, resulting in the blocks still being controlled after the falling is completed.

[0042] In order to make up for this deficiency, this application introduces an element completion mechanism based on semantic analysis. After obtaining the demand semantic information, this application will further analyze the demand semantic information to determine whether there are missing elements and whether the demand semantic information is clear, and then complete the semantic information that is still unclear.

[0043] Preferably, the present application uses a large language model for semantic analysis and element clarity judgment, and conducts subsequent element completion dialogue based on natural language interaction. Thanks to the large language model's powerful natural language understanding and reasoning capabilities, and its learning of a large amount of game development-related data, the large language model can conduct a deeper analysis and understanding of the elements in game development, and identify potential missing or unclear elements.

[0044] Specifically, when it is determined that the semantic information of the requirement is unclear, the present application will trigger the first data completion dialogue, and guide the developer to provide the information of the elements through an interactive method to complete the missing parts in the semantics. Among them, the first data completion dialogue is preferably established based on a large language model, and can also be established by calling a dedicated conversational platform interface, or by a specific data structure such as a data linked list that is strongly associated with the task to model the dialogue process. The first data completion dialogue will generate a series of guiding questions based on the current semantic information of the requirements. These questions are intended to reveal the game elements and details that the developer is not clear about. The developer provides the corresponding first answer supplementary data by answering these questions, thereby completing the semantic information of the requirements. Among them, the first answer supplementary data is a semantic answer to each guiding question in the dialogue process. During the dialogue process, the first data completion dialogue can also adjust its question raising according to the clarity of the answer to the corresponding guiding question and the additional feature information, so as to achieve a more effective and reliable dialogue, thereby completing the semantic information of the requirements through the first answer supplementary data, and ensuring that the game development requirements are more complete and clear.

[0045] For example, in a game, it is necessary to update the scene texture. The original development requirement is "I want to update the texture of the residential building". After semantic analysis, the only residential building in the game is the residential building 101 in Street View 1. Then the semantic information of the requirement may be "update the texture of the residential building 101 in Street View 1". After further analysis by the large language model, it is determined that the appearance features and resolution information of the texture to be updated need to be clarified. Then the first data completion dialogue may first ask "What are the appearance features of this building?", so that the corresponding answer can be "Residential building style 2". Then the first completion dialogue will then ask, "What is the resolution size?", so that the corresponding answer can be "General distant view texture resolution" or specifically "20ppi", so as to achieve the update of the residential building texture. For another example, in the answer to the appearance features, the answer is "Residential building style 2 in distant view texture format", which is equivalent to answering the question related to the resolution size. Therefore, the first completion dialogue for subsequent requests to supplement the resolution size will not be initiated, thereby improving the efficiency of the dialogue.

[0046] After completing the conversation, developers can continue to propose more development requirements, and then continuously update and optimize game development requirements through natural language interaction. After determining the game development requirements, they can also actively trigger subsequent game generation through controls or instructions, or set it to automatically trigger subsequent game generation after the game development requirements are clarified. At the same time, corresponding element configuration tools can also be set during the conversation. For example, when it is necessary to clarify the target object and location coordinates, you can click on the corresponding element in the image or code to automatically determine its corresponding element information. When it is necessary to clarify color information, you can select it through the color picker or set the color customization tool for configuration, which improves the convenience of answering and ensures the accuracy and completeness of game development requirements.

[0047] When the semantic information of the requirements is clear, detailed and accurate game development requirements can be generated based on the semantic information of the requirements. These game development requirements can be directly used to guide the game development team in subsequent game development work as a blueprint for game development, ensuring that the entire development process can proceed in an orderly manner, while ensuring that the final developed game meets the original intention of the developer. In addition, when generating game development requirements based on the large language model, the cross-domain learning and integration capabilities of the large language model can also be used to combine game design with knowledge in related fields, such as considering the physics logic, economic logic, aesthetic logic, and plot rationality logic in the game scene, etc., to provide developers with more creative and practical suggestions, making the subsequent game development process more flexible and efficient, and helping to improve the quality of the game and user experience.

[0048] Step 20: Generate game content data based on game development requirements.

[0049] After the aforementioned clear and complete game development requirements are obtained, the game content data can be further generated. Among them, the game content data covers various elements and details of the game, including but not limited to the game's plot, characters, scenes, props, music, sound effects, animations and special effects.

[0050] Game content data is mainly presented in two ways, namely game code and game assets. Game code is the programming language code of game logic and functions, including the code implementation of various aspects of the game's control logic, physics engine, network communication and data storage, which is used to ensure that the game can run as expected, while game assets are various media resources used in the game, such as pictures, audio, video, animation and effects, etc., which are used to ensure the audio-visual and operating experience of the game.

[0051] In the process of generating game content data based on game development requirements, we first plan and design various elements in the game in detail according to the game development requirements, then encode the game content and bind assets to obtain the corresponding game code and game assets, thereby determining the corresponding game content data, and through engine testing and interface adaptation, make the generated game content data more suitable for subsequent game engine drivers.

[0052] Specifically, in the generation process, there are different development modes for different types of games. For example, for story-based games, the game plot will be written according to the game development requirements, and then the scenes, characters, dialogues and other information will be introduced based on the game plot. For strategy games, scenes and rules will be designed based on the core gameplay of the game strategy to build a variety of levels and challenges. The development process can be based on the traditional waterfall development model or agile development model, or it can be achieved through rapid iterative development based on automated generation and collaboration tools, or it can be achieved through game generation technology based on deep learning and big data.

[0053] In this application, it is preferred to implement the game content data in an automated manner based on a large language model. The large language model can understand and parse complex game development requirements, and convert the requirements into specific game content data through natural language processing technology. In this process, the large language model can not only generate game code, but also automatically create or configure game assets, greatly improving the efficiency and accuracy of game development. When generating game content data, the large language model will comprehensively consider various factors in the game development requirements, such as game type, style, target user group, etc., to ensure that the generated content is consistent with the overall game design. At the same time, the large language model also has the ability of self-learning and optimization, and can continuously learn lessons from past development experience to improve the quality and creativity of generated content.

[0054] Optionally, see Figure 3 To generate game content data based on game development requirements, you can follow the steps below:

[0055] Step 21: Analyze game development requirements and generate a development plan.

[0056] Step 22: Break down the development plan into at least one development subtask.

[0057] Step 23: Generate pending game code corresponding to the development subtask, and obtain pending game assets corresponding to the pending game code.

[0058] Step 24: In response to determining that the pending game asset is incomplete, an asset request is initiated, and asset completion data for the asset request is received to complete the pending game asset.

[0059] Step 25: In response to determining that the parameters in the pending game code are complete and the pending game assets are complete, the pending game code and the pending game assets are integrated to obtain the pending game content.

[0060] Step 26: In response to determining that the pending game content meets the game running condition, the pending game content is determined as game content data.

[0061] Step 27: In response to determining that the pending game content does not meet the game running conditions, obtaining running error information.

[0062] Step 28: Re-decompose the development plan based on the operation error information to determine new development subtasks.

[0063] In this optional embodiment, the process of generating game content based on game development requirements is specifically decoupled into multiple implementation steps, which can be executed separately by multiple executors through different servers, processing units or functional modules, or by splitting a single executor into multiple execution units through threads, processes or microservices. The process of automating each step can be specifically implemented by multiple pre-trained models, or can be implemented using traditional algorithms or tools for different tasks. In this solution, it is preferred to implement it using multiple intelligent agents based on a large language model. For details, please refer to the relevant description of the game generation device described later, which will not be repeated here.

[0064] Among them, parsing game development requirements refers to deeply understanding and analyzing the requirements and purposes of game development requirements provided by developers. This process involves a careful study of game development requirements and converting them into structured, actionable tasks and indicators, thereby determining the development plan used to guide subsequent development work and ensuring that they can be effectively disassembled and tasks executed later. In the process of automated implementation of this step, game development requirements can be parsed through corresponding preset models or algorithms, and key information such as game type, target users, and core gameplay can be extracted, and a development plan can be generated based on this information.

[0065] After the development plan is determined, it can be further broken down into several specific development subtasks. These subtasks usually correspond to a specific module or function in the game, such as character design, scene construction, and audio and video matching. By breaking down the complex development plan into a series of simple subtasks, it is easier to allocate resources and synchronize progress during the game development process, thereby improving development efficiency. In the process of automatic implementation of this step, the development plan can be broken down into specific development subtasks through a task decomposition algorithm or rule engine. During the decomposition process, factors such as task dependencies, priorities, and resource allocation can be considered to improve the subsequent integration efficiency of the corresponding tasks.

[0066] Subsequently, for each development subtask, the corresponding pending game code is generated, and the corresponding pending game assets are obtained to convert the abstract development subtask into a specific programming language and resource file, ensuring that the code and assets of each subtask can work together to form the preliminary framework of the game. Specifically, the generation of pending game code can be implemented based on a generative pre-trained model (Generative Pre-Trained Transformer, GPT), using the knowledge stored in the unstructured model vector space to generate the pending game code corresponding to the sub-development task requirements, or based on RAG (Retrieval-Augmented Generation) to achieve generation by combining the ability to retrieve information in an external knowledge base and the ability to generate text. The acquisition of pending game assets can be determined by automatically retrieving the required assets corresponding to the pending game code from the resource library through an automated tool or API (Application Programming Interface, application programming) interface, or it can be obtained by generation, using asset generation models such as pictures, audio, video, animation and effects, and automatically generating the required resources according to the requirements of the pending game code.

[0067] After determining the pending game code and pending game assets corresponding to the development subtask, there may still be incomplete parameters. Specifically, it may be because all details are not fully considered when the task is disassembled, or the requirements have changed during the game development process, and further supplementation is required. In this case, the lack of some parameters will cause the subsequent generated pending game content to fail to meet the game running conditions, such as the lack of dynamic link library files (DLL), the lack of driver parameters for specific graphics cards or sound cards, and the lack of game configuration parameters stored in .ini, .cfg and other formats, etc., and the lack of these parameters will cause the game to fail to run effectively. However, these parameters are usually default. As long as the intelligent agent preferably used in this scheme has the knowledge of the configuration and storage methods related to these parameters, it can usually automatically complete and improve such parameters based on the error correction and self-checking mechanism. As long as a more comprehensive corpus is used for fine-tuning in the process of obtaining the intelligent agent, the problem of such parameter missing can usually be solved. In the case of such parameter missing, the subsequent testing process can also timely discover the situation where the game cannot run normally due to the missing parameters, and perform corresponding code debugging and patching.

[0068] In other cases, the absence of some parameters will not affect the operation of the game. These parameters usually have a certain degree of customizability, such as the costume configuration of the characters and scenes in the game, the cooldown and effect configuration of the game skills, etc. These can be further customized after the game engine is driven, or they can be configured by default. The game development requirements for modifying these parameters can be obtained and processed through natural language interaction. For details, please refer to step 30 and its optional embodiments described later, which will not be elaborated here.

[0069] At the same time, since the amount of asset data that can be provided manually is often limited, after the pending game code is determined, there may still be a situation where the corresponding assets are missing, which also needs to be completed. This application designs a corresponding asset completion mechanism for the problem of missing assets. Specifically, when it is detected that the pending game assets are incomplete, the asset request process will be automatically triggered. This process will first analyze the specific requirements of the pending game code for assets, including key parameters such as asset type, asset style and asset size, and generate detailed asset requests based on these parameters. On this basis, by initiating a request to an external resource library, find assets that meet the requirements. The external resource library can specifically be the Internet or a specific asset generator, which searches on the Internet or a specific asset library based on the asset request, or automatically generates corresponding assets based on the asset request to supplement the pending game assets. In addition, after receiving the returned asset completion data, the requested assets can be further screened and verified to ensure that the completed assets match the game requirements and performance requirements, and there are no copyright or legal issues. Once the verification is passed, these asset completion data will be integrated into the pending game assets to complete the asset completion work.

[0070] Optionally, the asset completion data is generated based on the asset request by at least one of an image generation model, a video generation model, a 3D generation model, a music generation model, a special effect generation model and a multimodal generation model, and / or the asset completion data is generated using retrieval enhancement.

[0071] This application preferably adopts a generation method to generate corresponding assets based on asset requests to complete the assets. Among them, since the types of game assets include various types, and their modalities include images, videos, music, special effects, and 3D models, etc., when completing assets, appropriate generation models or methods can be selected according to different asset types.

[0072] For example, for image assets, image generation models such as GAN (Generative Adversarial Network) and VAE (Variational Autoencoder) can be used. By training these models to learn a large amount of image data, images of specific styles, sizes and themes can be generated. Similarly, video assets can use video generation models, which can generate continuous video content based on image sequences or video clips. For audio assets, such as background music and sound effects, music generation models can be used. These models can generate audio data that conforms to specific music styles, rhythms and emotions. In addition, for special effects assets, such as explosions, flames and other visual effects, special effects generation models can be used. These models can simulate complex physical phenomena and visual effects to generate realistic special effects scenes. Multimodal generation models can simultaneously process multiple types of asset data, such as images, text, audio, etc., to generate richer multimedia content.

[0073] In addition, when completing assets, in order to improve the quality and accuracy of generated assets, retrieval-enhanced generation technology can also be applied. This technology combines the advantages of generative methods and retrieval methods. Specifically, first, through retrieval technology, find samples or fragments that are most similar to asset requests in existing resource libraries. These samples or fragments can be used as input or reference for the generation model to guide the model to generate results that are closer to real needs. At the same time, the generation model will also process and modify the retrieved samples according to the specific requirements of the asset request to generate new assets that meet the requirements. This retrieval-enhanced generation method can not only ensure the quality and accuracy of generated assets, but also improve generation efficiency and diversity.

[0074] When the parameters in the pending game code are complete and the pending game assets are complete, the pending game code and the pending game assets can be integrated to obtain the pending game content. The integration method includes but is not limited to associating, embedding or referencing the code with the corresponding assets to ensure that the game content can be correctly loaded and displayed at runtime. It should be understood that in the intelligent agent solution based on fine-tuning of the large language model preferably adopted in this scheme, the judgment of parameter completeness is based on the existing knowledge of the model. As long as the model confirms that it has the knowledge to configure the parameter and completes the configuration of the parameters to be configured, the parameter can be determined to be complete. In addition, the judgment of parameter completeness can also avoid the configuration of custom parameters that can be randomly configured or configured by default, and the custom parameters are completed through the subsequent step 30 and the steps of its optional embodiment.

[0075] The determined pending game content already has the basic conditions for game implementation, and can usually directly drive the game engine to obtain the corresponding game, but in order to ensure the effectiveness and reliability of the final user-oriented game, it is necessary to further test and verify the pending game content. In this test process, the game can be fully tested and evaluated by simulating the actual game running environment, such as different operating systems, game engines, hardware configurations, network environments, etc., through automated testing tools and / or manual testing methods, and the corresponding pending game content can be debugged in the game engine. This application specifically judges whether the pending game content passes the test through the game running conditions. The game running conditions can be multi-dimensional specifications and requirements, such as the game's frame rate stability, loading time, memory usage, CPU (Central Processing Unit, Central Processing Unit) occupancy rate and GPU (Graphics Processing Unit) occupancy rate and other performance indicators, as well as the availability and stability of various functional modules in the game. These conditions together constitute an evaluation system for game operation, which is used to measure the performance and stability of the game in different environments. When it is determined that the pending game content meets the game running conditions, the pending game content can be determined as the game content data that is ultimately used to drive the game engine.

[0076] In the case of determining that the pending game content does not meet the game running conditions, the operation error information can be obtained, and the development plan can be disassembled again based on the operation error information to determine the new development subtask. Among them, the operation error information can specifically include information such as error code, error description, and the specific location where the error occurred, which can directly point out the problems in the game content. Therefore, after obtaining the operation error information, the developer or the large language model preferably used in this application can conduct detailed investigation and analysis of the game content based on this information, determine the root cause of the problem, and correct the error code, or optimize the corresponding development plan, subtask disassembly and other steps, regenerate the corresponding pending game content, and test and verify the game running conditions again to ensure that the new game content can meet the requirements of the game running conditions. If there are still problems, the developer or the large language model can continue to optimize and repair the pending game content iteratively until the pending game content can pass the test smoothly. Through this judgment based on the game running conditions and automatic optimization process, the application can efficiently handle complex game development needs, obtain more effective and reliable high-quality game content data, and greatly improve the efficiency and accuracy of game development.

[0077] Step 30: Drive the game engine based on the game content data to obtain the target game.

[0078] The game engine is a set of code instructions that integrates functions such as image rendering, physical modeling, audio processing, collision detection, and user input processing, which can be recognized by computer equipment to realize the operation and presentation of the game. The game engine can specifically be Unity engine, Unreal Engine (Unreal Engine), Godot, Defold, Solar2D, etc., and this application does not make specific restrictions on this, as long as the game content data generated by the aforementioned steps can be adapted to the game engine. By inputting the game content data into the game engine, the game engine can be driven to generate the corresponding target game. Specifically, the game engine will parse the game content data, initialize the game according to the instructions and configurations in the game content data, determine the corresponding game logic, scene layout, storage path and other information, and render and apply the assets, thereby generating the corresponding target game. For example, when a game engine is loading the game content data of a Flappy Bird game, it will first import the font files needed by the game, then create the blueprint assets needed by the game and the UI (UserInterface) needed by the game, and add components, variables, functions and events to it. Finally, it will build the game level based on the created content and set the game mode corresponding to the level.

[0079] Optionally, see Figure 4 , based on the game content data, drive the game engine and get the target game. You can follow the following steps:

[0080] Step 31: Send an application request, and in response to receiving an application permission response to the application request, perform a custom parameter detection on the game content data.

[0081] Step 32: In response to determining that there are uncompleted custom parameters in the game content data, a second data completion dialogue is initiated, and a second response supplementary data to the second data completion dialogue is received to complete the custom parameters and update the game content data.

[0082] Step 33: In response to determining that there are no uncompleted custom parameters in the game content, the game engine is driven based on the game content data to obtain the target game.

[0083] This optional embodiment determines whether to use game content data to drive the game engine through an application request. The application request can be triggered after the corresponding game content data is obtained in the above determination, and is presented by displaying a box or a specific control popping up, while the application operation response is triggered manually by the developer or by a preset model recognition method or trigger mechanism, for example, by clicking a virtual button such as "Run" or "Generate". After the trigger, this solution also detects custom parameters that may be omitted or not clearly specified, and completes these custom parameters through natural language interaction to meet the configuration requirements of some configurable parameters.

[0084] The custom parameters can be information that needs to be filled in for initialization, such as the initial character name, archive name, partition name, and initial difficulty of the game, which require custom identification information. They can also be parameters that can be adjusted within the game, such as picture quality, sound effect size, control sensitivity, and other user preference settings. The existence of these custom parameters not only increases the playability and personalized experience of the game, but also provides the possibility for further expansion and optimization of game content. In addition, some custom parameters can be automatically obtained after the game is deployed. For example, the storage path of the game installation package can be automatically obtained, and the screen resolution, processor model, and graphics card information of the current user's device can be automatically obtained.

[0085] By initiating a second data completion dialogue, the present solution can actively guide developers or users to supplement these possibly omitted custom parameters to ensure the integrity and accuracy of the game content data. The embodiment of the second data completion dialogue is similar to the aforementioned first data completion dialogue and will not be described in detail here. After receiving the second response supplementary data, the game content data is updated, and then the game engine can be driven based on the updated game content data to obtain the target game.

[0086] During the game development process, as the project progresses and requirements change, the assets in the game content data may need to be updated or supplemented. In addition to the aforementioned retrieval and acquisition methods and automatic generation methods, this solution also provides active supplementation of assets, that is, through the supplementation of asset paths, where the asset path can be the storage path of the local asset in the file system, or it can be the acquisition link of the network asset. By sending a change in the asset path to the preset asset library, the preset asset library can be updated. Due to the independence of each asset, in the process of generating games or game applications, the corresponding assets can be updated in real time according to the changes in the assets in the preset asset library, without the need for additional file transfer or configuration operations, and the development process will not affect the game content data or other data in the target game, which improves the efficiency, flexibility and convenience of game development.

[0087] See also Figure 5 , Figure 5 4 is a schematic diagram of a structure of an embodiment of a game generation device provided by the present application. The game generation device 40 includes: an interaction module 41, which is used to obtain game development requirements based on natural language interaction; a generation module 42, which is used to generate game content data based on the game development requirements; and a driving module 43, which is used to drive a game engine based on the game content data to obtain a target game.

[0088] Optionally, the interaction module 41 is also specifically used to: obtain original development requirements based on natural language, and perform semantic analysis on the original development requirements to obtain requirement semantic information; in response to determining that the requirement semantic information is unclear, initiate a first data completion dialogue, and receive a first response supplementary data to the first data completion dialogue to complete the requirement semantic information; in response to determining that the requirement semantic information is clear, generate game development requirements based on the requirement semantic information.

[0089] Optionally, the generation module 42 is further used to: analyze the game development requirements and generate a development plan. Decompose the development plan into at least one development subtask; generate pending game code corresponding to the development subtask, and obtain pending game assets corresponding to the pending game code; in response to determining that the parameters in the pending game code are complete and the pending game assets are complete, integrate the pending game code and the pending game assets to obtain pending game content; in response to determining that the pending game content meets the game running conditions, determine the pending game content as game content data;.

[0090] Optionally, the generation module 42 is further configured to: in response to determining that the pending game asset is incomplete, initiate an asset request, and receive asset completion data for the asset request to complete the pending game asset.

[0091] Optionally, the generation module 42 is further specifically used for: in response to determining that the pending game content does not meet the game running conditions, obtaining running error information; and re-disassembling the development plan based on the running error information to determine new development subtasks.

[0092] Optionally, the driver module 43 is also specifically used to: send an application request, and in response to receiving an application permission response to the application request, perform custom parameter detection on the game content data; in response to determining that there are incomplete custom parameters in the game content data, initiate a second data completion dialogue, and receive second response supplementary data to the second data completion dialogue to complete the custom parameters and update the game content data; in response to determining that there are no incomplete custom parameters in the game content, drive the game engine based on the game content data to obtain the target game.

[0093] Optionally, the asset completion data is generated based on the asset request by at least one of an image generation model, a video generation model, a 3D generation model, a music generation model, a special effect generation model and a multimodal generation model, and / or the asset completion data is generated using retrieval enhancement.

[0094] Optionally, the driver module 43 is also specifically used to: send an application request, and in response to receiving an application permission response to the application request, perform custom parameter detection on the game content data; in response to determining that there are incomplete custom parameters in the game content data, initiate a second data completion dialogue, and receive second response supplementary data to the second data completion dialogue to complete the custom parameters and update the game content data; in response to determining that there are no incomplete custom parameters in the game content, drive the game engine based on the game content data to obtain the target game.

[0095] Since the embodiments of the device part correspond to the embodiments of the above-mentioned method, please refer to the above-mentioned game generation method embodiments for the introduction of the game generation device 40 provided in the embodiments of the present invention. The embodiments of the present invention will not be repeated here, and it has the same beneficial effects as the above-mentioned game generation method.

[0096] Further, see Figure 6 The generation module 42 in the game generation device 40 is preferably implemented based on a multi-agent architecture. Specifically, the generation module 42 includes:

[0097] Development framework agent 421, used to analyze game development requirements and generate development plans;

[0098] A development engineer agent 422 is used to decompose the development plan into at least one development subtask;

[0099] The engine engineer agent 423 is used to generate pending game codes corresponding to the development subtask and obtain pending game assets corresponding to the pending game codes;

[0100] Review agent 424, used to determine whether the pending game assets are complete, and in response to determining that the parameters in the pending game code are complete and the pending game assets are complete, integrate the pending game code and the pending game assets to obtain the pending game content;

[0101] The test engineer agent 425 is used to test the pending game content, and in response to determining that the pending game content meets the game running conditions, determine the pending game content as game content data.

[0102] An agent refers to an entity that can act autonomously, receive information and perform roles, and the generation module 42 of the present application is specifically a multi-agent system (MAS) that integrates multiple agents. It uses the interaction and coordination between multiple agents to complete corresponding tasks and implement corresponding execution steps. It can realize a standard workflow (Standard Operating Procedure, SOP) for generating game content data based on game development requirements in the above-mentioned game generation method.

[0103] Specifically, the development architecture agent 421 is responsible for the planning and design of the overall game development. It can understand and analyze the needs of game development, and then generate a detailed development plan based on these needs. Next, the development engineer agent 422 is responsible for breaking down this grand development plan into a series of specific and executable development subtasks. Then, the engine engineer agent 423 uses its generation and retrieval capabilities to determine the pending game code and pending game assets corresponding to each development subtask. At the same time, the review agent 424 will conduct a comprehensive review of the pending game code and pending game assets to ensure their integrity and correctness. If any problems are found, it will promptly notify the relevant agents to make corrections or completions to ensure the smooth progress of game development. When all development subtasks are completed and all codes and assets have been strictly reviewed, the test engineer agent 425 will test the pending game content to ensure that the game can operate normally under various circumstances. The above details can be referred to as follows Figures 1 to 5 The game generation method described will not be repeated here.

[0104] In the generation module 42, different agents can be obtained in different ways, specifically by training a large language model framework or other deep learning networks, or by prompting a large language model framework, or by fine-tuning the existing large language model by instructions (Instruction Tuning) or establishing a chain of thought (CoT), thereby combining various agents to obtain the generation model 42 of a multi-agent architecture. Among them, the large language model used in each method preferably adopts a generative pre-trained Transformer network architecture, and establishes a reliable, stable and effective agent by relying on the advantages of the Transformer network architecture, such as powerful parallel computing capabilities, effective modeling of long-distance dependencies, high flexibility and scalability.

[0105] Optionally, taking the engine engineering agent as an example, the engine engineer agent 423 in the present application is obtained by fine-tuning the instructions of a preset large language model using a preset code corpus, and the preset code corpus satisfies the preset high decoupling degree condition and the preset high annotation degree condition, and corresponds to the game engine.

[0106] Instruction Fine-Tuning is a technology for optimizing large language models, which aims to adjust model parameters so that it can better understand and execute human natural language instructions. In this application, the engine engineer agent 423 can be obtained by fine-tuning the instructions of the preset large language model based on the preset code corpus. Specifically, the corresponding preset code corpus can be obtained from the code corpus such as the developer's standard code document, program development manual and standardized compilable program file. These code corpora preferably use highly decoupled and highly annotated corpora to ensure the quality of subsequent code generation, and then use these instructions to fine-tune the pre-trained large language model. The fine-tuning process includes from simple code snippet generation to complex logic implementation, so as to gradually enhance the ability of the engine engineer agent 423 to understand and generate game code, so that the model can more accurately understand and execute natural language instructions related to game engine development, produce good generalization effect, and obtain the corresponding engine engineer agent 423. The obtained engine engineer agent 423 can handle development subtasks more efficiently, quickly determine the corresponding pending game code and pending game assets, thereby improving the efficiency and quality of game development.

[0107] Among them, code corpora that meet the preset high decoupling degree conditions and the preset high annotation degree conditions usually have strong content quality and applicability. Meeting the preset high decoupling degree conditions means that the code snippets of the preset code corpus have high independence, that is, the coupling degree between them is low, and the dependencies between each other are simple and clear. Correspondingly, the high decoupling degree condition can refer to determining that there is no circular dependency between code snippets, or the complexity of the circular dependency is low. Such a design helps to flexibly combine and reuse these code snippets according to specific needs during the game generation process, thereby improving the flexibility and scalability of game development. At the same time, highly decoupled code is also easier to maintain and upgrade, which facilitates subsequent modification and debugging.

[0108] The preset code corpus that meets the preset high annotation conditions has strong readability and is easy to be understood and parsed by the corresponding large language model and related semantic parsing methods, providing developers with detailed descriptions of the code function, purpose and usage. Correspondingly, the high decoupling condition can refer to determining whether the code snippet contains sufficient annotation information, such as whether each function, class or module has clear annotations to explain its function and usage. A highly annotated code corpus means that each code snippet is equipped with detailed and accurate annotations, so that non-original authors can quickly understand and use the codes. It not only improves the efficiency of game generation, but also ensures the quality and accuracy of the generated game content data. It also makes the code easier to maintain and upgrade, and facilitates subsequent modification and debugging.

[0109] The specific settings of the high decoupling degree condition and the high annotation degree condition can be determined based on the best practices of the game development industry, the actual needs of the project, and the development habits of the team. For example, the high decoupling degree condition may include a high degree of modularization of code snippets, high recognition of interface definitions, and minimization of dependencies, etc., while the high annotation degree condition may include high completeness and accuracy of the annotations, and close correspondence with the code logic, etc., which are not specifically limited in this application.

[0110] Matching the code corpus with the game engine is used to ensure that the code corpus is compatible with the specific game engine and meets the requirements of the engine in terms of code language, architecture, interface, version and protocol, so that the pending game code determined by the code corpus can be effectively parsed by the game engine, thereby driving the game engine and increasing the probability of successful game operation.

[0111] In order to make the code corpus correspond to the game engine, specifically, a mapping relationship table between the preset code corpus and the game engine can be established to clearly record key information such as the game engine version, API interface and programming language applicable to each code corpus fragment. In the code generation and integration stage, the corresponding code corpus fragments can be automatically selected and adapted according to these mapping relationships to reduce manual intervention and error risks. In addition, the natural language processing and code understanding capabilities of the large language model can be used to automatically analyze and adapt the code corpus. By training the large language model, it can be able to recognize the characteristics and requirements of different game engines, and then automatically adjust the code style, grammatical structure and API calls when generating code to ensure that the generated code meets the standards and specifications of the target game engine.

[0112] Through the above steps, it is ensured that the code corpus in the preset code corpus meets the high decoupling degree conditions and the high annotation degree conditions, and that the code corpus corresponds to the target game engine, which can significantly improve the efficiency, flexibility, readability and maintainability of game development, while reducing the risk of manual intervention and errors, improving the quality and accuracy of game content, ensuring the standardization and regularization of the game development industry, and helping to alleviate the illusion problem in the game code generation task, generate game code that can run well, and avoid the adverse effects of redundant code in the generated game on game performance.

[0113] Through the multi-agent-based generation module 42, it is possible to undertake the complex game content generation from the analysis of game development requirements to the game engine drive in the game development process, realize the automatic generation of game content based on the game development requirements, and help better meet the game generation requirements of users or developers. At the same time, this embodiment ensures the consistency and efficiency of the entire development process through data exchange and collaboration between multiple agents, not only reduces manual intervention, but also improves the standardization and repeatability of the development process, so that the generation module 42 can quickly understand the development requirements and generate high-quality codes and assets, thereby greatly improving the efficiency and quality of game development, helping to alleviate the illusion problem in the game code generation task, generate game code that can run well, and avoid the adverse effects of redundant codes in the generated game on game performance. In addition, the use of the generation module 42 can also effectively lower the threshold for game development, allowing more people to participate in the ranks of game creation to better meet the needs of commercialization and industrialization.

[0114] Optionally, the review agent 424 is further used to initiate an asset request in response to determining that the pending game asset is incomplete; the development framework agent 421 is further used to complete the pending game asset in response to receiving asset completion data for the asset request.

[0115] Optionally, the review agent 424 is also used to obtain operation error information in response to determining that the pending game content does not meet the game operation conditions; the development engineer agent 422 is also used to re-disassemble the development plan based on the operation error information to determine new development subtasks.

[0116] Optionally, the audit agent 424 is also used to review the task planning and output content after the development architecture agent 421, the development engineer agent 422, the engine engineer agent 423 and the test engineer agent 425 complete the corresponding processing steps to ensure that the generation module 42 is executed according to standard operating procedures.

[0117] It should be noted that the intelligent agent execution subjects of the above-mentioned different steps can be adaptively adjusted. For example, the same intelligent agent can be trained to implement the executable steps of the review intelligent agent 424 and the test engineer intelligent agent 425, and part of the functions of an intelligent agent can be implemented in another intelligent agent, etc. As long as the above-mentioned steps are implemented specifically through multiple intelligent agents, they can be understood to be within the protection scope of the present application.

[0118] Since the optional embodiments of part of the generation module 42 under the above-mentioned intelligent agent architecture correspond to each other with the embodiments of the above-mentioned method, please refer to the above-mentioned game generation method embodiment for the introduction of the game generation device provided in the embodiment of the present invention. The embodiment of the present invention will not be repeated here, and it also has the same beneficial effects as the above-mentioned game generation method.

[0119] See also Figure 7 , Figure 7 It is a structural diagram of an embodiment of the storage medium provided by the present application.

[0120] The storage medium 50 stores program data 51. When the program data 51 is executed by the processor, the following is achieved: Figures 1 to 5 The described game generation method.

[0121] The program data 51 is stored in a storage medium 50, and includes a number of instructions for enabling a network device (such as a router, a personal computer, a server, etc.) or a processor to execute all or part of the steps of the methods of the various embodiments of the present application.

[0122] Optionally, the storage medium 50 may be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or other medium that can store the program data 51 .

[0123] See also Figure 8 , Figure 8 It is a structural schematic diagram of an embodiment of a computer device provided by the present application.

[0124] The computer device 60 includes a processor 62 and a memory 61 connected to each other. The memory 61 stores a computer program. When the processor 62 executes the computer program, the following is achieved: Figures 1 to 5 The game generation method described herein. The memory 61 may include the storage medium 50, or may be other independently developed memory.

[0125] Different from the prior art, the present application discloses a game generation method, device, storage medium and computer equipment. After obtaining the game development requirements based on natural language interaction, the present application generates game content data based on the game development requirements, and drives the game engine based on the game content data to obtain the target game, effectively realizing the automatic development of games based on the game development requirements, being able to meet the game development requirements of more different developers, and realizing efficient and high-quality game generation. At the same time, by combining the generated game content data with the game engine, the game content can be run more effectively and reliably in the game engine, and the required target game is obtained, thereby improving the game development efficiency, and its intelligence and automation level, lowering the game development threshold, meeting the game development requirements of more developers, and being conducive to alleviating the illusion problem in the game code generation task, generating game code that can run well, avoiding the adverse effects of redundant code in the generation model on game performance, and being conducive to realizing the commercialization and industrialization of the game generation method based on the large language model.

[0126] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, storage medium embodiment, and computer device embodiment, since they are basically similar to the game generation method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0127] The above are merely embodiments of the present application and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A game generation method, characterized in that: include: Obtain game development requirements based on natural language interaction; Generate game content data based on the game development requirements; A game engine is driven based on the game content data to obtain a target game.

2. The game generation method according to claim 1, characterized in that: The obtaining of game development requirements based on natural language interaction includes: Acquire original development requirements based on natural language, and perform semantic analysis on the original development requirements to obtain semantic information of the requirements; In response to determining that the demand semantic information is ambiguous, initiating a first data completion dialogue, and receiving first response supplementary data to the first data completion dialogue to complete the demand semantic information; In response to determining that the requirement semantic information is clear, the game development requirement is generated based on the requirement semantic information.

3. The game generation method according to claim 1, characterized in that: The generating of game content data based on the game development requirements includes: Analyze the game development requirements and generate a development plan; Decomposing the development plan into at least one development subtask; Generate pending game code corresponding to the development subtask, and obtain pending game assets corresponding to the pending game code; In response to determining that the parameters in the pending game code are complete and the pending game assets are complete, integrating the pending game code and the pending game assets to obtain pending game content; In response to determining that the pending game content meets the game running condition, the pending game content is determined as the game content data.

4. The game generation method according to claim 3, characterized in that: The obtaining of pending game assets corresponding to the pending game code includes: In response to determining that the pending game asset is incomplete, an asset request is initiated, and asset completion data for the asset request is received to complete the pending game asset.

5. The game generation method according to claim 4, characterized in that: The asset completion data is generated based on the asset request by at least one of an image generation model, a video generation model, a 3D generation model, a music generation model, a special effect generation model and a multimodal generation model, and / or the asset completion data is generated using retrieval enhancement.

6. The game generation method according to claim 3, characterized in that: In response to determining that the parameters in the pending game code are complete and the pending game assets are complete, integrating the pending game code and the pending game assets to obtain the pending game content, the method further includes: In response to determining that the pending game content does not meet the game running condition, obtaining running error information; The development plan is re-disassembled based on the operation error information to determine new development subtasks.

7. The game generation method according to claim 1, characterized in that: The step of driving a game engine based on the game content data to obtain a target game comprises: sending an application request, and in response to receiving an application permission response to the application request, performing a custom parameter detection on the game content data; In response to determining that there are uncompleted custom parameters in the game content data, initiating a second data completion dialogue, and receiving second response supplementary data to the second data completion dialogue to complete the custom parameters and update the game content data; In response to determining that there are no uncompleted custom parameters in the game content, the game engine is driven based on the game content data to obtain the target game.

8. A game generating device, characterized in that: include: Interaction module, used to obtain game development requirements based on natural language interaction; A generation module, used to generate game content data based on the game development requirements; A driving module is used to drive a game engine based on the game content data to obtain a target game.

9. The game generating device according to claim 8, characterized in that: The generation module comprises: Develop an architecture agent to analyze the game development requirements and generate a development plan; A development engineer agent, used to decompose the development plan into at least one development subtask; The engine engineer agent is used to generate pending game codes corresponding to the development subtask and obtain pending game assets corresponding to the pending game codes; A review agent, used to determine whether the parameters in the pending game code are complete and whether the pending game assets are complete, and in response to determining that the parameters in the pending game code are complete and the pending game assets are complete, integrate the pending game code and the pending game assets to obtain pending game content; The test engineer agent is used to test the pending game content, and in response to determining that the pending game content meets the game running conditions, determine the pending game content as the game content data.

10. The game generating device according to claim 9, characterized in that: The engine engineer agent is obtained by fine-tuning a preset large language model with instructions using preset code corpus. The preset code corpus satisfies preset high decoupling conditions and preset high annotation conditions and corresponds to the game engine.

11. A storage medium having program data stored thereon, characterized in that: When the program data is executed by a processor, the steps of the game generation method according to any one of claims 1 to 7 are implemented.

12. A computer device, characterized in that: It comprises a processor and a memory connected to each other, the memory stores a computer program, and when the processor executes the computer program, the steps of the game generation method according to any one of claims 1 to 7 are implemented.

Images