AI animation full-process automatic generation method and system based on multi-agent cooperation

Abstract

The application provides an AI animation full-process automatic generation method and system based on multi-agent cooperation, and relates to the technical field of AI animation generation. The application generates creation intention data containing entity attributes, relationships and logical rules by receiving professional theme text data, combining a knowledge base and user annotation information. A script agent carries out multi-agent symbolic reasoning accordingly to generate scene description information with constraint conditions, entity relationships and animation driving rules. A visual agent generates a three-dimensional model and obtains initial visual data through dynamic rehearsal. An animation effect agent configures a basic motion mode for the entity to form a primary animation sequence. A synthesis agent verifies the frame-level logical consistency and generates a feedback signal to drive the visual agent and the animation effect agent to iteratively adjust, so as to output a final animation product after the consistency is met. The application can realize multi-agent cooperative AI animation full-process automatic generation and guarantee the consistency of animation logic and the efficiency of the generation process.

Description

Technical Field

[0001] This application relates to the field of AI animation generation technology, and in particular to a method and system for fully automated AI animation generation based on multi-agent collaboration. Background Technology

[0002] The AI-powered fully automated animation generation method leverages artificial intelligence to complete the entire production process from text to finished animation, significantly simplifying the complex procedures of traditional animation. This method has broad application prospects in film, games, science education, and other fields, effectively improving animation production efficiency and lowering the professional creation threshold.

[0003] Current mainstream AI animation generation methods mostly adopt a serial execution mode, with each production stage handled independently by a single module, lacking collaborative interaction and information synchronization between stages. These methods generate visual and animation content directly from text without systematically verifying the creative logic, entity relationships, and frame-level states throughout the entire process.

[0004] Independent operation of each stage can easily lead to inconsistencies in entity attributes, motion rules, and scene logic, requiring extensive manual corrections to the final animation, impacting generation efficiency and content reliability. Manual iterative adjustments also lengthen the production cycle, making it difficult to meet the logical rigor requirements of professional-themed animations. Therefore, existing technologies suffer from insufficient coordination throughout the animation generation process and difficulty in ensuring content logical consistency. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for automated generation of AI animations throughout the entire process based on multi-agent collaboration, so as to solve the problems of insufficient collaboration in the entire animation generation process and difficulty in ensuring the consistency of content logic in the existing technology.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a method for automated generation of AI animations throughout the entire process based on multi-agent collaboration, comprising:

[0007] Receive text data on specialized topics;

[0008] Based on the text data, the knowledge base is accessed and user annotation information is integrated to generate creative intent data, which includes entities and the attributes, relationships and logical rules of the entities;

[0009] Based on the creative intent data, the scriptwriting agent runs a multi-agent symbolic reasoning process to generate scene description information, which includes entities with constraints, relationships between entities, and rules that drive the animation state.

[0010] The visual intelligent agent generates a three-dimensional geometric model based on the scene description information and performs dynamic pre-simulation to form initial visual data;

[0011] Based on the initial visual data, the motion effects agent configures the basic motion patterns for the entity, forming a primary animation sequence;

[0012] The synthetic agent receives the primary animation sequence, performs logical consistency verification on the frame-level content of the primary animation sequence based on the scene description information, and generates a feedback signal according to the verification result to drive the visual agent or motion effect agent to perform iterative adjustments until the frame-level content of the primary animation sequence meets the logical consistency, and outputs the final animation product.

[0013] Optionally, the step of generating scene description information by having the screenwriter agent perform a multi-agent symbolic reasoning process based on the creative intent data includes:

[0014] The screenwriter's intelligent agent initiates multiple specialized intelligent agents to perform text parsing, knowledge concept association, and animation scene rule extraction operations in parallel on the creative intent data, and obtains the processing results of each specialized intelligent agent.

[0015] The processing results of each of the aforementioned specialized intelligent agents are interactively challenged to identify and eliminate conflicting content, and the eliminated content is jointly verified to fill in the missing content.

[0016] Extract entities from the verified and completed content and add scene creation constraints. Clarify the relationships between the entities and extract the content that guides the changes in the animation screen state as the rules that drive the animation state.

[0017] By integrating entities with constraints, the relationships between these entities, and the rules that drive the animation state, scene description information is generated.

[0018] Optionally, the step of generating a three-dimensional geometric model and performing dynamic pre-simulation based on the scene description information by the visual agent to form initial visual data includes:

[0019] The visual intelligent agent parses the scene description information, extracts entities, relationships between entities, and rules driving the animation state from the scene description information, so as to extract numerical attributes from the entities and determine geometric construction parameters;

[0020] Based on the geometric construction parameters, generate three-dimensional geometric shapes for each entity, integrate the three-dimensional geometric shapes to form a three-dimensional geometric model, and configure spatial association conditions and state change triggering conditions for the entities;

[0021] The physical simulation process is driven by the spatial association conditions and the state change triggering conditions. The three-dimensional geometric model is dynamically pre-simulated, the key information of the pre-simulation image is extracted as the initial key frame, and the view trajectory content is planned as the camera path.

[0022] Based on the three-dimensional geometric models of each entity, the initial keyframes and camera paths are integrated to form initial visual data, and the three-dimensional geometric models are output synchronously with the initial visual data as generated related content.

[0023] Optionally, the step of receiving the primary animation sequence by the synthetic agent, performing logical consistency verification on the frame-level content of the primary animation sequence based on the scene description information, and generating a feedback signal according to the verification result to drive the visual agent or motion effects agent to perform iterative adjustments until the frame-level content of the primary animation sequence satisfies logical consistency, and outputting the final animation product, includes:

[0024] The synthetic agent receives the primary animation sequence and retrieves the corresponding scene description information as the basis for frame-level logical consistency verification.

[0025] The primary animation sequence is broken down frame by frame to extract the content of each frame. The content of each frame includes entity state, inter-entity relationship information, and changes in the state of the frame.

[0026] The extracted image content is compared with the scene description information one by one to determine whether the content at each frame level meets the logical consistency requirements.

[0027] Problems are located in frame-level content that do not meet the logical consistency requirements, corresponding feedback signals are generated, and the feedback signals are sent to the corresponding visual agent or motion effect agent to drive the visual agent or motion effect agent to perform iterative adjustments.

[0028] Repeat the above steps of comparison, positioning, feedback, and adjustment. If the content of all frames meets the logical consistency requirements, the adjusted animation sequence will be integrated and optimized to generate and output the final animation product.

[0029] Optionally, the step of driving the physical simulation process based on the spatial association conditions and the state change triggering conditions to dynamically pre-simulate the three-dimensional geometric model includes:

[0030] The visual intelligent agent transforms the spatial association conditions and the state change triggering conditions into executable parameters for physical simulation and divides them into pre-simulation stages, which include entity individual pre-simulation stage, entity association pre-simulation stage and full scene integration pre-simulation stage.

[0031] During the individual entity pre-play phase, it is verified whether the state changes of a single entity meet the state change triggering conditions and its own constraints.

[0032] During the entity association pre-simulation phase, the linkage of state changes of associated entities is verified to ensure consistency with the entity relationships in the scenario description information.

[0033] During the full-scene integrated pre-rehearsal phase, the pre-rehearsal rate is controlled in conjunction with the initial keyframe extraction requirements, and entity state anomalies and spatial position deviations are captured in real time during the pre-rehearsal process.

[0034] Optionally, the step of configuring basic motion patterns for the entity based on the initial visual data by the motion effects agent to form a primary animation sequence includes:

[0035] The initial visual data is parsed by the motion effects intelligent agent to extract entities, initial keyframes, and camera paths;

[0036] Based on the attributes and constraints corresponding to the entities, the motion mode is matched and adapted, and the adapted motion mode is determined as the basic motion mode of each entity and the configuration is completed.

[0037] The basic motion pattern is temporally correlated with the initial keyframe, and the perspective of the physical motion image is adjusted in conjunction with the camera path.

[0038] The adjusted motion frames of each entity are sorted and integrated according to time nodes to generate a primary animation sequence.

[0039] Optionally, the step of accessing a knowledge base and integrating user annotation information based on the text data to generate creative intent data includes:

[0040] Feature extraction is performed on the text data and a retrieval identifier is constructed. Based on the retrieval identifier, the knowledge base is searched layer by layer to obtain knowledge base association information.

[0041] The user-annotated information is classified and extracted to obtain annotation feature information and user-specified association requirements. The knowledge base association information is matched and fused with the annotation feature information, and the fused information is associated with content according to the association requirements.

[0042] Extract entities from the information after content association and assign attributes, sort out the relationships between the entities, and extract the behavioral constraints that the entities must follow as logical rules;

[0043] The entities, their attributes, relationships, and logical rules are integrated to generate creative intent data.

[0044] Secondly, this application provides an AI animation full-process automated generation system based on multi-agent collaboration, including:

[0045] The receiving module is used to receive text data on professional topics;

[0046] The first generation module is used to access the knowledge base and integrate user annotation information based on the text data to generate creative intent data, wherein the creative intent data includes entities and the attributes, relationships and logical rules of the entities;

[0047] The second generation module is used by the screenwriter agent to run a multi-agent symbolic reasoning process based on the creative intent data to generate scene description information, which includes entities with constraints, relationships between entities, and rules that drive the animation state.

[0048] The first forming module is used by the visual intelligent agent to generate a three-dimensional geometric model and perform dynamic pre-simulation based on the scene description information to form initial visual data;

[0049] The second forming module is used by the motion effects agent to configure basic motion patterns for the entity based on the initial visual data, thereby forming a primary animation sequence;

[0050] The output module is used to receive the primary animation sequence by the synthetic agent, perform logical consistency verification on the frame-level content of the primary animation sequence based on the scene description information, and generate a feedback signal according to the verification result to drive the visual agent or motion effect agent to perform iterative adjustments until the frame-level content of the primary animation sequence meets the logical consistency, and output the final animation product.

[0051] Thirdly, this application provides an electronic device, comprising:

[0052] Memory, used to store computer programs;

[0053] A processor, configured to execute the computer program to implement the steps of the AI ​​animation full-process automated generation method based on multi-agent collaboration as described in the first aspect above.

[0054] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the AI ​​animation full-process automated generation method based on multi-agent collaboration as described in the first aspect above.

[0055] The AI ​​animation full-process automated generation method based on multi-agent collaboration provided in this application can clarify the basic materials and thematic direction of animation creation by receiving professional subject text data. Combined with knowledge base and user annotation to generate creative intention data, it can accurately extract core creative elements and standardize basic logic. The screenwriting agent generates scene description information through multi-agent symbolic reasoning, which can structure and clarify animation production constraints and execution rules. The visual agent generates a 3D model and dynamically pre-plays it, which can lay the visual foundation of the animation. The motion effect agent configures motion modes to form preliminary animation content. The synthesis agent conducts frame-level logical consistency verification and feedback iteration, which can ensure the logical unity of the animation content and finally output a compliant and reliable animation product.

[0056] Furthermore, the scriptwriting agent initiates multiple specialized agents to perform parallel operations on the creative intent data, including text parsing, knowledge concept association, and animation scene rule extraction. The results of each process interact and challenge each other to eliminate conflicting content, then jointly verify and fill in missing information. Subsequently, entities are extracted, scene creation constraints are added, entity relationships are clarified, and animation state-driven rules are refined. Finally, all of the above content is integrated to generate scene description information. This step improves the efficiency of scene description generation through parallel processing by multiple specialized agents, and eliminates content conflicts and fills information gaps through interactive challenges and joint verification. This makes the entity constraints, entity relationships, and animation-driven rules in the scene description more accurate and complete, providing a more standardized and logically rigorous scene basis for subsequent animation production stages. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 A flowchart illustrating a fully automated AI animation generation method based on multi-agent collaboration, provided in an embodiment of this application;

[0059] Figure 2 A flowchart illustrating another AI animation fully automated generation method based on multi-agent collaboration provided in this application embodiment;

[0060] Figure 3 This is a schematic diagram of the structure of an AI animation full-process automated generation system based on multi-agent collaboration, provided in an embodiment of this application. Detailed Implementation

[0061] Current AI animation full-process automated generation mostly adopts a modular independent serial processing mode. The production stages lack effective collaboration and information exchange, and no full-process logic verification and iterative optimization mechanism has been built. This easily leads to problems such as inconsistencies in entity attributes, motion states, and scene logic. The consistency of animation frame-level content is difficult to guarantee, and a lot of manual correction and adjustment are required. The overall generation efficiency is low, the quality of professional content cannot be consistently met, and it is difficult to adapt to the creative requirements of professional-themed animation.

[0062] To address the aforementioned issues, this application proposes a fully automated AI animation generation method based on multi-agent collaboration. This method involves the collaborative work of scriptwriters, visual designers, motion designers, and compositing agents, who sequentially complete creative intent analysis, scene reasoning, 3D modeling, motion effect generation, and logic verification. A feedback iterative closed loop is formed based on frame-level consistency verification by the compositing agent. This solution replaces independent, segmented processing with multi-agent collaboration, combining a knowledge base and user-annotated specifications to standardize creative logic. Through full-process linkage and closed-loop optimization, it effectively eliminates content logic conflicts, ensures consistency of animation entities, movements, and scenes, reduces human intervention, and fundamentally solves the shortcomings of existing animation generation technologies, such as insufficient collaboration and difficulty in ensuring content logic consistency. This improves the efficiency and quality of automated professional animation generation.

[0063] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0064] The core of this application is to provide a fully automated AI animation generation method based on multi-agent collaboration, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0065] S101, Receive text data on professional topics.

[0066] Among them, professional-themed text data refers to text materials related to specific professional fields and used for animation creation, including the core content, related descriptions and creative requirements of that field.

[0067] In one specific implementation, text data that meets the requirements of the professional field can be acquired through various conventional methods, such as receiving document files uploaded by users, text content manually entered in a designated input interface, or importing the required data by calling external professional text libraries or relevant knowledge bases via API calls. This acquired professional-themed text data will serve as the basic material for the entire process of subsequent automated animation generation, providing the original basis for refining the creative core and building animation scenes.

[0068] S102. Based on the text data, access the knowledge base and integrate user annotation information to generate creative intent data.

[0069] The creative intent data serves as the core guiding information for adapting AI animation creation. It includes the entities to be presented in the animation, the specific attributes of each entity, the relationships between entities, and the logical rules that entity behaviors must follow, providing a clear basis for subsequent animation scene generation. The knowledge base is a database storing knowledge related to the specific subject matter, containing basic concepts and entity attributes to supplement the professional details of the text data. User annotation information consists of annotations added to the text data by users based on their creative needs, clarifying the creative focus.

[0070] S102 specifically includes:

[0071] S1021. Extract features from the text data and construct a retrieval identifier. Based on the retrieval identifier, perform a layer-by-layer retrieval of the knowledge base to obtain knowledge base association information.

[0072] The retrieval identifier is a combination of search keywords composed of the core features of the text, used to accurately match knowledge base content. Knowledge base related information is professional knowledge retrieved from the knowledge base that is relevant to the text, used to supplement the text with professional details.

[0073] S1022. Classify and extract the user annotation information to obtain annotation feature information and user-specified association requirements, match and merge the knowledge base association information with the annotation feature information, and perform content association on the merged information according to the association requirements.

[0074] Among these, the annotation feature information is the core content extracted from user annotations, including creative focus and specific entity requirements. Association requirements are the user-defined rules for linking various pieces of information, used to standardize the information fusion logic.

[0075] S1023. Extract entities from the information after content association and assign attributes, sort out the relationships between the entities, and extract the behavioral constraints that the entities must follow as logical rules.

[0076] In this context, an entity is a concrete object that needs to be presented in the animation, such as items or scene elements. Entity attributes are its specific characteristics, such as size and parameters. Logical rules are the criteria that constrain the behavior of entities, ensuring that the animation conforms to professional logic.

[0077] S1024. Integrate the entity and its attributes, relationships between entities, and logical rules to generate creative intent data.

[0078] In one specific implementation, this step generates creative intent data through a process of retrieval, supplementation, fusion matching, and extraction and organization. The overall process first supplements professional information, then integrates user needs, and finally organizes the data into standardized core information. Specifically, in step S1021, the TF-IDF algorithm is used to extract core text features. This algorithm calculates the frequency and domain scarcity of keywords in the text, filters key information, constructs retrieval identifiers, and then performs layer-by-layer matching retrieval from the knowledge base. First, it matches the basic information of core keywords, then supplements details to obtain knowledge base-related information. Next, in step S1022, user-annotated information is categorized and extracted. A feature matching algorithm is used to match and fuse knowledge base-related information with annotation feature information, eliminating irrelevant content, and then organizing the information coherence according to user association requirements. Then, in step S1023, entities are selected from the related information, attributes are assigned based on the knowledge base and user annotations, relationships between entities are organized, and logical rules that meet professional requirements are extracted. Finally, in step S1024, the above content is integrated to form logically clear creative intent data.

[0079] As an example, assuming the professional text is "a mechanical demonstration animation of measuring the weight of a hook using a spring balance in junior high school physics classroom teaching", in step S1021, the text is first subjected to keyword and semantic feature extraction, and core features such as spring balance, hook, gravity measurement, reading change, and mechanical experiment are selected. Based on these, a retrieval identifier containing topic tags and entity names is constructed. Then, according to the hierarchical order of topic, entity, and parameter, the physics professional knowledge base is searched layer by layer. First, the major category of mechanical experiments is matched, and then the relevant entries of measuring tools and experimental equipment are located. Finally, the relevant knowledge information such as the spring balance range of 0-5N, the scale division value of 0.2N, the standard mass of a single hook is 50g, the corresponding gravity is about 0.5N, and the experimental vertical suspension operation specifications are obtained.

[0080] Secondly, in step S1022, the user-annotated content "highlighting the real-time changes in the spring balance reading, keeping the hook weight vertically suspended, and the experimental data conforming to physical laws" is classified and extracted. The annotation feature information and the user-specified association requirements are separated. The annotation feature information is the core demonstration object and the screen presentation requirements. The association requirements are that the reading and the weight of the hook weight must strictly correspond and the operating posture must conform to the experimental specifications. Then, the tool parameters and equipment specifications in the knowledge base are matched and integrated with the annotation features one by one. Irrelevant extended experimental content is eliminated. According to the association requirements, the weight weight, the weight value and the spring balance reading are precisely logically bound to form a coherent and unified integrated information.

[0081] Then, in step S1023, the two core animation entities, spring balance and weight, are extracted from the fusion information that has been associated. Specific attributes are assigned to each of them. The attributes of the spring balance include measurement range, scale division, and appearance. The attributes of the weight include single weight mass, standard gravity, and shape specifications. At the same time, the spatial relationship of the weight suspended below the hook of the spring balance is sorted out. Furthermore, the experimental operation specifications, linear change of reading with gravity, and vertical suspension without tilting are extracted as the logical rules that the animation must follow.

[0082] Finally, in step S1024, the above entities, corresponding attributes, relationships between entities, and logical rules are structurally integrated according to the animation creation requirements to form complete creative intent data, providing a clear core basis for subsequent scene generation.

[0083] This application, through the above steps, transforms fragmented professional text data into complete and standardized core creative information, supplements professional details, and caters to user needs. It solves the problem that the original text is difficult to use directly for animation creation, and provides accurate and reliable guidance for subsequent animation scene generation, balancing professionalism and personalization.

[0084] S103. The screenwriter agent runs a multi-agent symbolic reasoning process based on the creative intent data to generate scene description information.

[0085] Among them, the scriptwriting agent is the core agent responsible for animation scene planning and logic organization, and is responsible for coordinating the collaborative work of multiple agents. Multi-agent symbolic reasoning is the process of transforming creative intentions into structured, reasonable scene rules, and completing logical deduction and content standardization through the collaboration of multiple agents. Scene description information is the structured execution basis for animation production, including entities with constraints, relationships between entities, and rules that drive the animation state, used to guide subsequent visual modeling and motion effect generation.

[0086] S103 specifically includes, such as Figure 2 As shown:

[0087] S1031. The screenwriter agent initiates multiple specialized agents to perform text parsing, knowledge concept association, and animation scene rule extraction operations in parallel on the creative intent data, and obtains the processing results of each specialized agent.

[0088] Among them, the specialized intelligent agents are sub-intelligent agents that undertake a single specialized task, respectively responsible for text parsing, knowledge association, and rule extraction, and improve the efficiency of scene generation through parallel processing.

[0089] S1032. The processing results of each of the division of labor intelligent agents are interactively challenged to identify and remove conflicting parts of the content, and the removed content is jointly verified to fill in the missing content.

[0090] Interactive adversarial interaction involves multiple specialized intelligent agents mutually verifying each other's outputs and identifying contradictory information through logical comparison. Joint verification involves conducting a completeness check on conflict-free content and supplementing missing scene elements and logical information.

[0091] S1033. Extract entities from the verified and completed content and add scene creation constraints, sort out the relationships between the entities, and extract the content that guides the changes in the animation screen state as the rules that drive the animation state.

[0092] Among these, scene creation constraints are the production limitations that entities must adhere to within the scene, ensuring that the animation meets professional standards and creative requirements. The rules driving the animation state are the core basis for controlling the changes in entity form, position, and state, determining the dynamic change logic of the animation.

[0093] S1034. Integrate the entities with constraints, the relationships between the entities, and the rules that drive the animation state to generate scene description information.

[0094] In one specific implementation, this step transforms creative intent into standardized scene description information through a process of multi-agent parallel processing, interactive verification, rule extraction, and structured integration. The overall process begins with specialized agents processing basic information in parallel, then eliminating conflicts and supplementing content through interactive adversarial processes. Subsequently, constrained entities, relationships, and driving rules are extracted, and finally, a complete output is formed.

[0095] As an example, in a classroom demonstration animation scene using a spring balance to measure the weight of a hook in junior high school physics, step S1031 first involves the scriptwriting agent acting as the core scheduling unit, simultaneously activating three types of specialized agents: a text parsing agent, a knowledge concept association agent, and an animation scene rule extraction agent, to achieve parallel processing. The text parsing agent performs semantic decomposition and structured analysis on the creative intent data, accurately extracting core entities such as the spring balance, the hook, and the experimental table, as well as entity attributes and basic behavioral logic. The knowledge concept association agent matches professional knowledge such as the operating procedures, tool usage standards, and physical parameter correspondences of junior high school physics mechanics experiments based on the professional knowledge base. The animation scene rule extraction agent combines the creative intent with professional knowledge to sort out the process logic, key points of screen presentation, and state change nodes of the animation demonstration. The three types of specialized agents operate independently and output their respective processing results synchronously, without interfering with each other and improving overall processing efficiency.

[0096] Secondly, in step S1032, the output results of the three types of intelligent agents are summarized and interactive adversarial verification is initiated. Each intelligent agent judges the rationality of the output content of other intelligent agents based on its own processing logic. Conflicting content is identified through logic matching and rule comparison. For example, some intelligent agents output content that does not conform to experimental specifications and physical logic, such as the spring balance being placed horizontally, the number of hooks being suspended exceeding the range, and reading the reading in a non-static state. The system automatically marks and removes such conflicting content. After removing conflicting content, the system performs joint verification by multiple intelligent agents to check the completeness of scene elements and supplement missing content such as experimental scene background, operation safety boundaries, reading change triggering conditions, and entity position references. This forms intermediate processing content without logical contradictions and with complete information, laying the foundation for subsequent rule extraction.

[0097] Then, through step S1033, four core animation entities—spring balance, single hook, experimental workbench, and background blackboard—are stably extracted from the complete content after verification and completion. Clear scene creation constraints are added to each entity, including spring balance range constraints, vertical placement constraints, static reading constraints, upper limit constraints on the number of hooks, vertical suspension constraints, and workbench size and position constraints. Simultaneously, the fixed relationships between entities are clarified, including hooks suspended from the lower hooks of the spring balance, the spring balance placed vertically above the workbench, and the blackboard as a background located behind the workbench. Furthermore, the core rules controlling the changes in the animation screen state are refined, including a linear increase in the spring balance reading when the number of hooks increases, a stable reading after the hooks come to rest, and a synchronous dynamic change in the reading when the suspension state changes. These serve as the core basis for driving animation state switching and screen updates.

[0098] Finally, through step S1034, the entity information with constraints, the spatial and relational relationships between entities, and the animation state-driven rules are classified and integrated according to the standardized format of animation production. The priority and execution logic of each content are clarified, forming scene description information with clear structure, complete elements, and clear constraints. This directly provides standardized guidance for subsequent visual intelligent agent modeling and motion logic design of motion effect intelligent agents.

[0099] The model used by the specialized intelligent agents is based on a general semantic understanding model and is fine-tuned and trained on datasets of professional science and education animations and experimental demonstrations. It learns the corresponding capabilities of text parsing, knowledge association, and scene rule extraction. The model training data includes professional text, animation scene specifications, entity constraint rules, and other content. Parallel processing employs a task scheduling algorithm to synchronously allocate the three types of processing tasks to the corresponding specialized intelligent agents, improving processing efficiency.

[0100] This application extracts constrained and logically complete scene description information through multi-agent parallel collaboration and interactive verification, providing a unified standard for subsequent visual and motion effect production, avoiding scene logic conflicts, and improving the accuracy and standardization of animation scene planning.

[0101] S104. The visual agent generates a three-dimensional geometric model based on the scene description information and performs dynamic pre-simulation to form initial visual data.

[0102] The visual agent is responsible for transforming structured scene descriptions into visual content. Its core functions are 3D modeling and dynamic pre-visualization, bridging the scene planning and motion design stages. The 3D geometric model is a digital 3D model simulating the shape, size, and structure of animated entities, used to intuitively present their forms. Dynamic pre-visualization is the process of simulating and demonstrating the motion state and scene presentation effects of the 3D geometric model in advance, used to verify the model's compatibility with scene requirements. Initial visual data is the basic visual data containing the 3D geometric model, key pre-visualization shots, and viewpoint paths, providing an intuitive visual foundation for subsequent motion design.

[0103] S104 specifically includes:

[0104] S1041. The visual intelligent agent parses the scene description information, extracts entities, relationships between entities, and rules driving the animation state from the scene description information, so as to extract numerical attributes from the entities and determine geometric construction parameters.

[0105] Among them, geometric construction parameters are the core parameters used to generate three-dimensional geometric models. These parameters include numerical information such as the size, shape, proportion, and structural details of the entity. They are derived from the numerical attributes of the entity and directly determine the shape and specifications of the three-dimensional model.

[0106] S1042. Generate three-dimensional geometric shapes for each entity based on the geometric construction parameters, integrate the three-dimensional geometric shapes to form a three-dimensional geometric model, and configure spatial association conditions and state change triggering conditions for the entities.

[0107] Spatial association conditions are rules that define the relative positions and orientations of entities in a 3D scene, ensuring that the spatial relationships between entities conform to the scene description requirements. State change trigger conditions are the prerequisites for triggering changes in the form, position, and state of entities. They correspond to the rules that drive the animation state and are used to control the state switching in dynamic pre-show.

[0108] S1043. Based on the spatial association conditions and the state change triggering conditions, drive the physical simulation process, perform dynamic pre-playing on the three-dimensional geometric model, extract key information from the pre-playing images as initial keyframes, and plan the view trajectory content as the camera path.

[0109] S1043 specifically includes:

[0110] The visual agent transforms the spatial association conditions and the state change triggering conditions into executable parameters for physical simulation and divides the simulation into pre-simulation stages. These pre-simulation stages include individual entity pre-simulation, entity association pre-simulation, and full-scene integrated pre-simulation. In the individual entity pre-simulation stage, it verifies whether the state changes of a single entity conform to the state change triggering conditions and its own constraints. In the entity association pre-simulation stage, it verifies the linkage of state changes of associated entities to ensure consistency with the entity relationships in the scene description information. In the full-scene integrated pre-simulation stage, it controls the pre-simulation rate based on the initial keyframe extraction requirements and captures entity state anomalies and spatial position deviations in real time during the pre-simulation process.

[0111] The physical simulation process simulates real-world physical laws, ensuring that the movement and state changes of the 3D geometric model conform to natural logic and scene rules. Initial keyframes are frames that capture the core state of entities and key scenes during the dynamic pre-show, forming the foundation for subsequent animation sequences. Camera paths plan the camera's shooting angle and movement trajectory during dynamic pre-show and subsequent animation presentation, controlling the presentation angle and switching logic of the animation footage. The pre-show phase is a step-by-step demonstration to ensure comprehensiveness, progressively verifying the model's compatibility with scene requirements through layered pre-shows, preventing omissions or errors in the overall pre-show.

[0112] S1044. Based on the three-dimensional geometric model of each entity, the initial keyframes and camera paths are integrated to form initial visual data.

[0113] The three-dimensional geometric model is output synchronously with the initial visual data as part of the generated related content.

[0114] In one specific implementation, this step is coordinated by a visual intelligent agent, following the process of parsing and extracting, modeling and configuring, hierarchical pre-playing, and integrated output. First, the scene description information is parsed, entities, relationships and state rules are extracted and geometric construction parameters are determined. Then, based on the parameters, the three-dimensional geometric shapes of each entity are generated and integrated into a three-dimensional geometric model. Spatial association and state triggering conditions are configured. Subsequently, the conditions are converted into physical simulation parameters. Dynamic pre-playing is carried out in three stages: individual entities, entity association, and full scene integration. Initial keyframes are extracted and camera paths are planned. Finally, the three-dimensional model, keyframes and camera paths are integrated to form and output the initial visual data.

[0115] As an example, using the experimental animation scene of measuring the weight of a hook with a spring balance from junior high school physics, the visual agent first analyzes the scene description information in step S1041, extracting four entities: the spring balance, the hook, the experimental workbench, and the background blackboard. The spatial relationships and animation state driving rules between the entities are then sorted out. Numerical information is extracted from the entity attributes to determine the geometric construction parameters. The parameters of the spring balance are: length 15cm, range 0-5N, and scale division 0.2N; the parameters of the hook are: diameter 2cm and single weight 50g; the parameters of the experimental workbench are: length 60cm, width 40cm, and height 80cm; and the parameters of the background blackboard are: length 120cm and width 80cm.

[0116] Next, in step S1042, based on the above geometric construction parameters, a three-dimensional modeling algorithm is used to generate the three-dimensional geometric shape of each entity. The spring balance, weight, experimental operating table, and background blackboard are integrated according to the spatial relationship described in the scene to form a complete three-dimensional geometric model of the experimental scene. At the same time, spatial association conditions and state change triggering conditions are configured for each entity. The spatial association conditions include the spring balance being placed vertically on the left side of the experimental operating table, the weight being suspended from the hook at the lower end of the spring balance, and the background blackboard being located behind the operating table. The state change triggering conditions include the spring balance reading changing synchronously when the weight is suspended, and the reading remaining stable after the weight stops.

[0117] Then, through step S1043, the visual agent transforms the above conditions into executable parameters for physical simulation, dividing them into three pre-playing stages: the entity-individual pre-playing stage, which simulates the range of changes in the spring balance reading and the suspension state of the hook, verifying that the state changes of a single entity conform to its own constraints and triggering conditions; the entity-associated pre-playing stage, which simulates the linkage changes between the hook and the spring balance after the hook is suspended, verifying the rationality of the linkage between the hook's gravity and the spring balance reading, ensuring no logical deviation; and the full-scene integrated pre-playing stage, which controls the pre-playing rate to 10 frames per second, adapting to the initial keyframe extraction requirements, capturing in real time whether the spring balance is tilted or the hook deviates from its suspension position during the pre-playing process, extracting three core images as initial keyframes: the moment the hook is suspended, when the reading is stable, and when the hook is removed, and planning camera paths from a frontal view of the experimental operating table, a side view of the spring balance reading, and an overall overview of the experimental scene.

[0118] Finally, in step S1044, based on the three-dimensional geometric model of the experimental scene, the extracted initial keyframes are integrated with the planned camera path to form initial visual data containing the three-dimensional model, key images, and view trajectory. At the same time, the three-dimensional geometric model is output synchronously to provide a visual basis for configuring the motion mode of the subsequent motion effect intelligent agent.

[0119] Through the above steps, this application transforms structured scene description information into concrete and standardized initial visual data, ensuring that the 3D model conforms to entity attributes and scene constraints, dynamically pre-visualizes scene rules, provides a reliable visual foundation for subsequent motion effect configuration and animation optimization, reduces adjustment costs in subsequent stages, and ensures the professionalism and rationality of the animation visual presentation.

[0120] S105. Based on the initial visual data, the motion effects agent configures a basic motion pattern for the entity to form a primary animation sequence.

[0121] Among them, the motion effects agent is responsible for configuring motion logic for animated entities and generating dynamic images. Its core function is to transform static 3D models into animated content with motion states. The basic motion mode is the basic motion method that adapts to the attributes of each entity and scene constraints, conforming to the characteristics of the entity and the needs of animation demonstration, and is the core logic of entity motion. The primary animation sequence is a preliminary collection of dynamic images containing the basic motions of each entity and conforming to the viewpoint path, and is the basic prototype of the finished animation.

[0122] S105 specifically includes:

[0123] S1051 The motion effects agent parses the initial visual data and extracts entities, initial keyframes, and camera paths.

[0124] S1052. Match and adapt the motion mode based on the attributes and constraints corresponding to the entity, determine the adapted motion mode as the basic motion mode of each entity, and complete the configuration.

[0125] Among them, the adaptive motion mode is a motion mode that is selected by combining entity attributes and scene constraints and conforms to the characteristics of the entity and the requirements of the scene, so as to ensure that the motion logic conforms to professional standards and creative needs.

[0126] S1053. The basic motion mode is temporally correlated with the initial keyframe, and the perspective of the physical motion image is adjusted in conjunction with the camera path.

[0127] Among these, temporal correlation maps the basic motion pattern of an entity to specific time points along with the initial keyframes, ensuring that the motion rhythm is synchronized with the key shots. Viewpoint adaptation adjustment adjusts the presentation angle of the entity's motion image based on the camera's shooting angle and movement trajectory, ensuring that the motion image fits the changes in viewpoint, resulting in a smooth and natural presentation.

[0128] S1054. Sort and integrate the adjusted motion frames of each entity according to time nodes to generate a primary animation sequence.

[0129] In one specific implementation, this step is coordinated by the motion effects agent and follows the process of parsing and extracting, matching and configuring, associating and adjusting, and sorting and integrating. First, the initial visual data is parsed to extract the core elements. Then, the basic motion modes that are matched and configured for each entity are matched and configured. The motion modes are temporally correlated with key frames and the perspective is adjusted in combination with the camera path. Finally, the primary animation sequence is generated by sorting and integrating according to time, ensuring that the motion logic and visual presentation meet the requirements of the scene.

[0130] The motion effects agent uses a model based on a general motion effects configuration model, fine-tuned and trained on professional animation motion datasets and entity motion constraint cases. It focuses on learning the matching logic between entity attributes, constraints, and motion patterns, as well as methods for temporal correlation and viewpoint adaptation, ensuring that the configured motion patterns accurately adapt to the scene and entity characteristics. Motion pattern matching employs a feature matching algorithm, comparing entity attributes and constraints with features in a pre-set motion pattern library to select the optimal motion pattern. Temporal correlation uses a timeline mapping algorithm to precisely map keyframes to motion stages, ensuring a smooth motion rhythm.

[0131] As an example, using the experimental animation scene of measuring the weight of a hook with a spring balance in junior high school physics, the motion effect agent first parses the initial visual data in step S1051, extracting four entities: spring balance, hook, experimental operating table, and background blackboard, as well as three initial keyframes: the moment the hook is suspended, when the reading is stable, and when the hook is removed, and three preset camera paths: front, side, and overhead.

[0132] Next, through step S1052, the motion modes are matched and adapted by combining the attributes and constraints of each entity. The spring balance is configured with a stretching and contracting motion mode to conform to the range constraint, and the hook is configured with a vertical lifting motion mode to conform to the vertical suspension constraint. The experimental operating table and the background blackboard do not need to move, so a static mode is configured, thus completing the configuration of the basic motion modes of each entity.

[0133] Then, through step S1053, the basic motion mode is temporally correlated with the initial keyframe. The spring force gauge starts to stretch when the hook is suspended, the spring force gauge is stretched to the end and the reading is stable when the hook is stationary, and the spring force gauge retracts and resets when the hook is removed. At the same time, the presentation angle of each entity's motion image is adjusted in combination with the three camera paths to ensure that the stretching changes of the spring force gauge and the suspension state of the hook can be clearly seen from different perspectives.

[0134] Finally, in step S1054, the adjusted motion images of each entity are sorted sequentially according to time nodes and integrated to form a preliminary dynamic image containing the lifting and lowering of the hook and the stretching and contraction of the spring balance, i.e., the primary animation sequence.

[0135] This application adds motion logic that fits the characteristics of the entity and the requirements of the scene to static visual data through the above steps, generates a coherent and standardized primary animation sequence, ensures that the motion mode is adapted and the perspective is presented reasonably, provides a foundation for subsequent synthesis verification and iterative optimization, and reduces the adjustment cost of the finished animation.

[0136] S106. The synthetic agent receives the primary animation sequence, performs logical consistency verification on the frame-level content of the primary animation sequence based on the scene description information, and generates a feedback signal according to the verification result to drive the visual agent or motion effect agent to perform iterative adjustments until the frame-level content of the primary animation sequence meets the logical consistency, and outputs the final animation product.

[0137] The synthetic agent is the core agent responsible for logically verifying the initial animation sequence, driving iterative optimization, and outputting the final product. It coordinates the final verification and optimization stages of animation generation. Frame-level logical consistency verification involves checking each frame of the initial animation sequence to ensure it conforms to entity constraints, entity relationships, and animation-driving rules in the scene description information, guaranteeing the logical rigor of every frame. Feedback signals are instructions generated by the synthetic agent in response to detected logical problems, clearly informing the visual or motion effects agent of the problem and the direction for adjustment. The final animation product is a complete animation that has undergone multiple rounds of iterative verification, meets frame-level logical consistency requirements, and aligns with professional subject matter needs and creative intent.

[0138] S106 specifically includes:

[0139] S1061. The synthetic agent receives the primary animation sequence and retrieves the corresponding scene description information as the basis for frame-level logical consistency verification.

[0140] S1062. The primary animation sequence is decomposed frame by frame to extract the content of each frame. The content of each frame includes the entity state, the relationship information between entities, and the content of changes in the state of the frame.

[0141] Frame-by-frame decomposition is the process of breaking down the initial animation sequence into individual frames in chronological order, used to achieve fine-grained verification of the animation content. The content of each frame contains all the core information presented in the animation, directly reflecting the state, relationships, and dynamic changes of entities within that frame, and is the core object for logical consistency verification.

[0142] S1063. The extracted image content is compared with the scene description information in turn to determine whether the content of each frame meets the logical consistency requirements.

[0143] The process of sequential comparison involves comparing each piece of information in a single frame with the corresponding entity constraints, entity relationships, and animation-driven rules in the scene description information, ensuring that no logical verification point is missed. Logical consistency requirements encompass all specifications explicitly stated in the scene description information, including entity attribute constraints, spatial relationships, and motion rules; these are the core criteria for determining whether frame-level content is acceptable.

[0144] S1064. Locate the problem in the frame-level content that does not meet the logical consistency requirements, generate the corresponding feedback signal, and send the feedback signal to the corresponding visual agent or motion agent to drive the visual agent or motion agent to perform iterative adjustments.

[0145] Problem localization is the process of identifying frame-level content that does not meet logical requirements, the specific problem type, and the root cause, which is used to accurately guide subsequent adjustment operations. Iterative adjustment is the process by which the visual agent or motion agent modifies its own output content (3D model or motion pattern) based on feedback signals to resolve logical conflicts until the corresponding frame-level content meets the requirements.

[0146] S1065. Repeat the above steps of comparison, positioning, feedback, and adjustment. If the content of all frames meets the logical consistency requirements, integrate and optimize the adjusted animation sequence to generate and output the final animation product.

[0147] Integration and optimization involves re-integrating all frame-level content after multiple rounds of iterative adjustments in chronological order, while optimizing details such as screen continuity and perspective transitions to ensure a smooth and natural overall animation presentation.

[0148] In one specific implementation, this step is coordinated by a synthetic agent and follows a process of receiving and retrieving, frame-by-frame decomposition, cross-directional comparison, problem feedback, and iterative output. First, the initial animation sequence is received and scene description information is retrieved as the basis for verification. Then, the animation is decomposed to extract the content of each frame and compare it with the scene description to determine the consistency of the logic. For unqualified content, the problem is located and a feedback signal is sent to drive the corresponding agent to iteratively adjust. The above process is repeated until all frames meet the requirements. Finally, the final animation product is integrated, optimized, and output.

[0149] The synthetic agent's model is based on a general logic verification model, fine-tuned and trained on professional animation logic verification cases, scene rule datasets, and frame-level problem localization samples. It focuses on learning the comparison logic between frame-level content and scene description, problem localization methods, and feedback signal generation rules to ensure accurate verification and localization. Frame-by-frame decomposition uses a frame parsing algorithm to break down the primary animation sequence sequentially along the timeline, efficiently extracting the core content of each frame. Progressive comparison uses a feature comparison algorithm to match the image content with the corresponding features of the scene description one by one to determine whether it meets the logical consistency requirements.

[0150] In another specific implementation, the verification accuracy can be adjusted according to the complexity of the animation. For simple professional scenarios, a combination of sampling verification and frame-by-frame verification can be used to improve efficiency. For complex scenarios, multiple rounds of cross-comparison can be added, with the synthetic agent linking with the scripting agent to assist in verification, further improving the accuracy of logical consistency judgment.

[0151] As an example, using the experimental animation scene of measuring the weight of a hook with a spring balance from junior high school physics, the process begins with step S1061. The synthesized agent receives a primary animation sequence containing the lifting and lowering of the hook and the stretching and contraction of the spring balance, and retrieves the corresponding scene description information, which serves as the core basis for frame-level logical consistency verification. Next, in step S1062, a frame parsing algorithm is used to decompose the primary animation sequence frame by frame, separating each independent frame in chronological order. The content of each frame is extracted, including the stretching state of the spring balance, changes in the reading, the suspension position of the hook, the spatial relationship between entities, and the dynamic changes in the state of the entities within the frame.

[0152] Then, in step S1063, a feature comparison algorithm is used to compare the extracted image content of each frame with the scene description information one by one, and to determine whether each frame meets the logical consistency requirements. For example, it verifies whether the spring balance is always placed vertically, whether the change in the reading matches the number of hooks, and whether the hooks are always suspended on the hook of the spring balance, so as to ensure that the content of each frame meets the constraints and rules of the scene description.

[0153] Next, in step S1064, the frame-level content that does not meet the logical consistency requirements is compared and the problem is located. For example, specific problems such as the spring balance tilting in a few frames, the hook's hanging position deviating from the hook in a certain frame, and the reading and the weight of the hook not matching in a certain frame are located. Corresponding feedback signals are generated for different problems. If it is a problem of spring tilting or position deviation, the feedback signal is sent to the visual agent to guide it to adjust the spatial position of the 3D model. If it is a problem of mismatch between the reading and the weight of the hook, the feedback signal is sent to the motion effect agent to guide it to adjust the motion mode and the logic of the reading linkage, driving the corresponding agent to perform iterative adjustments.

[0154] Finally, through step S1065, the above process of comparison, positioning, feedback, and adjustment is repeated. After each round of adjustment, the synthetic agent re-verifies the adjusted animation sequence frame by frame until the content of all frames meets the logical consistency requirements. Then, all the adjusted frame-level images are re-integrated in chronological order, optimizing details such as the smoothness of image transitions and the naturalness of perspective switching, generating and outputting a complete, standardized, and logically rigorous junior high school physics mechanics experiment demonstration animation. The above example is only one example of this application. In practical applications, it can be set according to needs, and this application does not limit it.

[0155] This application performs refined frame-level logical verification on the primary animation sequence through the above steps, accurately locates and resolves logical conflicts, and ensures that the content of each frame of the animation meets the scene requirements through iterative optimization. Finally, it outputs a logically rigorous, professionally compliant, and smoothly presented animation product, thus guaranteeing the quality of animation generation.

[0156] Figure 3 This application provides a schematic diagram of a specific implementation of an AI animation full-process automated generation system based on multi-agent collaboration, as illustrated in the embodiments of this application. Figure 3 The system may include:

[0157] The receiving module 31 is used to receive text data on professional topics;

[0158] The first generation module 32 is used to access the knowledge base and integrate user annotation information based on the text data to generate creative intent data, wherein the creative intent data includes entities and the attributes, relationships and logical rules of the entities;

[0159] The second generation module 33 is used by the screenwriter agent to run a multi-agent symbolic reasoning process based on the creative intent data to generate scene description information, which includes entities with constraints, relationships between entities, and rules that drive the animation state.

[0160] The first forming module 34 is used to generate a three-dimensional geometric model and perform dynamic pre-simulation based on the scene description information by the visual intelligent agent to form initial visual data.

[0161] The second forming module 35 is used by the motion effects agent to configure basic motion patterns for the entity based on the initial visual data, thereby forming a primary animation sequence;

[0162] The output module 36 is used to receive the primary animation sequence by the synthetic agent, perform logical consistency verification on the frame-level content of the primary animation sequence based on the scene description information, and generate a feedback signal according to the verification result to drive the visual agent or motion effect agent to perform iterative adjustments until the frame-level content of the primary animation sequence meets the logical consistency, and output the final animation product.

[0163] The AI ​​animation full-process automated generation system based on multi-agent collaboration in this application embodiment is used to implement the aforementioned AI animation full-process automated generation method based on multi-agent collaboration. Therefore, the specific implementation of the AI ​​animation full-process automated generation system based on multi-agent collaboration can be found in the embodiment section of the AI ​​animation full-process automated generation method based on multi-agent collaboration mentioned above. The specific implementation can be referred to the description of the corresponding embodiment, which will not be repeated here.

[0164] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described method for fully automated generation of AI animation based on multi-agent collaboration.

[0165] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for the fully automated generation of AI animation based on multi-agent collaboration.

[0166] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0167] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the AI ​​animation full-process automated generation method based on multi-agent collaboration.

[0168] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0169] The above provides a detailed description of the AI ​​animation full-process automated generation method and system based on multi-agent collaboration provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A fully automated AI animation generation method based on multi-agent collaboration, characterized in that, include: Receive text data on specialized topics; Based on the text data, the knowledge base is accessed and user annotation information is integrated to generate creative intent data, which includes entities and the attributes, relationships and logical rules of the entities; Based on the creative intent data, the scriptwriting agent runs a multi-agent symbolic reasoning process to generate scene description information, which includes entities with constraints, relationships between entities, and rules that drive the animation state. The visual intelligent agent generates a three-dimensional geometric model based on the scene description information and performs dynamic pre-simulation to form initial visual data; Based on the initial visual data, the motion effects agent configures the basic motion patterns for the entity, forming a primary animation sequence; The synthetic agent receives the primary animation sequence, performs logical consistency verification on the frame-level content of the primary animation sequence based on the scene description information, and generates a feedback signal based on the verification result to drive the visual agent or motion effects agent to perform iterative adjustments until the frame-level content of the primary animation sequence satisfies logical consistency, and outputs the final animation product, including: The synthetic agent receives the primary animation sequence and retrieves the corresponding scene description information as the basis for frame-level logical consistency verification. The primary animation sequence is broken down frame by frame to extract the content of each frame. The content of each frame includes entity state, inter-entity relationship information, and changes in the state of the frame. The extracted image content is compared with the scene description information one by one to determine whether the content at each frame level meets the logical consistency requirements. Problems are located in frame-level content that do not meet the logical consistency requirements, corresponding feedback signals are generated, and the feedback signals are sent to the corresponding visual agent or motion effect agent to drive the visual agent or motion effect agent to perform iterative adjustments. Repeat the above steps of comparison, positioning, feedback, and adjustment. If the content of all frames meets the logical consistency requirements, the adjusted animation sequence will be integrated and optimized to generate and output the final animation product.

2. The method according to claim 1, characterized in that, The process by which the screenwriter's intelligent agent generates scene description information by performing a multi-agent symbolic reasoning process based on the creative intent data includes: The screenwriter's intelligent agent initiates multiple specialized intelligent agents to perform text parsing, knowledge concept association, and animation scene rule extraction operations in parallel on the creative intent data, and obtains the processing results of each specialized intelligent agent. The processing results of each of the aforementioned specialized intelligent agents are interactively challenged to identify and eliminate conflicting content, and the eliminated content is jointly verified to fill in the missing content. Extract entities from the verified and completed content and add scene creation constraints. Clarify the relationships between the entities and extract the content that guides the changes in the animation screen state as the rules that drive the animation state. By integrating entities with constraints, the relationships between these entities, and the rules that drive the animation state, scene description information is generated.

3. The method according to claim 1, characterized in that, The step of generating a three-dimensional geometric model and performing dynamic pre-simulation based on the scene description information by the visual intelligent agent to form initial visual data includes: The visual intelligent agent parses the scene description information, extracts entities, relationships between entities, and rules driving the animation state from the scene description information, so as to extract numerical attributes from the entities and determine geometric construction parameters; Based on the geometric construction parameters, generate three-dimensional geometric shapes for each entity, integrate the three-dimensional geometric shapes to form a three-dimensional geometric model, and configure spatial association conditions and state change triggering conditions for the entities; The physical simulation process is driven by the spatial association conditions and the state change triggering conditions. The three-dimensional geometric model is dynamically pre-simulated, the key information of the pre-simulation image is extracted as the initial key frame, and the view trajectory content is planned as the camera path. Based on the three-dimensional geometric models of each entity, the initial keyframes and camera paths are integrated to form initial visual data, and the three-dimensional geometric models are output synchronously with the initial visual data as generated related content.

4. The method according to claim 3, characterized in that, The process of driving the physical simulation based on the spatial correlation conditions and the state change triggering conditions, and dynamically pre-simulating the three-dimensional geometric model, includes: The visual intelligent agent transforms the spatial association conditions and the state change triggering conditions into executable parameters for physical simulation and divides them into pre-simulation stages, which include entity individual pre-simulation stage, entity association pre-simulation stage and full scene integration pre-simulation stage. During the individual entity pre-play phase, it is verified whether the state changes of a single entity meet the state change triggering conditions and its own constraints. During the entity association pre-simulation phase, the linkage of state changes of associated entities is verified to ensure consistency with the entity relationships in the scenario description information. During the full-scene integrated pre-rehearsal phase, the pre-rehearsal rate is controlled in conjunction with the initial keyframe extraction requirements, and entity state anomalies and spatial position deviations are captured in real time during the pre-rehearsal process.

5. The method according to claim 1, characterized in that, The process involves a motion effects agent configuring basic motion patterns for the entity based on the initial visual data, forming a primary animation sequence, including: The initial visual data is parsed by the motion effects intelligent agent to extract entities, initial keyframes, and camera paths; Based on the attributes and constraints corresponding to the entities, the motion mode is matched and adapted, and the adapted motion mode is determined as the basic motion mode of each entity and the configuration is completed. The basic motion pattern is temporally correlated with the initial keyframe, and the perspective of the physical motion image is adjusted in conjunction with the camera path. The adjusted motion frames of each entity are sorted and integrated according to time nodes to generate a primary animation sequence.

6. The method according to claim 1, characterized in that, The step of accessing the knowledge base and integrating user annotation information based on the text data to generate creative intent data includes: Feature extraction is performed on the text data and a retrieval identifier is constructed. Based on the retrieval identifier, the knowledge base is searched layer by layer to obtain knowledge base association information. The user-annotated information is classified and extracted to obtain annotation feature information and user-specified association requirements. The knowledge base association information is matched and fused with the annotation feature information, and the fused information is associated with content according to the association requirements. Extract entities from the information after content association and assign attributes, sort out the relationships between the entities, and extract the behavioral constraints that the entities must follow as logical rules; The entities, their attributes, relationships, and logical rules are integrated to generate creative intent data.

7. A fully automated AI animation generation system based on multi-agent collaboration, characterized in that, include: The receiving module is used to receive text data on professional topics; The first generation module is used to access the knowledge base and integrate user annotation information based on the text data to generate creative intent data, wherein the creative intent data includes entities and the attributes, relationships and logical rules of the entities; The second generation module is used by the screenwriter agent to run a multi-agent symbolic reasoning process based on the creative intent data to generate scene description information, which includes entities with constraints, relationships between entities, and rules that drive the animation state. The first forming module is used by the visual intelligent agent to generate a three-dimensional geometric model and perform dynamic pre-simulation based on the scene description information to form initial visual data; The second forming module is used by the motion effects agent to configure basic motion patterns for the entity based on the initial visual data, thereby forming a primary animation sequence; The output module is used to receive the primary animation sequence from the synthetic agent, perform logical consistency verification on the frame-level content of the primary animation sequence based on the scene description information, and generate a feedback signal according to the verification result to drive the visual agent or motion effects agent to perform iterative adjustments until the frame-level content of the primary animation sequence meets the logical consistency, and output the final animation product, including: The synthetic agent receives the primary animation sequence and retrieves the corresponding scene description information as the basis for frame-level logical consistency verification. The primary animation sequence is broken down frame by frame to extract the content of each frame. The content of each frame includes entity state, inter-entity relationship information, and changes in the state of the frame. The extracted image content is compared with the scene description information one by one to determine whether the content at each frame level meets the logical consistency requirements. Problems are located in frame-level content that do not meet the logical consistency requirements, corresponding feedback signals are generated, and the feedback signals are sent to the corresponding visual agent or motion effect agent to drive the visual agent or motion effect agent to perform iterative adjustments. Repeat the above steps of comparison, positioning, feedback, and adjustment. If the content of all frames meets the logical consistency requirements, the adjusted animation sequence will be integrated and optimized to generate and output the final animation product.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the fully automated AI animation generation method based on multi-agent collaboration as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the automated generation method for the entire AI animation process based on multi-agent collaboration as described in any one of claims 1 to 6.

Images