Method, system and medium for generative ai creation running based on world state snapshot

Abstract

The application discloses a generative AI creation operation method and system based on world state snapshots, and a medium, relates to the technical field of artificial intelligence, and aims to solve the technical problems of difficult state tracing, poor creation coherence, insufficient consistency of multi-round iteration, and inability to accurately reuse the creation state in the existing generative AI creation process. The method comprises the following steps: constructing the world state dimension of generative AI creation, collecting real-time state data of the whole creation process and generating standardized world state snapshots; structurally storing and indexing the world state snapshots, establishing a snapshot retrieval and rapid calling mechanism; after receiving a creation instruction, matching the target world state snapshot and loading it into the AI creation model; based on the loaded state snapshot, executing AI creation generation, synchronously updating the state data in real time and generating a new snapshot; realizing state rollback, breakpoint continuation, multi-version parallel creation and accurate state reuse in the creation process. The system comprises a state collection module, a snapshot generation module, a storage retrieval module, a state loading module, a creation generation module and a state updating module. The application can greatly improve the coherence, controllability and iteration efficiency of generative AI creation, reduce the overhead of repeated calculation, and is suitable for multi-type AI creation scenes such as text, image, audio and video.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a generative AI creation and operation method, system, and medium based on world state snapshots. Background Technology

[0002] With the rapid development of generative artificial intelligence technology, AI has been widely applied in various fields such as text creation, image generation, audio and video production, and code writing, greatly improving the efficiency of content creation and the richness of creative forms. Existing generative AI creation systems mostly adopt a real-time inference, single-generation operation mode, which has the following technical shortcomings in the creation process:

[0003] First, the creative state cannot be accurately tracked and reused. During the operation of existing AI creation models, real-time status data such as model parameters, creative content status, and user interaction preferences cannot be completely saved. Once creation is interrupted, or modifications or iterative optimizations are needed, the historical creative state cannot be quickly restored. The creation command must be restarted, resulting in a large amount of repetitive calculation, low creation efficiency, and difficulty in ensuring consistency across multiple rounds of creation.

[0004] Second, the continuity and controllability of creation are insufficient. In the creation of large-scale content such as long texts, complex images, and long audio and video recordings, AI models are prone to problems such as state drift, content logic breaks, and inconsistent styles. They lack standardized control over the entire creation process, and users cannot accurately intervene in creation nodes, resulting in poor controllability of the creation results.

[0005] Third, the system lacks the ability to create multiple versions in parallel and resume creation from where interruption occurred. Existing generative AI systems do not support generating multiple differentiated versions based on the same creative state, nor can they resume creation from where interruption occurred. When the creative task is interrupted or multiple solutions need to be created in parallel, the operation process is cumbersome and cannot meet the needs of efficient and flexible creation.

[0006] Fourth, the management of status data is chaotic, resulting in low retrieval efficiency. The existing system does not provide structured classification and standardized encapsulation of AI creation status. Status data is stored in a scattered manner and lacks an efficient retrieval and matching mechanism, making it impossible to quickly locate and retrieve the target creation status, further reducing the efficiency of creation iteration.

[0007] To address the aforementioned issues, no perfect solution has yet been proposed in the existing technology. Therefore, there is an urgent need to develop a generative AI creation operation solution that can achieve complete collection, accurate storage, rapid retrieval, and real-time updates of the creative state, thereby solving the technical pain points of existing technologies such as difficulty in state tracing, poor continuity, and low iteration efficiency. Summary of the Invention

[0008] The purpose of this invention is to provide a generative AI creation method, system, and medium based on world state snapshots to solve the problems mentioned in the background art above:

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A generative AI creation method based on world state snapshots includes the following steps:

[0011] S1. Construct a world state dimension system corresponding to generative AI creation, collect state data of the entire AI creation process in real time based on the state dimension system, and generate corresponding world state snapshots after standardizing the collected state data.

[0012] Specifically, the world state dimension system includes four core dimensions: model operation state dimension, content creation state dimension, user interaction state dimension, and environmental parameter state dimension. The model operation state dimension data includes AI model computing power usage, parameter configuration, inference progress, model activation layer state, and inference cache data, comprehensively reflecting the AI ​​model's own operational status. The content creation state dimension data includes semantic features of the created text, image pixel and style features, audio and video frame information, creation completion progress, content logical structure, creation style parameters, and material reference information, accurately representing the real-time status of the created content. The user interaction state dimension data includes user creation instructions, personalized creation preferences, historical modification records, interaction permissions, version selection instructions, and intervention parameters, reflecting the user's control needs over the creation process. The environmental parameter state dimension data includes hardware parameters of the running device, network connection status, storage resource usage, data transmission rate, timestamps, and scene identifiers, ensuring the adaptability of the creation state to the operating environment.

[0013] When generating a world state snapshot, the collected multi-dimensional state data is first normalized and denoised to remove redundant and invalid data and extract core state features that can fully represent the creative state. Then, the core state features are structured and encoded according to a preset unified encoding rule and compressed into a snapshot data package containing complete state information. Finally, the snapshot data package is bound with attributes such as unique identifier, timestamp, creation type, state tag, and version information to form a standardized world state snapshot that can be stored and called independently, ensuring the integrity and uniqueness of the snapshot data.

[0014] S2. The generated world state snapshot is structurally encapsulated, a unique snapshot identifier and index label are established, the encapsulated world state snapshot is stored in a preset snapshot database, and a snapshot retrieval and matching mechanism is constructed.

[0015] The system provides a structured encapsulation of world state snapshots, employing a hierarchical storage architecture. Snapshots are categorized and stored according to creation type, time point, status type, and user identifier, improving storage efficiency and subsequent retrieval speed. Each snapshot is assigned a unique ID, and multi-level index tags are set, covering dimensions such as creation domain, status type, timestamp, user information, and creation progress, achieving precise snapshot classification.

[0016] The constructed snapshot retrieval and matching mechanism includes four modes: keyword retrieval, state similarity matching, timeline retrieval, and creation type matching, supporting multi-condition combination retrieval. After a user initiates a creation command, the system automatically parses the command to extract retrieval features, and uses a feature matching algorithm to filter candidate snapshots that meet the calling conditions from the snapshot database. If multiple candidate snapshots exist, they are sorted by state similarity calculation, and the target world state snapshot with the highest matching degree to the creation needs is output first.

[0017] S3. Receive the AI ​​creation instruction initiated by the user, parse the creation instruction to obtain the creation requirements and state call requirements, and match the corresponding target world state snapshot from the snapshot database based on the retrieval and matching mechanism.

[0018] The system listens for and receives AI creation commands input by the user in real time. Through natural language parsing and feature extraction technology, it breaks down the core creation requirements, target creation type, state invocation requirements, and personalized parameters in the commands. The parsed feature information is matched with the index tags in the snapshot database to select target world state snapshots that meet the requirements. If the user does not specify state invocation requirements, the system defaults to using the initial blank state snapshot and starts a new creation process.

[0019] S4. Load the target world state snapshot into the generative AI creation model, restore the corresponding creation state, and perform AI creation generation operation based on the loaded creation state.

[0020] Before loading the target world state snapshot, a state verification process is initiated to check the version compatibility, data integrity, and parameter validity of the snapshot state data with the current AI creation model, and to identify issues such as data corruption and parameter mismatch. If the verification passes, the snapshot data is directly parsed and loaded into the AI ​​creation model to fully restore the creation state corresponding to the snapshot, including model parameters, content progress, and interaction preferences. If the verification fails, the system automatically performs adaptive adjustments to the snapshot state data, optimizes parameter configurations, and repairs abnormal data until the loading and running requirements are met, and then completes the state restoration.

[0021] Once the state is restored, the creation generation module uses AI reasoning to generate content based on the loaded creation state and the user's creation instructions, ensuring that the creation process continues the historical state and avoiding issues such as state drift and content inconsistency.

[0022] S5. During the AI ​​creation process, the updated state data is collected in real time, and a new world state snapshot is generated and updated to the snapshot database simultaneously, so as to realize the real-time iteration and traceability of the creation state.

[0023] Throughout the entire AI-generated creation process, the state acquisition module continuously collects updated multi-dimensional state data in real time. Following the rules of step S1, it generates new world state snapshots in real time and simultaneously uploads them to the snapshot database for overwriting or incremental storage, enabling dynamic iteration of the creation state. Simultaneously, the system retains historical snapshot data, allowing users to retrieve snapshots from any point in time at any time, enabling precise retrospection of the creation state, continuation of creation from breakpoints, and comparison of multiple versions.

[0024] Furthermore, the method of the present invention also includes state backtracking and breakpoint resumption steps: when the creation process is interrupted or the user needs to revert to the historical creation state, the user's state backtracking instruction is received, the target backtracking time point or the unique identifier of the target snapshot is obtained, the corresponding world state snapshot is accurately retrieved from the snapshot database, and the current running state of the AI ​​creation model is seamlessly switched to the creation state corresponding to the target snapshot. Without restarting the creation process, breakpoint resumption or restoration of the historical creation version can be directly achieved.

[0025] Furthermore, the method of the present invention also includes a multi-version parallel creation step: based on the user-selected basic world state snapshot, the system automatically copies and generates multiple independent state copies, loads each state copy into different AI creation instances, issues differentiated creation instructions to each creation instance, realizes multi-version parallel creation based on the same basic state, and each version of the creation state is updated and stored independently without interfering with each other, thus meeting the user's multi-scheme creation needs.

[0026] II. Generative AI Creation and Operation System Based on World State Snapshots

[0027] A generative AI creation and execution system based on world state snapshots includes:

[0028] The state acquisition module is used to build a world state dimension system for generative AI creation. It collects state data in real time from four dimensions: model operation, creation content, user interaction, and environmental parameters throughout the entire AI creation process, and completes preliminary data cleaning and normalization.

[0029] The snapshot generation module, connected to the state acquisition module, receives the processed state data transmitted by the state acquisition module, performs structured encoding and encapsulation according to preset encoding rules, generates a standardized world state snapshot, and binds a unique identifier and attribute tag to the snapshot.

[0030] The storage and retrieval module, connected to the snapshot generation module, is used to store world state snapshots in a hierarchical and structured manner, build a multi-level index and retrieval matching mechanism, and realize fast retrieval, accurate matching and efficient access to snapshots.

[0031] The state loading module, connected to the storage and retrieval module, is used to parse user creation instructions, extract retrieval features, and call the storage and retrieval module to match target snapshots, complete snapshot data verification and adaptive adaptation, and load the snapshots into the AI ​​creation model.

[0032] The creation generation module, connected to the state loading module, is used to receive user creation instructions based on the loaded and restored creation state, and drive the generative AI model to perform content creation and inference generation operations.

[0033] The status update module is connected to the creation generation module and the snapshot generation module respectively. It is used to collect updated status data in real time during the creation process, transmit it to the snapshot generation module to generate a new snapshot, and synchronize it to the storage and retrieval module to complete the data update.

[0034] Furthermore, the system also includes:

[0035] The state verification module, connected to the state loading module, is used to verify the compatibility, data integrity, and parameter validity of the target world state snapshot and the AI ​​creation model, and to adaptively repair abnormal data.

[0036] The status rewind module, connected to the storage and retrieval module, is used to receive user rewind commands, retrieve target historical snapshots, control the AI ​​creation model to switch to the target creation state, and realize breakpoint continuation creation and version rewind.

[0037] The parallel creation module, connected to the storage and retrieval module and the creation generation module, is used to generate multiple independent state copies based on the basic snapshot and distribute them to different AI creation instances to achieve multi-version parallel creation.

[0038] III. Computer-readable storage media

[0039] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described generative AI creation method based on world state snapshots.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] 1. This invention constructs a multi-dimensional world state system, generates standardized world state snapshots, and fully collects and saves state data of the entire AI creation process, enabling accurate tracking and reuse of the creation state. This completely solves the problems of state loss and inability to trace back in existing technologies, significantly reduces redundant calculation overhead, and improves creation efficiency.

[0042] 2. This invention ensures the stability, style consistency, and logical coherence of the AI ​​creation process by loading and restoring state snapshots, effectively avoiding problems such as model state drift and content fragmentation, significantly improving the controllability and consistency of AI creation results, and adapting to large-scale content creation scenarios such as long texts and complex images.

[0043] 3. This invention has complete breakpoint resume creation and multi-version parallel creation capabilities, supports the generation of multiple differentiated creation versions based on the same state, meets users' flexible creation needs, simplifies the multi-scheme creation operation process, and improves the flexibility and diversity of creation.

[0044] 4. This invention employs a structured storage and multi-level index retrieval mechanism to achieve efficient storage and rapid retrieval of world state snapshots, simplifying the state management process, improving the speed of creation iteration, and is applicable to various generative AI creation scenarios such as text, images, audio and video, and code. It is highly versatile and practical. Attached Figure Description

[0045] Figure 1 is a flowchart of the generative AI creation and operation method based on world state snapshots of the present invention;

[0046] Figure 2 is a block diagram of the world state dimension architecture of the present invention;

[0047] Figure 3 is a block diagram of the generative AI creation and operation system based on world state snapshots of the present invention.

[0048] In Figure 1, S1-S5 are the core steps of the method of the present invention, which sequentially complete the state acquisition and snapshot generation, snapshot storage and retrieval mechanism construction, creation instruction parsing and snapshot matching, snapshot loading and state restoration, and real-time state update and iteration, forming a complete closed-loop creation operation process.

[0049] In Figure 2, the world state dimension system is divided into four core dimensions: model running state dimension, content creation state dimension, user interaction state dimension, and environmental parameter state dimension. Each dimension contains multiple subdivided state parameters, comprehensively covering the state information of the entire AI creation process.

[0050] In Figure 3, the core modules of the system include a status acquisition module, a snapshot generation module, a storage and retrieval module, a status loading module, a creation generation module, and a status update module. It is also equipped with a status verification module, a status backtracking module, and a parallel creation module. All modules work together to achieve full-process control of AI creation based on world status snapshots. Detailed Implementation

[0051] Please see Figure 1 - Figure 3 A generative AI creation method based on world state snapshots includes the following steps:

[0052] S1. Construct a world state dimension system corresponding to generative AI creation. In this embodiment, text AI creation is used as an example. The state dimension system includes model running state dimension (model parameters, inference progress, computing power usage), creation content state dimension (text semantics, paragraph structure, creation progress, writing style parameters), user interaction state dimension (user instructions, creation preferences, modification records), and environmental parameter state dimension (timestamp, device parameters). The above-mentioned state data are collected in real time, normalized and denoised, and the core state features are extracted. A structured snapshot data package is generated according to JSON encoding rules, and a timestamp and a unique ID are bound to form a text creation world state snapshot.

[0053] S2. The snapshots are encapsulated in a structured manner, and classified storage directories are established according to the text creation field and time node. Multi-level index tags of keywords, time and creation type are set and stored in a distributed snapshot database. A keyword + status similarity combination retrieval mechanism is built to support fast matching of target snapshots.

[0054] S3: Receive the user's creation instruction "continue writing the novel, continuing the state at the end of Chapter 3", parse and obtain the creation requirement as novel continuation and the state call requirement as the state at the end of Chapter 3; match the corresponding world state snapshot through the retrieval mechanism and filter out the unique target snapshot.

[0055] S4. Load the target snapshot into the text generation AI model, verify the compatibility between the snapshot and the model, and after confirming that the data is complete and valid, restore the novel to the state up to the end of the third chapter, including writing style, plot logic, character settings, etc.

[0056] S5. Based on the restored state, the novel continuation is generated. The updated state data is collected in real time. A new world state snapshot is generated for each paragraph and stored synchronously in the database. Users can retrieve the snapshot of any continuation node at any time for modification or rewind.

[0057] Meanwhile, this embodiment supports resume creation from the breakpoint: if the device is turned off during the creation process, it can be restarted and the latest snapshot can be retrieved to continue creation directly from the breakpoint; it also supports parallel creation of multiple versions: based on the snapshot at the end of Chapter 3, two state copies are made, and different plot direction instructions are executed to generate two different versions of the novel's subsequent content.

[0058] Example 2

[0059] As shown in Figure 3, a generative AI creation system based on world state snapshots includes:

[0060] The status acquisition module constructs a multi-dimensional world status system, collects real-time status data of the entire AI image creation process, including model parameters, image features, drawing progress, user commands, device status, and other data, and completes data cleaning.

[0061] The snapshot generation module receives state data, performs structured encoding to generate a snapshot of the world state of image creation, and binds it with a unique identifier, timestamp, and image style tag.

[0062] The storage and retrieval module uses a hierarchical storage method to store snapshots and builds multi-level indexes based on image type, style, and time to enable fast snapshot retrieval.

[0063] The state loading module parses the user's image optimization instructions, matches the target historical snapshot, verifies snapshot compatibility, and then loads it into the image generation AI model.

[0064] The creation and generation module performs image optimization and detail modification operations based on the loaded image creation status to maintain consistency between the image style and the overall effect.

[0065] The status update module collects optimized image status data in real time, generates new snapshots, and stores them synchronously.

[0066] Meanwhile, the system's status verification module completes snapshot data verification, the status backtracking module supports restoring to historical drawing states, and the parallel creation module supports generating different styles of creation versions based on the same image state.

[0067] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A generative AI creation and execution method based on world state snapshots, characterized in that, Includes the following steps: S1. Construct a world state dimension system corresponding to generative AI creation, collect state data of the entire AI creation process in real time based on the state dimension system, and generate corresponding world state snapshots after standardizing the collected state data. S2. The generated world state snapshot is structurally encapsulated, a unique snapshot identifier and index label are established, the encapsulated world state snapshot is stored in the preset snapshot database, and a snapshot retrieval and matching mechanism is constructed at the same time. S3. Receive the AI ​​creation instruction initiated by the user, parse the creation instruction to obtain the creation requirements and state call requirements, and match the corresponding target world state snapshot from the snapshot database based on the retrieval and matching mechanism. S4. Load the target world state snapshot into the generative AI creation model, restore the corresponding creation state, and perform AI creation generation operation based on the loaded creation state; S5. During the AI ​​creation process, the updated state data is collected in real time, and a new world state snapshot is generated and updated to the snapshot database simultaneously, so as to realize the real-time iteration and traceability of the creation state.

2. The operating method according to claim 1, characterized in that, In step S1, the world state dimension system includes the model running state dimension, the created content state dimension, the user interaction state dimension, and the environmental parameter state dimension. The model's operational status data includes AI model computing power usage, parameter configuration, inference progress, and model activation layer status. The creation content status dimension data includes creation text semantics, image features, audio and video frame information, creation progress, content logical structure, and style parameters; The user interaction status dimension data includes user commands, creation preferences, modification records, interaction permissions, and version selection commands; The environmental parameter status dimension data includes operating equipment parameters, network status, storage resource usage, and timestamp information.

3. The operating method according to claim 2, characterized in that, Step S1, which generates a world state snapshot, specifically includes: Normalize and reduce noise in multi-dimensional state data to extract core state features; The core state features are structured and encoded according to preset encoding rules to generate a snapshot data packet containing complete state information; The snapshot data package is bound with timestamp, creation type, and status tag attribute information to form a standardized world state snapshot.

4. The operating method according to claim 1, characterized in that, In step S2, the snapshot retrieval matching mechanism includes keyword retrieval, status similarity matching, timeline retrieval, and creation type matching; Based on the requirements information in the user's creation instructions, the search features are extracted, and world state snapshots that meet the calling conditions are selected from the snapshot database through feature matching. If multiple candidate snapshots are matched, they are sorted according to state similarity and the optimal target snapshot is output.

5. The operating method according to claim 1, characterized in that, In step S4, after loading the target world state snapshot into the generative AI creation model, a state verification step is also included: Verify the compatibility of snapshot status data with the current AI creation model, and verify the completeness and validity of the status data; If the verification passes, the creation status will be restored directly. If the verification fails, the status data will be adaptively adjusted until the loading requirements are met.

6. The operating method according to claim 1, characterized in that, It also includes state backtracking and breakpoint restart steps: Receive user state backtracking instructions, obtain the target backtracking time point or target snapshot identifier, and retrieve the corresponding world state snapshot from the snapshot database; Switch the current state of the AI ​​creation model to the creation state corresponding to the target snapshot to enable breakpoint continuation of the creation process or rollback of historical versions.

7. The operating method according to claim 1, characterized in that, It also includes a multi-version parallel creation process: Based on the same basic world state snapshot, multiple independent state copies are generated. Each state copy is loaded into a different AI creation instance, and differentiated creation instructions are executed to generate multiple parallel creation versions, with each version's state being updated and stored independently.

8. A generative AI creation and operation system based on world state snapshots, characterized in that, include: The state acquisition module is used to build a world state dimension system for generative AI creation and to collect multi-dimensional state data of the entire AI creation process in real time. The snapshot generation module is used to standardize state data and generate corresponding world state snapshots. The storage and retrieval module is used to store world state snapshots in a structured manner, build a snapshot retrieval and matching mechanism, and realize fast snapshot retrieval and retrieval. The state loading module is used to parse creation instructions, match and load the target world state snapshot into the generative AI creation model, and restore the creation state; The creation generation module is used to perform AI creation generation operations based on the loaded creation state; The status update module is used to update status data in real time during the creation process, generate new world status snapshots, and synchronize them to the storage and retrieval module.

9. The operating system according to claim 8, characterized in that, Also includes: The state verification module is used to verify the compatibility and data integrity between the target world state snapshot and the AI ​​creation model; The state rewind module is used to retrieve a snapshot of the target world state based on the user's rewind command, enabling the creation state to be rewound and resumed from the breakpoint. The parallel creation module is used to generate multiple state copies based on a base state snapshot, enabling multi-version parallel AI creation.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the generative AI creation method based on world state snapshots as described in any one of claims 1-7.

Images