Picture sharing method and device based on AI platform, electronic equipment and storage medium

Abstract

The invention relates to the technical field of picture sharing, and discloses a picture sharing method and device based on an AI platform, electronic equipment and a storage medium, and the method comprises the steps: obtaining a generated data stream of a picture uploaded by a first user, converting the generated data stream into a structured JSON file, and enabling a built-in cross-platform compatible tag to support multi-terminal sharing; based on the second user creation preference portraits and the key features, non-key parameters are screened through an AI semantic matching model, an adaptive priority adjustable component is generated, and a difference thermodynamic diagram and a new picture are generated after user adjustment; the enhanced copyright metadata containing original author identification, content fingerprints and the like are embedded into the JSON parameter stream; and calculating a hierarchical hash value and carrying out block chain evidence storage during each parameter iteration, calculating contribution degree distribution earnings based on a multi-dimensional index, generating a corresponding NFT copyright certificate, and supporting on-chain reverse tracing. According to the method and the device, the problem of low efficiency in secondary creation parameter adjustment caused by lack of standardized structured packaging and dependency relationship labeling in parameter generation can be solved.

Description

Technical Field

[0001] This application relates to the technical field of image sharing, and in particular to an image sharing method, apparatus, electronic device, and storage medium based on an AI platform. Background Technology

[0002] Existing AI image generation and sharing technologies, with their efficient creative capabilities, have been deeply integrated into various fields such as design, social media, and marketing, providing users with convenient channels for generating and disseminating personalized images. However, current technologies mostly achieve image sharing in the form of finished files, and the core parameters in the generation process are not standardized or structurally encapsulated. This not only leads to compatibility issues when adapting to multiple terminals, but also prevents users from performing precise secondary optimizations based on the core logic of the original creation, severely limiting the continuity and scalability of creation.

[0003] This technical deficiency has led to key problems in secondary creation scenarios: due to the lack of structured organization and dependency labeling of generated parameters, it is difficult for users to know the correlation and influence between parameters when adjusting them later, which can easily lead to parameter adjustment conflicts or ineffective adjustments, resulting in low efficiency of secondary creation; at the same time, the lack of unified parameter adaptation rules when adapting to multiple terminals means that the same parameters will have inconsistent effects on different terminals, which further exacerbates the sense of fragmentation in the creative experience and hinders the large-scale promotion of AI image collaborative creation.

[0004] As can be seen from the above, the problem of inefficient parameter adjustment in secondary creation due to the lack of standardized, structured encapsulation and dependency annotation of generated parameters still needs to be solved. Summary of the Invention

[0005] To address the problem of inefficient parameter adjustment in secondary creation due to the lack of standardized, structured encapsulation and dependency annotation of generated parameters, this application provides an image sharing method, device, electronic device, and storage medium based on an AI platform.

[0006] Firstly, this application provides an image sharing method based on an AI platform, employing the following technical solution: An AI-based image sharing method includes: Obtain the generated data stream corresponding to the first image uploaded by the first user. The generated data stream includes a unique identifier of the generation model, a customized random seed, the number of iteration steps, the semantic parsing results of the prompt words, and the model weight slices. Convert the generated data stream into a structured JSON file containing a parameter dependency graph. The JSON file has built-in cross-platform compatible tags to support sharing to at least one second user on multiple terminals. Based on the second user's creative preference profile and key input features, non-critical parameters in the JSON file are filtered through an AI semantic matching model to generate adjustable components with adaptive priority sorting. Each component is associated with a constraint threshold for parameter adjustment and an AI-recommended adjustment range, and the default value is dynamically optimized based on the user's historical adjustment behavior. After the second user adjusts the parameters through the adjustable components, a heatmap of the difference before and after the adjustment is generated in real time, and a new image is generated based on the heatmap of the difference and the adjusted parameters. The enhanced copyright metadata is embedded in the JSON parameter stream. The enhanced copyright metadata includes the original author identifier, creation timestamp, license type, AI-generated content fingerprint information, and contribution history including contributor identifier, contribution timestamp, contribution type, contribution rating, and adjustment parameter snapshot. During each parameter iteration, a Merkle tree algorithm is used to calculate the hierarchical hash value of the JSON parameter stream, generating a blockchain-based notarized record that includes a version number, timestamp, contributor information, hierarchical hash value, and a parameter dependency graph summary. Contribution is calculated based on multi-dimensional indicators, including the dimensional complexity of parameter adjustments, deviation of adjustment magnitude, AI-quantified improvement value of generation effect, and innovation score of adjustment behavior. The improvement value includes aesthetic score and semantic consistency score. Copyright revenue is dynamically allocated according to the contribution percentage, and an NFT copyright certificate bound to the hierarchical hash value is generated. The NFT copyright certificate supports reverse tracing of the corresponding parameter iteration version through on-chain queries.

[0007] Optionally, in the process of generating the adjustable component with adaptive priority ordering, the method further includes: The core semantic vector of the key features of the second user is extracted by the AI ​​semantic matching model, and the cosine similarity is calculated between the core semantic vector of the key features of the second user and the functional semantic vector of each parameter in the JSON file. Parameters with similarity below a preset similarity threshold are marked as non-critical parameters, and the non-critical parameters are prioritized by combining the commonly used adjustment dimensions and adjustment frequency weights in the second user creation preference profile. For each non-critical parameter after sorting, related parameters are identified based on the parameter dependency graph, and a component group of main parameter-related parameters is generated, in which parameter linkage rules are labeled within the component group; The system monitors the adjustment operations of the second user in real time. If the parameter adjustment is detected to exceed the constraint threshold, the AI ​​recommendation module will push multiple optimal adjustment schemes for the user to choose or refer to. The optimal adjustment scheme includes a preview image of the effect of the scheme.

[0008] Optionally, in the process of generating the difference heatmap before and after adjustment in real time, the method further includes: Pixel-level feature extraction is performed on the original image before adjustment and the preview image after adjustment to obtain feature data on color distribution, texture structure, and semantic region division. The AI ​​difference quantization model calculates the difference values ​​between the original image before adjustment and the preview image after adjustment in each feature dimension, and maps the difference values ​​to the color gradient of the heatmap. The higher the difference value, the more vivid the color of the heatmap. Semantic labels are overlaid on the heatmap to mark the parameter adjustment items corresponding to the differences, and the system supports automatic location of the corresponding adjustable component when the user clicks on the semantic label. If a second user continuously adjusts the same component group, the heatmap is updated in real time and the cumulative difference value is calculated. When the cumulative difference value exceeds the preset difference threshold, an effect solidification reminder is triggered and a snapshot of the parameters of the current adjusted version is saved.

[0009] Optionally, in the process of embedding enhanced copyright metadata into the JSON parameter stream, the method further includes: The first image and the generated data stream are processed by an AI content fingerprint generation model to extract a dual content fingerprint based on visual features and parametric features. Add an adjustment intent description field to the contribution history. The adjustment intent description field is automatically generated by parsing the input text or adjustment behavior of the second user through a natural language processing model. The enhanced copyright metadata is embedded into specified fields of the JSON parameter stream according to the hierarchical structure of original author information - core copyright information - contribution history, and the copyright metadata is encrypted. Each time the parameters are iterated, the contribution score in the contribution history is updated synchronously. The contribution score is dynamically calculated by the AI ​​copyright assessment model based on the improvement value of the adjustment effect and the innovation score.

[0010] Optionally, in the process of calculating the contribution based on multi-dimensional indicators, the method further includes: Set initial weights for each dimension of indicators and dynamically adjust the weights according to the image generation scenario, which includes commercial design, artistic creation, and daily sharing. Calculate the specific values ​​for each indicator, which includes dimensional complexity, deviation of adjustment range, improvement value, and innovation score; The total contribution is calculated based on the adjusted indicator weights. If there are multiple second users, the contribution percentage of each user is calculated separately to generate a contribution distribution matrix.

[0011] Optionally, in the process of generating the NFT copyright certificate bound to the hierarchical hash value, the method further includes: Based on the Merkle tree algorithm, the root hash value is extracted as the unique identifier of the NFT copyright certificate, and the hash values ​​of each layer are used as the metadata attributes of the NFT. Embed a parameter iteration traceability chain in the NFT copyright certificate. The parameter iteration traceability chain includes all iteration nodes from the original version to the current version. Each node is associated with a corresponding hierarchical hash value, contributor information and parameter adjustment summary. For secondary transactions of NFT copyright certificates, the copyright revenue distribution mechanism is automatically triggered during the transaction. The transaction revenue is split into the accounts of each contributor according to the current contribution distribution matrix, and an on-chain transaction record is generated. If the image content corresponding to an NFT is detected to be infringing, dual content fingerprints and parameter dependency graph summaries are extracted through on-chain queries to serve as core evidence for infringement protection.

[0012] Optionally, in the process of embedding cross-platform compatible tags in the JSON file to support multi-terminal sharing, the method further includes: We analyze the image rendering capabilities and interaction logic of current mainstream terminals and generate a rendering rule library adapted to each terminal. The current mainstream terminals include mobile terminals, PC terminals, tablet terminals and VR devices. Embed terminal type identification tags and rendering parameter adaptation tags in the JSON file. The rendering parameter adaptation tags include the image resolution, color space, and component interaction method corresponding to each terminal. When the second user receives the JSON file, the terminal automatically reads and identifies the tags, calls the corresponding adaptation rules from the rendering rule library, and dynamically adjusts the size, layout, and operation mode of the adjustable components. The JSON file is lightweighted by removing redundant parameter data based on the parameter dependency graph, compressing the file size using the LZ77 compression algorithm, retaining the complete parameter recovery interface, and allowing users to recover all parameter data when needed.

[0013] Secondly, this application provides an image sharing device based on an AI platform, employing the following technical solution: An AI-based image sharing device includes: The data stream structured cross-platform module obtains the generated data stream corresponding to the first image uploaded by the first user. The generated data stream includes a unique identifier of the generation model, a customized random seed, the number of iteration steps, the semantic parsing results of the prompt words, and the model weight slices. The generated data stream is converted into a structured JSON file containing a parameter dependency graph. The JSON file has built-in cross-platform compatible tags to support sharing to at least one second user on multiple terminals. The parameter adaptive adjustment visualization module, based on the second user's creative preference profile and key input features, filters non-critical parameters in the JSON file using an AI semantic matching model, generating adjustable components with adaptive priority order. Each component is associated with a constraint threshold for parameter adjustment and an AI-recommended adjustment range, and the default value is dynamically optimized based on the user's historical adjustment behavior. After the second user adjusts the parameters through the adjustable components, a heatmap of the difference before and after the adjustment is generated in real time, and a new image is generated based on the heatmap of the difference and the adjusted parameters. The copyright metadata embedding and updating module embeds enhanced copyright metadata into a JSON parameter stream. The enhanced copyright metadata includes the original author identifier, creation timestamp, license type, AI-generated content fingerprint information, and contribution history including contributor identifier, contribution timestamp, contribution type, contribution rating, and adjustment parameter snapshots. The blockchain copyright traceability contribution allocation module calculates the hierarchical hash value of the JSON parameter stream using the Merkle tree algorithm during each parameter iteration, generating a blockchain-based evidence record including version number, timestamp, contributor information, hierarchical hash value, and parameter dependency graph summary. It calculates contribution based on multi-dimensional indicators, including the dimensional complexity of parameter adjustments, deviation of adjustment magnitude, AI-quantified generation effect improvement value, and innovation score of adjustment behavior. The effect improvement value includes aesthetic score and semantic consistency score. Copyright revenue is dynamically allocated based on the contribution percentage, and an NFT copyright certificate bound to the hierarchical hash value is generated. This NFT copyright certificate supports reverse tracing of the corresponding parameter iteration version via on-chain queries.

[0014] Thirdly, this application provides an electronic device that adopts the following technical solution: An electronic device includes a processor in which a program for the image sharing method based on an AI platform, as described in any one of the preceding claims, is running.

[0015] Fourthly, this application provides a storage medium, which adopts the following technical solution: A storage medium storing a program for the image sharing method based on an AI platform as described in any one of the above.

[0016] In summary, this application includes at least one of the following beneficial technical effects: By converting the generated data stream into a structured JSON file containing a parameter dependency graph, standardized encapsulation of generated parameters and explicit labeling of dependencies are achieved, fundamentally solving the problem of unclear parameter relationships in secondary creation. Simultaneously, non-critical parameters are filtered based on an AI semantic matching model, and component groups with adaptive priority ranking are generated based on user creative preferences, with parameter linkage rules labeled. This allows users to clearly understand the influence logic between each parameter, avoiding adjustment conflicts. Combined with parameter adjustment constraint thresholds and AI recommendation schemes, ineffective adjustments are significantly reduced, greatly improving the efficiency of parameter adjustment in secondary creation.

[0017] By using pixel-level feature extraction and an AI-generated difference heatmap, the impact of parameter adjustments on images is visually presented. The addition of semantic tags and component positioning further lowers the operational barrier to parameter adjustment. Cumulative difference value monitoring and persistent alerts during continuous adjustments ensure the continuity of secondary creation. Furthermore, the cross-platform compatibility and lightweight processing of JSON files resolve inconsistencies across multiple devices, while enhanced copyright metadata embedding and blockchain-based notarization mechanisms enable precise definition and traceability of creative rights. This not only improves the efficiency of secondary creation but also strengthens the ecosystem for AI-powered image sharing. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an AI-based image sharing method according to an exemplary embodiment.

[0019] Figure 2 This is a structural block diagram of an AI-based image sharing device according to an exemplary embodiment. Detailed Implementation

[0020] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0021] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0022] This application discloses an image sharing method based on an AI platform, referring to... Figure 1 ,include: S100: Obtain the generated data stream corresponding to the first image uploaded by the first user. The generated data stream includes the unique identifier of the generation model, the customized random seed, the number of iteration steps, the semantic parsing results of the prompt words, and the model weight slices. Convert the generated data stream into a structured JSON file containing a parameter dependency graph. The JSON file has built-in cross-platform compatible tags to support sharing to at least one second user on multiple terminals.

[0023] The specific execution process of S100 is as follows: Step 1, Generate Data Stream Acquisition: First, the complete generated data stream corresponding to the first image uploaded by the first user is obtained. This data stream is not the finished image file, but the core "creation trajectory data" of the image generation process. Among them, the unique identifier of the generation model is used to accurately locate the original AI model that generated the image (to avoid the effect deviation caused by model mismatch during subsequent secondary creation); the customized random seed is the initial random value for image generation (to ensure that the original creation logic can be reproduced based on the seed); the iteration step count records the number of iterations when the model generates the image (reflecting the level of detail in image generation); the semantic parsing result of the prompt words is a structured decomposition of the user's original prompt words (such as decomposing "blue seaside sunset" into semantic units such as "scene: seaside, time: sunset, main color: blue", which facilitates subsequent parameter association); the model weight slice is the core parameter fragment of the original generation model (only the weight data directly related to the current image generation is extracted, taking into account both data integrity and lightweight).

[0024] Step 2, Structured JSON file conversion: The key to converting the collected generated data stream into a structured JSON file lies in embedding a parameter dependency graph. Specifically, the algorithm analyzes the correlation logic and influence weight between various parameters. For example, adjusting the "main color tone" parameter will affect parameters such as "light intensity" and "color saturation," while adjusting the "iteration step count" will affect parameters such as "detail richness" and "rendering speed." These relationships are written into the JSON file in the form of a graph (such as a node-edge structure), so that the originally scattered parameters form a logical whole, achieving parameter standardization and structured encapsulation.

[0025] Step 3, cross-platform compatible tag embedding: The JSON file incorporates two core tags: terminal type identification tags (for identifying the type of receiving device) and rendering parameter adaptation tags (containing adaptation rules for mainstream terminals such as mobile, PC, tablet, and VR devices, such as adapting to low resolution and simplifying component layout for mobile devices, and supporting high resolution and batch parameter operations for PC devices). Simultaneously, redundant data is removed based on the parameter dependency graph (e.g., deleting model weight fragments irrelevant to the current image generation), the file size is compressed using the LZ77 compression algorithm, and a "complete parameter recovery interface" is retained (ensuring users can restore all original parameters if needed).

[0026] Step 4, Multi-terminal sharing adaptation: After structuring and tag embedding are completed, the JSON file can be transmitted across terminals to at least one second user. The receiving terminal will automatically read the identification tags in the file and call the corresponding adaptation rules. Without the user having to make manual adjustments, the parameters can be adapted and compatible with the terminal.

[0027] By standardizing and structuring the generated data stream, constructing a parameter dependency graph, and embedding cross-platform tags, the problems of "dispersed and irregular generated parameters, unclear dependencies, and poor multi-terminal adaptability" in traditional technologies are fundamentally solved. On the one hand, the structured JSON file and parameter dependency graph make parameter adjustments during subsequent secondary creation systematic and provide a data foundation for precise optimization. On the other hand, cross-platform compatibility design and lightweight processing ensure the convenience of sharing and achieve consistent parameter presentation across different terminals.

[0028] S200, based on the second user's creative preference profile and key input features, uses an AI semantic matching model to filter non-critical parameters in the JSON file and generates adjustable components with adaptive priority sorting. Each component is associated with a constraint threshold for parameter adjustment and an AI-recommended adjustment range, and the default value is dynamically optimized based on the user's historical adjustment behavior. After the second user adjusts the parameters through the adjustable components, a heatmap of the difference before and after the adjustment is generated in real time, and a new image is generated based on the heatmap of the difference and the adjusted parameters.

[0029] The specific execution process of S200 is as follows: Step 1, User Characteristics and Key Data Retrieval: First, the second user's creative preference profile is retrieved (this profile is built based on the user's historical adjustment records, favorite content, creative tags and other data, including commonly used adjustment dimensions such as "color optimization" and "detail enhancement", and adjustment frequency weights such as high-frequency adjustments "brightness" and "contrast"). At the same time, the key features input by the second user are obtained (such as the core needs described in text or checked, such as "preserve the seaside scene and optimize the sky color" and "enhance the facial details of the characters").

[0030] Step 2, AI semantic matching and non-critical parameter filtering: The textual information of key features input by the second user is converted into core semantic vectors. These vectors are then used by an AI semantic matching model to perform cosine similarity calculations with the functional semantic vectors of each parameter in the JSON file (e.g., "dominant color tone" corresponds to "color adjustment," and "iteration steps" corresponds to "detail richness"). A preset similarity threshold (e.g., 0.6) is set. Parameters with similarity below this threshold are marked as non-critical parameters—parameters that the user has not explicitly focused on and can be flexibly adjusted. Key parameters with similarity above the threshold (e.g., the parameter corresponding to "sky color" mentioned by the user) are locked in a state where they cannot be arbitrarily adjusted, ensuring that core requirements are not compromised.

[0031] Step 3, Generate adaptive adjustable components: For the selected non-critical parameters, priority is assigned based on the "frequently used adjustment dimensions" and "adjustment frequency weight" from the second user creation preference profile (e.g., users who frequently adjust "saturation" are given priority). This generates adjustable components with adaptive priorities. Each component is associated with two core pieces of information: first, the constraint threshold for parameter adjustment (set based on a parameter dependency graph to avoid adjustments exceeding a reasonable range that could lead to image distortion; for example, the "brightness" adjustment threshold is limited to -30% to +30%); second, the AI-recommended adjustment range (generated based on similar image optimization cases and user preferences; for example, the recommended "saturation" adjustment range is +5% to +15%). Furthermore, the default value for each component is not fixed but dynamically optimized based on the user's historical adjustment behavior (e.g., if a user has a history of increasing the saturation of "warm-toned images" by 10%, the default value for that scenario is automatically set to +10%).

[0032] Step 4, Real-time generation and interaction of differential heatmaps: After the second user adjusts the parameters via the adjustable components, the system immediately initiates a dual-image comparison analysis: First, pixel-level feature extraction is performed on the original image before adjustment and the preview image after adjustment to obtain feature data on color distribution, texture structure, and semantic region division (such as "sky", "person", "background"); then, the difference value between the two images in each feature dimension is calculated using an AI difference quantization model, and the difference value is mapped to the color gradient of the heatmap (the higher the difference value, the more vivid the color, such as the "sky color" appearing as a deep red after adjustment if the difference value is high); then, semantic labels (such as "saturation +12%" and "brightness +8%) are superimposed on the heatmap, and interactive functions are supported. When the user clicks on a semantic label, the system automatically locates the corresponding adjustable component for accurate correction.

[0033] Step 5, New image generation: After the second user confirms the parameter adjustment, the system, based on the final adjusted parameters and the effective optimization data in the difference heatmap (excluding the difference values ​​corresponding to invalid adjustments), calls the appropriate generation model (matched by the unique identifier of the generation model in S100) to generate a new image that meets the user's needs, while retaining a snapshot of the parameters adjusted this time for subsequent traceability.

[0034] Through a full-process design of "user feature-driven intelligent parameter filtering, adaptive component generation, and visualization of adjustment effects," the problems of "blind parameter adjustment, unclear dependency relationships, and numerous inefficient and ineffective adjustments" in traditional secondary creation are precisely solved.

[0035] AI semantic matching and priority ranking allow users to focus only on non-critical parameters, avoiding disruption of core needs. Constraint thresholds and AI recommendation range reduce adjustment errors. In addition, the difference heatmap intuitively presents the adjustment effect and related parameters, lowering the operational threshold, and dynamically optimized default values ​​further align with user habits.

[0036] S300 embeds enhanced copyright metadata into the JSON parameter stream. The enhanced copyright metadata includes the original author identifier, creation timestamp, license type, AI-generated content fingerprint information, and contribution history containing contributor identifier, contribution timestamp, contribution type, contribution rating, and adjustment parameter snapshots.

[0037] The specific execution process of S300 is as follows: Step 1, Enhanced copyright metadata collection and generation: First, core basic copyright data is collected, including the original author identifier (the first user's unique account ID or identity identifier), creation timestamp (the precise time the first image was generated, accurate to milliseconds), and license type (such as commercial use license, non-commercial use license, derivative work license, etc., preset by the first user). Then, enhanced core data is generated: the first image (visual features) and the generated data stream (parameter features) in S100 are jointly processed using an AI content fingerprint generation model to extract dual content fingerprints (to prevent copyright invalidation due to single feature tampering); simultaneously, contribution history is initialized, including basic fields (contributor identifier, contribution timestamp, contribution type such as "parameter adjustment," "requirement optimization," contribution rating, and adjustment parameter snapshot), and a new field describing the adjustment intent is added. The second user's input text (such as "optimize sky color") or adjustment behavior (such as continuously adjusting "saturation" and "color temperature" parameters) is parsed using a natural language processing model, automatically generating a structured description to ensure the traceability of the adjustment behavior.

[0038] Step 2, Metadata Hierarchical Organization and Encryption: All enhanced copyright metadata is organized into a hierarchical structure of "Original Author Information - Core Copyright Information - Contribution History" to avoid data clutter. Original author information serves as the top-level data, core copyright information (including dual content fingerprints and license types) as the middle-level core, and contribution history as the bottom-level dynamically extended data. The entire layer of copyright metadata is then encrypted: the encryption key uses a hash combination of "Original Author Identifier + Creation Timestamp" (ensuring uniqueness and a strong association with the copyright holder), and the encrypted data is embedded in a specified hidden field of the JSON parameter stream to prevent unauthorized tampering or theft.

[0039] Step 3, embedding metadata into the JSON parameter stream and synchronously updating: The enhanced copyright metadata, after hierarchical encryption, is embedded into the JSON parameter stream generated by S100 and linked to the parameter dependency graph. Each parameter iteration corresponds to a complete copyright metadata record. A synchronous update mechanism is also implemented: each time a second user completes parameter adjustments (i.e., when S200 generates a new image), the system automatically adds a new record to the contribution history, supplementing it with the current contributor identifier (second user ID), contribution timestamp (adjustment completion time), contribution type (determined based on the adjustment operation), and a snapshot of the adjusted parameters (parameter name, original value, and adjusted value). Furthermore, the AI ​​copyright evaluation model dynamically calculates and updates the "contribution score" based on the improvement value and innovation score from the adjustments in S200, ensuring that the copyright metadata is synchronized with the creation process in real time.

[0040] The use of dual content fingerprinting and encryption ensures the uniqueness and security of copyright, effectively preventing infringement. In addition, detailed contribution history (including adjustment intent, parameter snapshots, and contribution scores) completely preserves the entire chain of multi-user collaborative creation, while making the boundaries of rights between the original author and contributors clearly traceable. This greatly enhances the copyright protection in AI image sharing and secondary creation, and stimulates the enthusiasm of multi-user collaborative creation.

[0041] S400 calculates the hierarchical hash value of the JSON parameter stream using the Merkle tree algorithm during each parameter iteration, generating a blockchain-based notarized record that includes a version number, timestamp, contributor information, hierarchical hash value, and a parameter dependency graph summary. Contribution is calculated based on multi-dimensional indicators, including the dimensional complexity of parameter adjustments, deviation of adjustment magnitude, AI-quantified improvement in generation effect, and innovation score of adjustment behavior. The improvement in effect includes aesthetic and semantic consistency scores. Copyright revenue is dynamically allocated based on contribution percentage, and an NFT copyright certificate bound to the hierarchical hash value is generated. This NFT copyright certificate allows for reverse tracing of the corresponding parameter iteration version via on-chain queries.

[0042] S400 utilizes blockchain technology to ensure traceability of the creation process, accurately calculates contribution based on multi-dimensional indicators, and generates transferable NFT copyright certificates. The specific execution process is as follows: Step 1, Hierarchical hash value calculation: Each time a second user completes parameter adjustments (i.e., one parameter iteration), the system calls the Merkle tree algorithm to process the current JSON parameter stream. First, the JSON parameter stream is split into multiple data blocks according to "core parameters - associated parameters - copyright metadata," and the hash value of each data block is calculated (forming the leaf nodes of the Merkle tree). Then, the hash values ​​of the parent nodes are calculated layer by layer upwards, ultimately generating the root hash value and the hash values ​​of each branch (i.e., hierarchical hash values). This ensures that any minor tampering with the parameter stream will result in a change in the hash value, achieving data integrity verification.

[0043] Step 2, Generation of Blockchain Evidence Records: Based on hierarchical hash values, structured blockchain evidence records are generated. These records contain five core fields: version number (incrementing by parameter iteration number, e.g., V1.0 corresponds to the initial version, V1.1 corresponds to the first adjustment), timestamp (the precise time the parameter iteration was completed), contributor information (the unique identifier of the second user in this adjustment), hierarchical hash value (including the root hash value and key branch hash values), and parameter dependency graph summary (a compressed representation of the graph in S100, used for quickly associating the original parameter logic). The generated records are synchronized to the blockchain network, leveraging the decentralized and immutable characteristics of blockchain to achieve end-to-end evidence preservation of the creation process.

[0044] Step 3, Calculation of multi-dimensional contribution: First, initial weights are set for each dimension indicator (dimensional complexity 0.2, adjustment deviation 0.15, effect improvement value 0.4, and innovation score 0.25), and dynamically adjusted according to the image generation scenario (commercial design, artistic creation, and daily sharing) (e.g., the effect improvement value weight is increased to 0.5 in the commercial design scenario). Then, the specific values ​​of each indicator are calculated: dimensional complexity is calculated as "number of adjusted dimensions × 1 + number of associated parameters × 0.5", adjustment deviation is the deviation ratio between the actual adjusted value and the average value of the AI ​​recommendation range, effect improvement value is the weighted average of the AI ​​aesthetic score (weight 0.6) and the semantic consistency score (weight 0.4), and innovation score is determined by comparing the platform's historical adjustment records (a repetition rate of less than 5% is a full score). Finally, the total contribution is calculated according to the adjusted weights. If there are multiple second users, a contribution percentage matrix for each user is generated.

[0045] Step 4, Dynamic Distribution of Copyright Revenue: Based on the contribution distribution matrix, the system automatically splits the copyright revenue into proportions. The original author receives a fixed percentage (e.g., 30%), and the remaining revenue is distributed according to the contribution percentage of each second-tier user (e.g., a user with a 20% contribution percentage receives 20% × 70% = 14% of the revenue). The revenue distribution rules and the distribution matrix are synchronously written into the blockchain for record-keeping, ensuring that the distribution process is transparent and traceable, and supporting subsequent revenue settlement and reconciliation.

[0046] Step 5, NFT copyright certificate generation and application: Merkel root hash value is extracted as the unique identifier of NFT copyright certificate, and hash values ​​of each layer and parameter iteration traceability chain (including hash values ​​of all version nodes, contributor information, and adjustment summary) are used as NFT metadata attributes; NFT supports secondary transactions, and the revenue distribution mechanism is automatically triggered when a transaction occurs, splitting the transaction revenue to each contributor's account according to the current contribution matrix and generating on-chain transaction records; if image infringement is detected, dual content fingerprints (generated by S300) and parameter dependency graph summaries are extracted through on-chain queries as core evidence for infringement protection.

[0047] By utilizing blockchain-based evidence storage, multi-dimensional contribution quantification, and NFT copyright certificate generation, the problems of "difficult copyright traceability, inability to quantify contribution, and unfair revenue distribution" in traditional multi-user collaborative creation are completely solved. On the one hand, layered hash values ​​and blockchain evidence storage ensure the immutability and full traceability of the creation process, providing irrefutable evidence of copyright ownership. On the other hand, multi-dimensional indicators accurately quantify the contribution value of each user, achieving fair and just revenue distribution and stimulating the enthusiasm of original authors and secondary creators. At the same time, NFT copyright certificates endow digital copyrights with transferable and tradable attributes, perfecting the closed-loop commercial ecosystem of AI image creation and providing core rights protection for the large-scale promotion of multi-user collaborative creation.

[0048] Based on the above solution, the following example illustrates this concept. Suppose user A generates a "blue seaside sunset" image using an AI platform and shares it with user B through this solution. User B's core requirement is to "preserve the core scene of the seaside and sunset, optimize the color gradation of the sky, and enhance the details of the waves." In traditional technologies, user B only receives the finished image and cannot know the original generation parameters and related logic. Adjustments require blind trial and error, which is extremely inefficient. This solution, however, precisely solves this problem through end-to-end design.

[0049] First, the S100 lays a structured data foundation for secondary creation: After user A uploads an image, the system collects and generates a data stream (including core data such as the unique identifier of the generation model, customized random seeds, and iteration steps), and converts it into a JSON file containing a parameter dependency graph. The file clearly marks the relationship logic between "dominant color tone" and "light and shadow intensity" and "color saturation," as well as the influence weight of "iteration steps" and "wave detail richness," completely solving the problems of scattered and irregular parameters and unclear dependencies in traditional methods. At the same time, the JSON file has built-in cross-platform tags, so user B can directly obtain the structured parameters on a mobile device without manual adaptation, removing data obstacles for efficient adjustments.

[0050] Secondly, the S200 reduces ineffective adjustments through intelligent filtering and adaptive component generation: The system calls upon user B's creative preference profile (frequently adjusting dimensions such as "color saturation" and "detail enhancement"), combines it with the key features of their input, and uses an AI semantic matching model to filter non-critical parameters, locking parameters related to "sky color" and "wave detail" as critical parameters (which cannot be arbitrarily adjusted), and sorting non-critical parameters such as "contrast" and "sharpening" according to user preferences to generate adaptive adjustable components. Each component is associated with a constraint threshold (e.g., "sharpening" is limited to -20% to +25%) and an AI recommended range (e.g., "contrast" is recommended to be +8% to +15%), and the default value is set to "contrast +10%" based on user B's historical adjustment habits, avoiding the inefficiency of "vague parameter ranges and blind trial and error" in traditional adjustments, allowing user B to focus only on core optimization dimensions.

[0051] Finally, the S200's difference heatmap visualization further enhances adjustment accuracy: After User B adjusts "Contrast" to +12% and "Sharpness" to +20% using the components, the system generates a difference heatmap in real time. The sky area appears deep red (high difference value) due to optimized color gradation, while the wave area appears light red (medium difference value) due to enhanced detail, with semantic labels "Contrast +12%" and "Sharpness +20%" overlaid. When User B clicks on the sky area label, the system automatically locates the "Color Saturation" component, allowing for precise correction without having to search for each component individually. After continuous adjustments, the system also calculates the cumulative difference value and triggers a fixed reminder, ensuring the adjustment effect is traceable. The entire process transforms User B from "blindly trying and failing" to "precise optimization," significantly improving the efficiency of secondary creation.

[0052] In this embodiment of the application, the method further includes the following steps in generating the adjustable component with adaptive priority sorting: Step 1, Semantic Vector Extraction and Similarity Measurement: First, the key features of the second user input (such as "optimize sky color") are converted into core semantic vectors that can be recognized by computers through an AI semantic matching model. At the same time, functional semantic vectors are preset for each parameter in the JSON file (such as "saturation", "color temperature", and "sharpening level"). For example, "saturation" corresponds to "color density adjustment". The matching degree of the two sets of vectors is calculated by using the cosine similarity algorithm to quantify the correlation strength between the parameters and the user's core needs.

[0053] Step 2, Filtering and prioritizing non-critical parameters: Set a preset similarity threshold (e.g., 0.6), and mark parameters with a matching degree lower than this threshold and weak correlation with the user's core needs as non-critical parameters (e.g., "contrast" and "shadow intensity" not mentioned by the user). Then, retrieve the creative preference profile of the second user, extract the "common adjustment dimensions" (e.g., the user frequently adjusts "color parameters") and "adjustment frequency weight" (e.g., adjusting "saturation" 15 times a month has a higher weight than adjusting "shadow intensity" 3 times a month), and prioritize the non-critical parameters, with frequently used parameters at the top.

[0054] Step 3, Component group generation and linkage rule annotation: Based on the parameter dependency graph constructed by S100, strongly correlated parameters are identified for each non-critical parameter after sorting (such as "main parameter - saturation" correlated with "color temperature" and "brightness"). Component groups of "main parameter - correlated parameters" are generated (such as the "saturation + color temperature + brightness" component group). The parameter linkage rules are clearly marked within the component group, such as "when the main parameter 'saturation' is increased by more than 10%, the correlated parameter 'color temperature' is automatically decreased by 3% to avoid color distortion," allowing users to clearly understand the chain reaction of parameter adjustments.

[0055] Step 4, Adjust monitoring and push the optimal solution: During the second user's operation of the adjustable components, the system monitors in real time whether the adjusted value exceeds the component's preset constraint threshold (such as the "saturation" constraint threshold of -30% to +30%). If the adjusted value is detected to exceed the threshold (such as increasing saturation by 35%), the AI ​​recommendation module is immediately activated. Based on similar image optimization cases on the platform, user creative preferences, and parameter dependency logic, it generates 3-5 optimal adjustment schemes (such as "saturation +25% and color temperature -5%" and "saturation +28% and brightness +3%). Each scheme is accompanied by a corresponding image effect preview image for users to intuitively select or refer to for adjustment.

[0056] By clarifying parameter linkage rules, providing real-time warnings for exceeding threshold adjustments, and offering visualized optimal solutions, the problems of "unclear parameter associations leading to adjustment conflicts" and "no guidance for adjustments beyond the range" in traditional secondary creation have been solved. This further improves the accuracy, security, and convenience of parameter adjustments, making secondary creation more efficient and better suited to user needs.

[0057] In this embodiment of the application, the method further includes the following steps in the process of generating the difference heatmap before and after adjustment in real time: Step 1, Pixel-level feature depth extraction from dual images: After the second user adjusts the parameters, the system simultaneously obtains the original image before adjustment and the preview image after adjustment. The system then uses computer vision algorithms to perform pixel-level traversal on the two images and extracts three major categories of core feature data: color distribution (RGB channel values, hue ratio), texture structure (edge ​​intensity, detail texture density), and semantic region division (using image segmentation technology to label independent semantic regions such as "sky", "waves", and "people"), providing accurate data support for difference calculation.

[0058] Step 2, Calculation of multi-dimensional difference values ​​and mapping of heatmap: The AI ​​differential quantization model is invoked to calculate the difference values ​​of two images in three major feature dimensions: color, texture, and semantic region (e.g., color difference value = adjusted RGB mean - original RGB mean). The comprehensive difference value of each region is then mapped to the color gradient of a heatmap (preset gradient rules: difference value 0-30 is green, 31-60 is yellow, and above 61 is red; the higher the difference value, the more vibrant the color). This visually presents the range and degree of the adjustment's impact.

[0059] Step 3, Semantic Tag Overlay and Interactive Positioning: In the generated heatmap, semantic tags are overlaid on each high-discrepancy area, with the tag content directly indicating the corresponding parameter adjustment item (such as "saturation +12%" or "sharpening +18%"). At the same time, an association mapping between tags and adjustable components is established. When a user clicks on a semantic tag, the system automatically jumps to and positions the corresponding adjustable component without requiring the user to manually search, achieving "what you see is what you adjust".

[0060] Step 4, continuously adjust the tracking and solidification of reminder triggers: The system monitors the second user's actions in real time. If it detects that the user continuously adjusts the same component group (such as the "Saturation + Color Temperature + Brightness" component group), it dynamically updates the heatmap and simultaneously calculates the cumulative difference value after multiple adjustments (cumulative difference value = weighted sum of the difference values ​​of each adjustment, with the weight decreasing in the order of adjustment). A preset difference threshold (such as 80) is set. When the cumulative difference value exceeds this threshold, the system triggers a "effect solidification reminder" through a pop-up window and sound effects (such as "The current adjustment has significantly optimized the image effect. Do you want to solidify this version?"). At the same time, it automatically saves a snapshot of the parameters of the current adjustment version (including the original value, adjusted value, and adjustment timestamp of all parameters of the component group), supporting subsequent rollback or restoration.

[0061] By continuously adjusting and tracking, calculating cumulative difference values, and solidifying intermediate versions, the problem of "uncontrollable continuous adjustment effects and easy loss of intermediate optimization versions" in traditional secondary creation is solved. This allows users to grasp the adjustment and superposition effects in real time, and retains key optimization nodes through parameter snapshots, further improving the coherence and traceability of secondary creation.

[0062] In this embodiment of the application, the method further includes the following steps in embedding enhanced copyright metadata into the JSON parameter stream: Step 1, Dual Content Fingerprint Extraction: First, the AI ​​content fingerprint generation model is invoked to extract features from the first image (visual dimension) and the generated data stream collected by the S100 (parameter dimension). The visual dimension extracts core visual features such as color distribution, texture features, and semantic region contours of the image, while the parameter dimension extracts core parameter features such as the unique identifier of the generation model, customized random seed, and key adjustment parameters. Then, the two types of features are fused to generate a unique dual content fingerprint (stored in Base64 encoding format to ensure tamper resistance and uniqueness).

[0063] Step 2, Adjustment Intent Description Field Generation: Add a "Adjustment Intent Description" subfield to the basic fields of the contribution history; use a Natural Language Processing (NLP) model to parse the input text of the second user (e.g., "enhance the sense of sky layering") and directly convert it into a structured description; if the user does not input text, parse their adjustment behavior (e.g., continuously adjust "saturation", "color temperature", "sharpening level"), and combine the parameter function semantics to automatically infer the adjustment intent (e.g., "optimize color richness and image sharpness"), ensuring that the intent of each contribution behavior is traceable.

[0064] Step 3, Hierarchical Embedding and Encryption Processing: A dedicated "copyright metadata field" (field name such as "copyright_metadata") is pre-defined in the JSON parameter stream. The enhanced copyright metadata is written into the corresponding sub-fields according to a hierarchical structure: "Original Author Information (Top Layer) - Core Copyright Information (Middle Layer) - Contribution History (Bottom Layer)". The top layer stores the original author identifier and creation timestamp; the middle layer stores the license type and dual content fingerprint; and the bottom layer stores the complete contribution history including the "adjustment intent description". Then, a hash combination of "original author identifier + creation timestamp" is used as the encryption key to symmetrically encrypt the entire "copyright metadata field" to prevent unauthorized tampering or illegal reading.

[0065] Step 4, Dynamic Update of Contribution Score: Each time the second user completes parameter adjustment (i.e., one parameter iteration), the system automatically triggers the AI ​​copyright assessment model, inputs the "adjustment effect improvement value" (a weighted value of aesthetic score + semantic consistency score) and "innovation score" (a repetition rate judgment value compared with the platform's historical adjustment records) calculated in S200, and the model calculates the current contribution score according to the preset algorithm (such as effect improvement value × 0.6 + innovation score × 0.4), and synchronously updates it to the corresponding entry in the contribution history record, ensuring that the quantification of contribution value is synchronized with the creation process in real time.

[0066] By strengthening copyright uniqueness through dual content fingerprinting, ensuring data security through hierarchical encryption, improving contribution tracing by adjusting intent descriptions, and quantifying contribution value through dynamic scoring, the integrity and reliability of copyright metadata have been further solidified.

[0067] In this embodiment of the application, the method further includes the following steps in calculating the contribution based on multi-dimensional indicators: Step 1, Dual Content Fingerprint Extraction: First, the AI ​​content fingerprint generation model is invoked to extract features from the first image (visual dimension) and the generated data stream collected by the S100 (parameter dimension). The visual dimension extracts core visual features such as color distribution, texture features, and semantic region contours of the image, while the parameter dimension extracts core parameter features such as the unique identifier of the generation model, customized random seed, and key adjustment parameters. Then, the two types of features are fused to generate a unique dual content fingerprint (stored in Base64 encoding format to ensure tamper resistance and uniqueness).

[0068] Step 2, adjust the generation of the intent description field: In the basic fields of contribution history, a new subfield "Adjustment Intent Description" has been added; the input text of the second user (such as "enhance the sense of sky layering") is parsed by a natural language processing (NLP) model and directly converted into a structured description; if the user does not input text, their adjustment behavior (such as continuously adjusting "saturation", "color temperature" and "sharpening level") is parsed, and combined with the semantics of parameter functions, the adjustment intent (such as "optimize color richness and image sharpness") is automatically inferred to ensure that the intent of each contribution behavior is traceable.

[0069] Step 3, Hierarchical Embedding and Encryption Processing: A dedicated "copyright metadata field" (field name such as "copyright_metadata") is pre-defined in the JSON parameter stream. The enhanced copyright metadata is written into the corresponding sub-fields according to a hierarchical structure: "Original Author Information (Top Layer) - Core Copyright Information (Middle Layer) - Contribution History (Bottom Layer)". The top layer stores the original author identifier and creation timestamp; the middle layer stores the license type and dual content fingerprint; and the bottom layer stores the complete contribution history including the "adjustment intent description". Then, a hash combination of "original author identifier + creation timestamp" is used as the encryption key to symmetrically encrypt the entire "copyright metadata field" to prevent unauthorized tampering or illegal reading.

[0070] Step 4, Dynamic Update of Contribution Score: Each time the second user completes parameter adjustment (i.e., one parameter iteration), the system automatically triggers the AI ​​copyright assessment model, inputs the "adjustment effect improvement value" (a weighted value of aesthetic score + semantic consistency score) and "innovation score" (a repetition rate judgment value compared with the platform's historical adjustment records) calculated in S200, and the model calculates the current contribution score according to the preset algorithm (such as effect improvement value × 0.6 + innovation score × 0.4), and synchronously updates it to the corresponding entry in the contribution history record, ensuring that the quantification of contribution value is synchronized with the creation process in real time.

[0071] The system strengthens copyright uniqueness through dual content fingerprinting, ensures data security through hierarchical encryption, improves contribution traceability by adjusting intent descriptions, and quantifies contribution value through dynamic scoring.

[0072] In this embodiment of the application, the method further includes the following steps in generating the NFT copyright certificate bound to the hierarchical hash value: Step 1, Hash value association and NFT identifier establishment: First, extract the root hash value (globally unique hash value) of the Merkle tree from the hierarchical hash values ​​generated in step 1 of S400, and use it as the unique identifier (TokenID) of the NFT copyright certificate to ensure that each creation version of the NFT is unique. At the same time, store the hash values ​​of each branch layer (such as the hash values ​​of the core parameter layer and the copyright metadata layer) in the key-value pair format of "layer-hash value" as the metadata attribute of the NFT, providing data support for subsequent on-chain traceability.

[0073] Step 2, embedding the parameter iteration tracing chain: Construct a parameter iteration traceability chain from "original version → all iteration versions", with each iteration node in the chain corresponding to a parameter adjustment. Each node contains three core pieces of information: the corresponding hierarchical hash value (associated with the on-chain evidence record), contributor information (a unique identifier for the second user), and a parameter adjustment summary (such as "saturation +12%, color temperature -3%", extracted from the parameter snapshot of S200). Embed this traceability chain into the metadata extension field of the NFT in JSON format to achieve a full-chain creation trajectory that traces back from the current version to the original version.

[0074] Step 3, Automatic Distribution of Profits from Secondary Transactions: A transaction trigger mechanism is preset for NFT copyright certificates. When an NFT is traded on a trading platform, the system automatically calls the contribution distribution matrix stored on the blockchain (generated in step 3 of S400). According to the proportion of each contributor in the matrix (e.g., original author 30%, secondary creator A 20%, secondary creator B 15%), the transaction revenue (e.g., NFT sales amount) is split and distributed to the corresponding contributor's on-chain account. At the same time, an on-chain transaction record is generated, including the transaction amount, revenue split details, and transaction timestamp, to ensure that the distribution process is transparent and traceable.

[0075] Step 4, Collection of evidence for infringement and rights protection: By monitoring the dissemination of images corresponding to NFTs in real time through blockchain nodes, if suspected infringing content is detected (such as unauthorized copying or commercial use), the system initiates the rights protection evidence extraction process: querying on-chain through the unique identifier (root hash value) of the NFT to retrieve the dual content fingerprint (visual + parametric features) generated by S300 and the parameter dependency graph summary of S100; packaging these data into a standardized evidence package (including on-chain evidence storage record hash and evidence generation timestamp) as the core basis for infringement litigation or platform complaints, ensuring efficient rights protection and legal validity.

[0076] By using strong hash value binding, embedding a full-chain traceability chain, automatically splitting transaction revenue, and solidifying evidence for rights protection, NFT copyright certificates are not only a symbol of digital copyright, but also a core carrier that ensures "traceable creation trajectory, fair revenue distribution, and evidence-based rights protection." This solves the problems of "non-unique identifiers, inconsistent traceability, difficulty in revenue distribution, and lack of irrefutable evidence" in the circulation of digital copyrights, and improves the full lifecycle protection of AI-created copyrights.

[0077] In this embodiment of the application, the method further includes, during the process of embedding cross-platform compatible tags in the JSON file to support multi-terminal sharing: Step 1: Building a mainstream terminal adaptation rule base: First, a technical characteristic survey was conducted on current mainstream terminals (mobile, PC, tablet, and VR devices) to analyze the image rendering capabilities (e.g., maximum resolution supported by mobile devices, 3D rendering protocols of VR devices) and interaction logic (e.g., mobile devices are mainly touch screen operated, while PCs support precise keyboard and mouse operation). Based on the survey results, a rendering rule library adapted to each terminal was generated. The rule library contains the corresponding "rendering parameter thresholds" and "component interaction specifications" for each terminal (e.g., mobile components need to support finger touch size, and PC components need to support batch import and export).

[0078] Step 2, Dual Tag Embedding and Parameter Association: Embed two types of functional tags in the JSON file: one is the terminal type identification tag; the other is the rendering parameter adaptation tag (stored in the field "render_adapt_tag", which includes the image resolution (e.g., 1080P for mobile devices, 4K for PC devices), color space (e.g., sRGB for mobile devices, DCI-P3 for VR devices), and component interaction method (e.g., vertical layout for mobile devices, horizontal layout for PC devices)) for each terminal). The tags are associated with the parameter dependency graph of S100 to ensure that the adaptation parameters are consistent with the core generation parameter logic.

[0079] Step 3, dynamic adaptation after terminal receives: After the second user's terminal receives the JSON file, it automatically reads the "device_type_tag" to identify the terminal type (e.g., iOS mobile device); it calls the corresponding adaptation rules from the rendering rule library to dynamically adjust the presentation of adjustable components, adapting the component size to the screen size (mobile components are enlarged by 20%), optimizing the layout according to the interaction logic (mobile components are arranged vertically), and adapting the operation method to the input device (mobile devices support sliding adjustment, and PC devices support numerical input), achieving "out-of-the-box" functionality without requiring manual settings from the user.

[0080] Step 4, file lightweighting and parameter restoration: After embedding tags, core data is filtered based on parameter dependency graphs, and redundant parameters unrelated to the current image generation (such as weight fragments of other models and invalid parameter default values) are removed. The JSON file is compressed using the LZ77 compression algorithm to reduce transmission bandwidth usage and storage size. At the same time, the "full parameter recovery interface" (the interface address is stored in the field "full_param_recover") is retained. If users need to view or edit all the original parameters later, they can restore the complete data with one click through this interface, balancing lightweight design and data integrity.

[0081] By building a terminal-specific rendering rule library, accurately adapting dual tags, dynamically adjusting component shapes, and balancing lightweight and complete parameters, the problems of "inconsistent parameter presentation, fragmented operation experience, and inefficient file transfer" in traditional cross-platform sharing have been completely solved, achieving convenience in parameter adjustment, consistency of effects, and integrity of data across multiple terminals.

[0082] This application discloses an image sharing device based on an AI platform, referring to... Figure 2 ,include: The cross-platform data stream structure module 001 obtains the generated data stream corresponding to the first image uploaded by the first user. The generated data stream includes the unique identifier of the generation model, the customized random seed, the number of iteration steps, the semantic parsing results of the prompt words, and the model weight slice. The generated data stream is converted into a structured JSON file containing a graph of parameter dependencies. The JSON file has built-in cross-platform compatible tags to support sharing to at least one second user on multiple terminals. The parameter adaptive adjustment visualization module 002, based on the second user's creative preference profile and key input features, uses an AI semantic matching model to filter non-key parameters in the JSON file and generate adjustable components with adaptive priority. Each component is associated with the parameter adjustment constraint threshold and the AI ​​recommended adjustment range, and the default value is dynamically optimized based on the user's historical adjustment behavior. After the second user adjusts the parameters through the adjustable components, a heatmap of the difference before and after the adjustment is generated in real time, and a new image is generated based on the heatmap of the difference and the adjusted parameters. The copyright metadata embedding update module 003 embeds enhanced copyright metadata into the JSON parameter stream. The enhanced copyright metadata includes the original author identifier, creation timestamp, license type, AI-generated content fingerprint information, and contribution history containing contributor identifier, contribution timestamp, contribution type, contribution rating, and adjustment parameter snapshots. The blockchain copyright traceability contribution allocation module 004 calculates the hierarchical hash value of the JSON parameter stream using the Merkle tree algorithm during each parameter iteration, generating a blockchain-based evidence record including version number, timestamp, contributor information, hierarchical hash value, and parameter dependency graph summary. It calculates contribution based on multi-dimensional indicators, including the dimensional complexity of parameter adjustments, deviation of adjustment magnitude, AI-quantified generation effect improvement value, and innovation score of adjustment behavior. The effect improvement value includes aesthetic score and semantic consistency score. Copyright revenue is dynamically allocated based on the contribution percentage, and an NFT copyright certificate bound to the hierarchical hash value is generated. This NFT copyright certificate allows for reverse tracing of the corresponding parameter iteration version via on-chain queries.

[0083] This application also discloses an electronic device, including a processor, wherein the processor runs a program for the image sharing method based on an AI platform as described in any one of the above embodiments.

[0084] This application also discloses a storage medium storing a program for the image sharing method based on an AI platform as described in any one of the above embodiments.

[0085] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for sharing images based on an AI platform, characterized in that, include: Obtain the generated data stream corresponding to the first image uploaded by the first user. The generated data stream includes a unique identifier of the generation model, a customized random seed, the number of iteration steps, the semantic parsing results of the prompt words, and the model weight slices. Convert the generated data stream into a structured JSON file containing a parameter dependency graph. The JSON file has built-in cross-platform compatible tags to support sharing to at least one second user on multiple terminals. Based on the second user's creative preference profile and key input features, non-critical parameters in the JSON file are filtered through an AI semantic matching model to generate adjustable components with adaptive priority sorting. Each component is associated with a constraint threshold for parameter adjustment and an AI-recommended adjustment range, and the default value is dynamically optimized based on the user's historical adjustment behavior. After the second user adjusts the parameters through the adjustable components, a heatmap of the difference before and after the adjustment is generated in real time, and a new image is generated based on the heatmap of the difference and the adjusted parameters. The enhanced copyright metadata is embedded in the JSON parameter stream. The enhanced copyright metadata includes the original author identifier, creation timestamp, license type, AI-generated content fingerprint information, and contribution history including contributor identifier, contribution timestamp, contribution type, contribution rating, and adjustment parameter snapshot. During each parameter iteration, a Merkle tree algorithm is used to calculate the hierarchical hash value of the JSON parameter stream, generating a blockchain-based notarized record that includes a version number, timestamp, contributor information, hierarchical hash value, and a parameter dependency graph summary. Contribution is calculated based on multi-dimensional indicators, including the dimensional complexity of parameter adjustments, deviation of adjustment magnitude, AI-quantified improvement value of generation effect, and innovation score of adjustment behavior. The improvement value includes aesthetic score and semantic consistency score. Copyright revenue is dynamically allocated according to the contribution percentage, and an NFT copyright certificate bound to the hierarchical hash value is generated. The NFT copyright certificate supports reverse tracing of the corresponding parameter iteration version through on-chain queries.

2. The image sharing method based on an AI platform according to claim 1, characterized in that, In the process of generating the adjustable component with adaptive priority ordering, the method further includes: The core semantic vector of the key features of the second user is extracted by the AI ​​semantic matching model, and the cosine similarity is calculated between the core semantic vector of the key features of the second user and the functional semantic vector of each parameter in the JSON file. Parameters with similarity below a preset similarity threshold are marked as non-critical parameters, and the non-critical parameters are prioritized by combining the commonly used adjustment dimensions and adjustment frequency weights in the second user creation preference profile. For each non-critical parameter after sorting, related parameters are identified based on the parameter dependency graph, and a component group of main parameter-related parameters is generated, in which parameter linkage rules are labeled within the component group; The system monitors the adjustment operations of the second user in real time. If the parameter adjustment is detected to exceed the constraint threshold, the AI ​​recommendation module will push multiple optimal adjustment schemes for the user to choose or refer to. The optimal adjustment scheme includes a preview image of the effect of the scheme.

3. The image sharing method based on an AI platform according to claim 1, characterized in that, The method also includes the following steps in the real-time generation of heatmaps showing the differences before and after adjustment: Pixel-level feature extraction is performed on the original image before adjustment and the preview image after adjustment to obtain feature data on color distribution, texture structure, and semantic region division. The AI ​​difference quantization model calculates the difference values ​​between the original image before adjustment and the preview image after adjustment in each feature dimension, and maps the difference values ​​to the color gradient of the heatmap. The higher the difference value, the more vivid the color of the heatmap. Semantic labels are overlaid on the heatmap to mark the parameter adjustment items corresponding to the differences, and the system supports automatic location of the corresponding adjustable component when the user clicks on the semantic label. If a second user continuously adjusts the same component group, the heatmap is updated in real time and the cumulative difference value is calculated. When the cumulative difference value exceeds the preset difference threshold, an effect solidification reminder is triggered and a snapshot of the parameters of the current adjusted version is saved.

4. The image sharing method based on an AI platform according to claim 1, characterized in that, In the process of embedding enhanced copyright metadata into the JSON parameter stream, the method further includes: The first image and the generated data stream are processed by an AI content fingerprint generation model to extract a dual content fingerprint based on visual features and parametric features. Add an adjustment intent description field to the contribution history. The adjustment intent description field is automatically generated by parsing the input text or adjustment behavior of the second user through a natural language processing model. The enhanced copyright metadata is embedded into specified fields of the JSON parameter stream according to the hierarchical structure of original author information - core copyright information - contribution history, and the copyright metadata is encrypted. Each time the parameters are iterated, the contribution score in the contribution history is updated synchronously. The contribution score is dynamically calculated by the AI ​​copyright assessment model based on the improvement value of the adjustment effect and the innovation score.

5. The image sharing method based on an AI platform according to claim 1, characterized in that, In the process of calculating contribution based on multi-dimensional indicators, the method further includes: Set initial weights for each dimension of indicators and dynamically adjust the weights according to the image generation scenario, which includes commercial design, artistic creation, and daily sharing. Calculate the specific values ​​for each indicator, which includes dimensional complexity, deviation of adjustment range, improvement value, and innovation score; The total contribution is calculated based on the adjusted indicator weights. If there are multiple second users, the contribution percentage of each user is calculated separately to generate a contribution distribution matrix.

6. The image sharing method based on an AI platform according to claim 1, characterized in that, In the process of generating the NFT copyright certificate bound to the hierarchical hash value, the method further includes: Based on the Merkle tree algorithm, the root hash value is extracted as the unique identifier of the NFT copyright certificate, and the hash values ​​of each layer are used as the metadata attributes of the NFT. Embed a parameter iteration traceability chain in the NFT copyright certificate. The parameter iteration traceability chain includes all iteration nodes from the original version to the current version. Each node is associated with a corresponding hierarchical hash value, contributor information and parameter adjustment summary. For secondary transactions of NFT copyright certificates, the copyright revenue distribution mechanism is automatically triggered during the transaction. The transaction revenue is split into the accounts of each contributor according to the current contribution distribution matrix, and an on-chain transaction record is generated. If the image content corresponding to an NFT is detected to be infringing, dual content fingerprints and parameter dependency graph summaries are extracted through on-chain queries to serve as core evidence for infringement protection.

7. The image sharing method based on an AI platform according to claim 1, characterized in that, In the process of embedding cross-platform compatible tags in the JSON file to support multi-terminal sharing, the method also includes: We analyze the image rendering capabilities and interaction logic of current mainstream terminals and generate a rendering rule library adapted to each terminal. The current mainstream terminals include mobile terminals, PC terminals, tablet terminals and VR devices. Embed terminal type identification tags and rendering parameter adaptation tags in the JSON file. The rendering parameter adaptation tags include the image resolution, color space, and component interaction method corresponding to each terminal. When the second user receives the JSON file, the terminal automatically reads and identifies the tags, calls the corresponding adaptation rules from the rendering rule library, and dynamically adjusts the size, layout, and operation mode of the adjustable components. The JSON file is lightweighted by removing redundant parameter data based on the parameter dependency graph, compressing the file size using the LZ77 compression algorithm, retaining the complete parameter recovery interface, and allowing users to recover all parameter data when needed.

8. A photo-sharing device based on an AI platform, characterized in that, include: The data stream structured cross-platform module obtains the generated data stream corresponding to the first image uploaded by the first user. The generated data stream includes a unique identifier of the generation model, a customized random seed, the number of iteration steps, the semantic parsing results of the prompt words, and the model weight slices. The generated data stream is converted into a structured JSON file containing a parameter dependency graph. The JSON file has built-in cross-platform compatible tags to support sharing to at least one second user on multiple terminals. The parameter adaptive adjustment visualization module, based on the second user's creative preference profile and key input features, filters non-critical parameters in the JSON file using an AI semantic matching model, generating adjustable components with adaptive priority order. Each component is associated with a constraint threshold for parameter adjustment and an AI-recommended adjustment range, and the default value is dynamically optimized based on the user's historical adjustment behavior. After the second user adjusts the parameters through the adjustable components, a heatmap of the difference before and after the adjustment is generated in real time, and a new image is generated based on the heatmap of the difference and the adjusted parameters. The copyright metadata embedding and updating module embeds enhanced copyright metadata into a JSON parameter stream. The enhanced copyright metadata includes the original author identifier, creation timestamp, license type, AI-generated content fingerprint information, and contribution history including contributor identifier, contribution timestamp, contribution type, contribution rating, and adjustment parameter snapshots. The blockchain copyright traceability contribution allocation module calculates the hierarchical hash value of the JSON parameter stream using the Merkle tree algorithm during each parameter iteration, generating a blockchain-based evidence record including version number, timestamp, contributor information, hierarchical hash value, and parameter dependency graph summary. It calculates contribution based on multi-dimensional indicators, including the dimensional complexity of parameter adjustments, deviation of adjustment magnitude, AI-quantified generation effect improvement value, and innovation score of adjustment behavior. The effect improvement value includes aesthetic score and semantic consistency score. Copyright revenue is dynamically allocated based on the contribution percentage, and an NFT copyright certificate bound to the hierarchical hash value is generated. This NFT copyright certificate supports reverse tracing of the corresponding parameter iteration version via on-chain queries.

9. An electronic device, characterized in that, Includes a processor, wherein the processor runs a program for the image sharing method based on an AI platform as described in any one of claims 1-7.

10. A storage medium, characterized in that, A program storing an AI-based image sharing method as described in any one of claims 1-7.

Images