An AIGC content right confirmation and dynamic account distribution system

Abstract

The application belongs to the technical field of dynamic distribution system, and particularly relates to an AIGC content right confirmation and dynamic distribution system, which comprises an AIGC generation platform or client, a generated track collection module, a multi-modal causal embedding generation topology construction module, a counterfactual re-generation experiment module, a contribution calculation module, a right interest hypergraph construction module, a verifiable re-generation scheduling and proof module, a blockchain interaction module, an on-chain settlement and authorization module and a blockchain network. The system realizes right confirmation and distribution through the following steps: collecting model calling, data input and output and editing operation track information in the AIGC content generation process, constructing a multi-modal causal embedding generation topology based on the track information, which is formalized as a seven-tuple, including node causal embedding, cross-modal alignment tensor, time sequence causal chain index and dynamic causal propagation operator.

Description

Technical Field

[0001] This invention belongs to the technical field of dynamic revenue sharing systems, and particularly relates to an AIGC content ownership confirmation and dynamic revenue sharing system. Background Technology

[0002] With the rapid development of AIGC technology, the creation threshold for various generative content has been significantly lowered. However, the generation process involves multiple contributing factors, such as basic models, fine-tuning plugins, prompt word design, and reference materials. Traditional rights confirmation mechanisms struggle to accurately define the ownership of rights among the participants. Currently, the industry generally lacks the ability to fully record the AIGC generation trajectory and model causal relationships, resulting in a lack of scientific basis for contribution quantification. This leads to frequent disputes among rights holders, severely restricting the commercial circulation of AIGC content.

[0003] Meanwhile, existing revenue-sharing systems mostly adopt a fixed-ratio allocation model, which cannot dynamically adjust the share of revenue based on the actual role of different contribution factors. Furthermore, the revenue-sharing process lacks a transparent and traceable verification mechanism, posing risks such as unfair settlement and data tampering. While blockchain technology provides a decentralized storage solution for rights confirmation, how to efficiently correlate the complex causal relationships of the generation process with on-chain data to achieve reliable calculation of contribution and automated execution of dynamic revenue sharing remains a pressing technical challenge. Summary of the Invention

[0004] The purpose of this invention is to address the aforementioned technical problems by providing an AIGC content ownership confirmation and dynamic revenue sharing system.

[0005] In view of this, the present invention provides an AIGC content ownership confirmation and dynamic revenue sharing system, comprising the following steps: The system includes an AIGC generation platform or client, a generation trajectory acquisition module, a multimodal causal embedding generation topology construction module, a counterfactual regeneration experiment module, a contribution calculation module, a stake hypergraph construction module, a verifiable regeneration scheduling and proof module, a blockchain interaction module, an on-chain settlement and authorization module, and a blockchain network. The system achieves ownership confirmation and revenue sharing through the following steps: Collect information on model calls, data input / output, and editing operation trajectories during the AIGC content generation process; A multimodal causal embedding topology is constructed based on trajectory information, which is formalized as a seven-tuple T=(V,E,Φ,A,τ,e,θ), containing node causal embedding, cross-modal alignment tensor, temporal causal chain index and dynamic causal propagation operator; Based on the MCEGT, a counterfactual experiment matrix is ​​generated, a counterfactual MCEGT is constructed, and off-chain counterfactual regeneration is performed to obtain multiple sets of counterfactual output content. Feature encoding and difference analysis are performed on the original output and counterfactual output. Combined with the gradient contribution calculated by automatic differentiation, the comprehensive contribution weight of each contributing factor is obtained. Construct an equity hypergraph that includes work nodes, contribution factor nodes, rights holder nodes, and hyperedges with contribution weights, and generate equity hypergraph commitment values ​​through a Merkle tree; Write the artwork identifier, artwork hash, and equity supergraph commitment value into the blockchain, and use smart contracts to maintain the authorization state machine and realize revenue splitting and settlement. The counterfact regeneration task is scheduled and sampled off-chain, and aggregated proofs are generated and stored on-chain.

[0006] Preferably, the construction of the multimodal causal embedding topology includes: creating model nodes, operation nodes, and data nodes and extracting local feature vectors; establishing directed edges based on the data flow, and configuring a dynamic causal propagation operator Φ for each edge. ij =MLP ij (e j ;θ ij )⊙g ij A unified mapping for multimodal node embeddings is achieved by comparing and learning a pre-trained cross-modal alignment tensor A; the temporal causal chain index τ of each node is calculated. i =(t i ,l i ,s i ); Based on the causal attention mechanism, recursively compute the causal embedding of nodes e i .

[0007] Preferably, the contribution factor includes the model class contribution factor F. model LoRA model contribution factor F lora , Prompt word class contribution factor F prompt Reference material contribution factor F ref and operational contribution factor F op The set of contribution factors is represented as F = {f1, f2, ..., f...} n}

[0008] Preferably, the counterfactual regeneration experiment includes: configuring baseline values ​​for various contribution factors, generating an experimental matrix containing single-factor replacement and combined-factor replacement; copying the original MCEGT and replacing the corresponding factors with the baseline configuration to obtain the counterfactual MCEGT; performing regeneration in an environment consistent with the original generation, and recording the task ID, model version, and random seed metadata.

[0009] Preferably, the contribution calculation includes: mapping the output content to a feature vector using a feature encoding model corresponding to the modality, and calculating the difference value Δ using cosine distance, Euclidean distance, or perceptual similarity. iNormalizing the difference values ​​yields the counterfactual contribution c. i The gradient contribution is obtained by calculating the gradient of the contribution with respect to the causal embedding of the node. ;pass The fusion yields the overall contribution, where λ∈[0,1].

[0010] Preferably, the serialized record of the hypergraph of interests is a quadruple R. k =(WorkID,FactorID_k,ActorAddr k ,c k Merkle leaf node hash h k =H(WorkID||FactorID k ||ActorAddr k ||c k The root hash Root_H is calculated from the bottom up and used as the commitment value of the hypergraph of equity.

[0011] Preferably, the state set S={S0(Original),S1(PlatformUse),S2(TrainingUse),S3(SubLicense),…} maintained by the on-chain authorization state machine is triggered by state transition functions δ:S×Event,S' responding to events such as use, train, and sublicense, and the state transition is recorded in the on-chain log.

[0012] Preferably, the calculation formula for the profit split and settlement is r. i =R×c i ×α i Where R is the total revenue, c i As the contribution weight, α i The profit coefficient is Σ i (c i ×α i =1; the final benefit of the rights holder The on-chain settlement is automatically calculated and completed by the smart contract.

[0013] Preferably, the verifiable regeneration scheduling and proof includes: sampling a portion of the regeneration tasks according to a sampling rate p and allocating them to the verification nodes, with a sampling quantity n. verify =⌈p×n total ⌉; Verification nodes verify the validity of tasks by comparing the hash results; aggregate the verified task signatures to form a lightweight proof π, which is stored on the blockchain in association with the stake hypergraph commitment value.

[0014] The beneficial effects of this invention are: This invention achieves deep causal modeling and semantic alignment of multimodal contribution factors in the AIGC generation process through innovative multimodal causal embedding topology generation. Combined with counterfactual regeneration experiments and gradient analysis for dual-dimensional contribution calculation, it accurately quantifies the equity proportion of each participant, completely resolving the industry pain point of traditional mechanisms' difficulty in defining the ownership of multiple stakeholders' rights. The combination of the equity supergraph and blockchain Merkle commitments transforms complex equity relationships into on-chain trusted credentials. Coupled with a smart contract-driven authorization state machine, it automates and makes the rights confirmation process transparent, effectively avoiding equity disputes and providing core technical support for the compliant circulation of AIGC content.

[0015] Meanwhile, the introduction of off-chain distributed regeneration and sampling verification mechanisms significantly reduces the on-chain storage and computational burden while ensuring the reliability of contribution calculation results. The dynamic revenue splitting algorithm can adjust the revenue sharing ratio in real time based on actual contributions, offering greater fairness and flexibility compared to a fixed revenue sharing model. The entire system establishes a closed loop encompassing the entire process of generation trajectory collection, contribution quantification, rights confirmation, and revenue settlement. This not only improves the efficiency of AIGC content commercialization but also incentivizes participants in all aspects, including model development and material creation, promoting the healthy and sustainable development of the AIGC industry ecosystem. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall architecture of the blockchain-based AIGC content ownership confirmation and dynamic revenue sharing system provided in this embodiment of the invention. It shows the interaction relationship and data flow between the AIGC generation platform, the generation trajectory acquisition module, the MCEGT construction module, the counterfactual regeneration experiment module, the contribution calculation module, the equity hypergraph construction module, the verifiable regeneration scheduling and proof module, the blockchain interaction module, the on-chain settlement and authorization module, and the blockchain network.

[0017] Figure 2 The schematic diagram of multimodal causal embedding generation topology construction provided in the embodiments of the present invention illustrates the method of abstracting the AIGC generation process into a seven-tuple T=(V,E,Φ,A,τ,e,θ), including a visual representation of innovative features such as node causal embedding, cross-modal alignment tensor, temporal causal chain index, and dynamic causal propagation operator.

[0018] Figure 3 This is a schematic diagram of the counterfactual regeneration experiment provided in an embodiment of the present invention, illustrating the generation of the experimental matrix M, the construction of the counterfactual MCEGT, and the acquisition of the counterfactual output.

[0019] Figure 4 The schematic diagram of the contribution calculation process provided in the embodiments of the present invention illustrates the process of feature encoding, difference measurement, normalization to obtain contribution weight, and calculation of gradient contribution based on MCEGT.

[0020] Figure 5 The diagram illustrates the stake hypergraph structure and Merkle tree commitment calculation provided in this embodiment of the invention, showing the composition of nodes and hyperedges of the stake hypergraph H, as well as the method for generating the root hash Root_H.

[0021] Figure 6 The diagram illustrates the verifiable regeneration scheduling and proof architecture provided in this embodiment of the invention, showing the process of task scheduling, sampling verification, and proof aggregation π. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0023] The AIGC generation process is formalized as a multimodal causal embedding generation topology T=(V,E,Φ,A,τ,e,θ). This high-order data structure not only includes model nodes, data nodes, and operation nodes, but also achieves a deep semantic representation of the generation process through innovative features such as node causal embedding mechanism, cross-modal alignment tensor, temporal causal chain index, and dynamic causal propagation operator, overcoming the limitations of traditional directed acyclic graphs in causal modeling and cross-modal alignment.

[0024] Secondly, based on the multimodal causal embedding generation topology, a series of counterfactual generation configurations are constructed to remove or replace a certain contributing factor. Counterfactual regeneration is performed while maintaining the consistency of the model environment, resulting in multiple sets of counterfactual outputs. .

[0025] Next, a feature encoding model is used to compare the original output Y with the counterfactual output. Mapping to a unified feature space, calculating the difference between the two, and normalizing to obtain the contribution weights C={c1,c2,c...} of each contributing factor. n Meanwhile, leveraging the end-to-end differentiability of MCEGT, the gradient of contribution to the causal embedding of each node is automatically differentiated, providing additional semantic basis for the interpretation of contribution.

[0026] Then, AIGC works, contribution factors, and rights holders are uniformly modeled into a rights hypergraph H=(V H E HVertices include work nodes, contribution factor nodes, and rights holder nodes; directed hyperedges represent the relationship of a group of contribution factors acting together on a certain work, and carry attributes such as contribution weight on the hyperedges; then, the entire rights hypergraph is compressed and committed through a Merkle tree, and a small number of commitment values ​​such as the root hash are written into the blockchain.

[0027] Next, an on-chain authorization state machine and revenue sharing logic are constructed using smart contracts. For each work, the following states are maintained: original authorization, usage authorization, training use authorization, sub-licensing, etc. Upon detecting a call or revenue event, the revenue share *r* of each rights holder is calculated based on the work's corresponding equity hypergraph commitment value and its expansion result. i And complete on-chain settlement and authorization status update.

[0028] Finally, a verifiable regeneration scheduling and proof mechanism is introduced to distribute the counterfactual regeneration task off-chain to multiple regeneration nodes for execution. Some tasks are sampled for verification and the results are aggregated and signed to form a lightweight proof that can be quickly verified on-chain, which reduces the on-chain burden while ensuring the credibility of the contribution calculation results.

[0029] The AIGC content ownership confirmation and dynamic revenue sharing system based on blockchain provided in this invention mainly includes the following modules: AIGC generation platform or client (used to provide content generation capabilities for end users), generation trajectory acquisition module, multimodal causal embedding generation topology construction module, counterfactual regeneration experiment module, contribution calculation module, equity hypergraph construction module, verifiable regeneration scheduling and proof module, blockchain interaction module, on-chain settlement and authorization module, and blockchain network.

[0030] The system's workflow is roughly as follows: Users initiate content generation requests through the AIGC generation platform, which internally calls basic models, fine-tuned models, and plugin models to execute the generation. The generation trajectory acquisition module records model calls, data input / output, and editing operations during the generation process through SDKs, API hooks, or log collection. The multimodal causal embedding generation topology construction module constructs an MCEGT structure based on the collected trajectory information, including node causal embedding, cross-modal alignment tensors, temporal causal chain indexes, and dynamic causal propagation operators. After generation, the counterfactual regeneration experiment module generates a counterfactual experiment matrix based on the MCEGT and performs counterfactual regeneration in an off-chain computing environment. The contribution calculation module performs feature encoding and difference analysis on the original output and counterfactual output to obtain the contribution weight of each contribution factor. The equity hypergraph construction module constructs an equity hypergraph based on this and generates equity hypergraph commitment values ​​through a Merkle tree structure. The blockchain interaction module writes the work identifier, work hash, and equity hypergraph commitment value into the blockchain. When a work is invoked or licensed, the on-chain settlement and licensing module automatically calculates the revenue split and completes the settlement based on the stake hypergraph and contribution weight, while updating the licensing status. The verifiable regeneration scheduling and proof module is responsible for coordinating the distribution and sampling verification of counterfactual regeneration tasks, and storing the aggregated proof and stake hypergraph commitments on-chain.

[0031] A multimodal causal embedding generative topology (MCEGT) is proposed as a high-order data structure to describe the AIGC generation process. The formal definition of MCEGT is a seven-tuple T=(V,E,Φ,A,τ,e,θ), and its construction process includes the following steps.

[0032] Regarding node creation and local feature extraction: When the AIGC platform calls a base model or fine-tuned model for inference, the trajectory acquisition module creates the corresponding model node v. model =(type model ,attr model ,h model ), where type model The node type is identified as a model call, attr model Includes attribute information such as model identifier, version number, and model provider address, h model The local feature vectors invoked by this model are obtained by embedding the model configuration parameters; when the platform performs a specific operation (such as generating an image, local redrawing, or applying a filter), an operation node v is created. op =(type op ,attr op ,h op ), where h opThis is obtained through embedding encoding of operation types and parameters; data nodes are created for input text, prompts, reference images, intermediate results, and final output. data =(type data ,attr data ,h data ), where h data Features are extracted through a pre-trained encoder for the corresponding modality.

[0033] Regarding the establishment of edge relationships and dynamic causal propagation operators: Based on the actual data flow, the flow of data from upstream nodes to downstream nodes is recorded as directed edges e. ij ∈E; Unlike traditional directed acyclic graphs, MCEGT assigns a dynamic causal propagation operator Φ to each edge. ij :R d →R d The definition of the dynamic causal propagation operator is as follows: Φ ij (e j )=MLP ij (e j ;θ ij )⊙g ij (Formula 1); MLP ij For edge-specific multilayer perceptrons, the parameter θ ij During MCEGT construction, g is initialized based on the edge type and context. ij Let g be the gate vector, and ⊙ represent element-wise multiplication. ij Calculated using the following formula: g ij =sigmoid(W g ·[e i ;e j ;f ij ]+b g )(Formula 2); Where f ij W represents the static feature vector of the edge (such as data type, transmission volume, operation latency, etc.). g and b g These are learnable parameters. The design of the dynamic causal propagation operator allows the propagation strength of causal effects to be adaptively adjusted according to the specific states of the source and target nodes, rather than using fixed propagation rules.

[0034] Regarding the construction of cross-modal alignment tensors: Since the AIGC generation process involves multiple modalities such as text, images, audio, and video, the local features h_i of nodes in different modalities have heterogeneous feature spaces. MCEGT introduces a cross-modal alignment tensor A∈R. (M ×d×d）To address the semantic gap problem, we use M to represent the number of modalities and d to represent the uniform embedding dimension. For the node embedding of modality m, the aligned uniform embedding is calculated as follows: (Formula 3); The cross-modal alignment tensor A is obtained through pre-training using a contrastive learning approach. Specifically, a set P of positive sample pairs (semantically similar cross-modal node pairs) is constructed, and the alignment loss function is defined as follows: (Formula 4); Where τ is the temperature parameter. By minimizing the alignment loss, we ensure that semantically similar content from different modalities has similar representations in a unified embedding space.

[0035] Regarding the computation of the temporal causal chain index: MCEGT assigns a temporal causal chain index τ to each node. i =(t i ,l i ,s i Its calculation follows these rules: t i =Actual execution timestamp (Formula 5); l i =max j∈Parent(i) (l j )+1, if Then l i =0 (Formula 6); s i =|{v k :l k =l i ∧t k <t i}|+1 (Formula 7); Based on the temporal causal chain index, the temporal causal distance between nodes is defined as: d temporal (v i ,v j )=|l i -l j |+β·|t i -t j | / Δt max (Formula 8); Where β is the time normalization coefficient, Δt max This represents the total duration of the generation process. The introduction of temporal causal chain index and temporal causal distance enables MCEGT to support temporal counterfactual analysis, i.e., to evaluate the impact of changes in the execution timing of a node on the final output, a function that traditional directed acyclic graphs cannot achieve.

[0036] Regarding the recursive computation of node causal embeddings: One of the core features of MCEGT is the computation of a d-dimensional causal embedding vector e for each node. i ∈R d Causal embeddings are recursively propagated from upstream nodes through a causal attention mechanism, and the calculation formula is as follows: e i =σ(W e ·[h i ;Σ j∈Parent(i) α ij ·Φ ij (e j )]+b e ) (Formula 9) Where h i Let Φ be the local feature vector of node i, and let Parent(i) be the set of direct predecessor nodes of node i in the MCEGT. ij For the dynamic causal propagation operator from node j to node i, α ij For causal attention weights, W e and b e σ is a learnable parameter, and σ is a non-linear activation function (such as ReLU or GELU). Causal attention weights α ij Calculated using a scaled dot product attention mechanism: (Formula 10); q i =W q ·h i k j =W k ·h j (Formula 11); Where q i Let k be the query vector for node i. j Let d be the key vector of node j. k W represents the dimension of the key vector. q and W k These are learnable parameters. This causal attention mechanism enables the causal embedding of each node to aggregate causal influences from all upstream nodes, and the aggregation weights are dynamically determined based on the semantic relevance between nodes, forming a globally consistent causal representation.

[0037] Regarding contribution factor labeling: For each base model, fine-tuning model, LoRA model, control network, etc., a model class contribution factor label F is generated. model The corresponding model nodes are then labeled; the complete prompt word is segmented or semantically fragmented, and each fragment generates a prompt word class contribution factor F. prompt ; Generate a reference material class contribution factor F for the reference images, audio, video, and other materials used in the generation. ref; Contribution factor F of key editing operations affecting the generated results (generation operation class) op The set of contribution factors can be represented as: F={F model ,F lora ,F prompt ,F ref ,F op ,..}={f1,f2,.f n} (Formula 12); Regarding the differentiable topology: The overall structure of MCEGT supports end-to-end gradient propagation, allowing contribution calculation to be directly obtained through automatic differentiation. Specifically, the contribution function C: MCEGT → R is defined. n The function, which maps from MCEGT to the contribution vectors of each contribution factor, is implemented through the following differentiable computational graph: (Formula 13); Among them ê i For the causal embedding of nodes after cross-modal alignment, γ i We aggregate weights for the contribution of node i, and calculate them using a differentiable attention mechanism: (Formula 14); in This is the causal embedding of the final output node. Since the entire computation process is differentiable, it can be calculated... That is, the gradient of the contribution with respect to the MCEGT parameters, thereby supporting gradient-based contribution interpretation and optimization.

[0038] The counterfactual regeneration experiment process in this embodiment includes the following steps.

[0039] Baseline configuration definition: For each type of contribution factor f i Pre-configure baseline value baseline(f) i The baseline for model-related factors is to not load the LoRA or plugin, or to use the default configuration of the base model; the baseline for prompt-related factors is to delete the fragment or replace it with neutral placeholder text; the baseline for reference material-related factors is to not use the material or to use a uniform standard reference material; the baseline for operation-related factors is to cancel the operation step or use the default parameters.

[0040] Experimental matrix generation: The counterfactual regeneration experimental module generates the matrix based on the list of contribution factors {f1, f2, ..., f...} contained in the MCEGT. n Generate the experiment matrix M. The experiment matrix is ​​defined as follows: M=[m ij ]_{k×n}, where m ij =1 indicates that factor f in the i-th experiment jReplaced with the baseline (Formula 15); Each row corresponds to a counterfactual experiment configuration; it includes at least an experiment configuration that masks or replaces each individual contributing factor (single-factor experiment, matrix in identity matrix form); optionally, a partial contributing factor combination experiment configuration (combined factor experiment) is introduced to evaluate the interaction contribution between factors.

[0041] Counterfactual MCEGT construction: For the configuration of the i-th row in the experiment matrix, copy the original MCEGT structure T, and for m... ij =1 contribution factor f j Perform the operation to replace the node or parameter corresponding to the factor with the baseline configuration (f). j Simultaneously, the local feature vector h and causal embedding e of the affected nodes are updated to obtain the counterfactual multimodal causal embedding generative topology. .

[0042] Counterfactual regeneration execution: Load the same version of the model and runtime environment as the original generation on the off-chain regeneration node, and execute according to counterfactual... Execute the generation process to obtain counterfactual output. To ensure verifiability, the regenerated node records metadata such as task ID, model version used, and random seed.

[0043] 3.2.4 Contribution Calculation Module; The contribution calculation process in this embodiment includes the following steps.

[0044] Feature Encoding: For image-based AIGC content, the contribution calculation module calls a visual encoding model (such as CLIP visual encoder) to encode the original output Y into a feature vector v, and outputs the counterfactual data. Encode as feature vector The feature encoding process can be represented as: (Formula 16); For text-based content, the same processing is performed using the text encoding model; for audio, video, etc., the corresponding feature encoding model can be selected respectively.

[0045] Difference metric: For each set of original feature vectors v, compare the original feature vectors with the counterfactual feature vectors. Calculate the difference value Δ i Difference measurement function It can be at least one of cosine distance, Euclidean distance, or perceptual similarity metrics: (Formula 17); The cosine distance is defined as: (Formula 18); The Euclidean distance is defined as: (Formula 19); Contribution normalization: Normalize the differences in all contribution factors to obtain the contribution weight c. i : (Formula 20); Where ε is a very small positive number to prevent the denominator from being zero (e.g., ε = 10). -8 The normalized contribution weights satisfy: (Formula 21); When it is necessary to distinguish the contributions of different dimensions such as semantics, style, and composition, multiple encoding models or multi-head features can be used to calculate the differences in each dimension separately. Then combine them into a multidimensional contribution vector: (Formula 22); Gradient Contribution Analysis Based on MCEGT: In addition to calculating the contribution through counterfactual experiments, this invention also utilizes the end-to-end differentiability of MCEGT to automatically differentiate and calculate the gradient of the contribution to the causal embedding of each node. The gradient contribution is defined as: (Formula 23); Where L output The objective function for the output content (e.g., the distance from the desired output), e i The causal embedding of node i. Gradient contribution. This reflects the sensitivity of each node to the final output and can be used as the counterfactual contribution c. i Supplementary verification or weighting factors are required. The final overall contribution can be calculated using the following formula: (Formula 24); Where λ∈[0,1] is the fusion coefficient of counterfactual contribution and gradient contribution.

[0046] The process of constructing the rights hypergraph in this embodiment includes the following steps.

[0047] Node creation: Create a work node for each AIGC work. work Record attributes such as work ID, work hash H(Y), and generation time t; create a contribution factor node for each contribution factor. Record factor ID, factor type, associated model or material hash, etc.; create a rights holder node for each rights holder. Record the subject's DID or blockchain address addr j The set of vertices in a hypergraph of interests is defined as follows: (Formula 25); Hyperedge creation: For work W, collect all contribution factors involved in the generation and their corresponding contribution weights {(f i ,c i Construct a directed hyperedge e from the set of contributing factor nodes to the work nodes, and record the contribution weight of each factor on this hyperedge. The hyperedge is defined as: (Formula 26); Rights Hypergraph Serialization: Serializes the work node, contribution factor node, rights holder node, and the aforementioned hyperedges and related edges into several records R. k Each record is a quadruple: R k =(WorkID,FactorID k ActorAddr k ,c k )(Formula 27); Merkle commitment computation involves: calculating the hash of each record, which becomes a leaf node in the Merkle tree; and then calculating the hashes of parent nodes from bottom to top until the Merkle root hash is obtained. The specific calculation process is as follows: h k =H(R k =H(WorkID||FactorID) k ||ActorAddr k ||c k )(Formula 28); h parent =H(h left ||h right )(Formula 29); Root H =MerkleRoot({h1,h2,...h n})(Formula 30); The root hash is denoted as the Hypergraph Commitment Value Root. H It can also record the number of participating nodes | V H | Generate timestamps and other summary information.

[0048] The on-chain settlement and authorization state machine in this embodiment mainly includes an authorization state machine submodule and a revenue splitting submodule.

[0049] Regarding the licensing state machine: For each work W, the licensing state machine contract maintains multiple licensing states S, and the set of states is defined as follows: S={S0(Original),S1(PlatformUse),S2(TrainingUse),S3(SubLicense),..} (Formula 31); The state transition function is defined as: δ:S×Event→S', where Event={use,train,sublicense,revoke,..} (Formula 32); When an event related to a work is detected (such as platform display, content call, training set sampling, or sublicensing request), a state transition is triggered based on the event type and the current state, and the corresponding on-chain event log is recorded; different authorization states can restrict subsequent usage behavior.

[0050] Regarding revenue splitting and settlement: For each revenue event related to the work (e.g., playback revenue, licensing fees, training usage revenue), the revenue splitting submodule receives the total revenue amount R and the work identifier WorkID. The revenue splitting calculation formula is as follows: r i =R×c i ×α i (Formula 33); Where R is the total revenue, and c i Let α be the contribution weight of the i-th contributing factor. i The revenue coefficient for contributing factors in this category (a basic revenue sharing ratio used to distinguish different types of entities such as model providers, data providers, creators, and platforms). The revenue coefficient must meet the following constraints: (Formula 34); The final profit share for each rights holder is as follows: (Equation 35).

[0051] This embodiment achieves lightweight verification of the counterfactual regeneration calculation results through a verifiable regeneration scheduling and proof architecture.

[0052] Task Scheduling: The scheduling node receives the experiment matrix M generated by the counterfactual regeneration experiment module and treats each experiment configuration as an independent task. i Assign to multiple off-chain regenerated nodes {Node1, Node2, .Node k Simultaneously, a portion of tasks are randomly selected from all tasks in proportion p as sampling verification tasks and assigned to independent verification nodes. The sampling quantity is calculated as follows: n verify =⌈p×n total ⌉, where p is the sampling rate (e.g., p=0.1) (Formula 36); Regeneration execution: Each regenerated node j Based on the assigned Task i Load the specified model version and parameters, perform counterfactual generation, and output the counterfactual results. And calculate the hash of the result. The node is regenerated using its own private key sk. j Digitally sign the task digest: (Formula 37); Sampling verification: The verification node independently recalculates the sampling verification task and generates its own result hash h. verify Compare the current result with the result of the regenerated node: (Formula 38); If the results are consistent, the task is considered to have passed verification; otherwise, it is marked as a failure and the corresponding regenerated node is penalized or removed.

[0053] Proof Aggregation: Aggregate the verified task signatures to form an aggregated proof π: π=Aggregate({σ1,σ2,.σ k},{TaskSummary},{VerifyResults}) (Formula 39); The aggregated proof π is associated with the corresponding stake hypergraph commitment value Root_H and stored on the blockchain: OnChainRecord=(WorkID,Root H ,π,timestamp) (Formula 40).

[0054] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A system for confirming ownership and dynamically distributing revenue from AIGC content, characterized in that: It includes an AIGC generation platform or client, a generation trajectory acquisition module, a multimodal causal embedding generation topology construction module, a counterfactual regeneration experiment module, a contribution calculation module, a stake hypergraph construction module, a verifiable regeneration scheduling and proof module, a blockchain interaction module, an on-chain settlement and authorization module, and a blockchain network; The system achieves ownership confirmation and revenue sharing through the following steps: Step 1: Collect information on model calls, data input / output, and editing operation trajectories during the AIGC content generation process; Step 2: Construct a multimodal causal embedding generation topology based on trajectory information, which is formalized as a seven-tuple T=(V,E,Φ,A,τ,e,θ), including node causal embedding, cross-modal alignment tensor, temporal causal chain index and dynamic causal propagation operator; Step 3: Generate a counterfactual experiment matrix based on MCEGT, construct a counterfactual MCEGT and perform off-chain counterfactual regeneration to obtain multiple sets of counterfactual output content; Step 4: Perform feature encoding and difference analysis on the original output and counterfactual output, and combine the gradient contribution calculated by automatic differentiation to obtain the comprehensive contribution weight of each contributing factor. Step 5: Construct a hypergraph of rights and interests containing work nodes, contribution factor nodes, rights holder nodes, and hyperedges with contribution weights, and generate the commitment value of the hypergraph of rights and interests through a Merkle tree; Step Six: Write the artwork identifier, artwork hash, and equity supergraph commitment value into the blockchain, and maintain the authorization state machine and realize revenue splitting and settlement through smart contracts; Step 7: Perform off-chain scheduling and sampling verification of the counterfactual regeneration task, generate aggregate proofs, and store them on-chain.

2. The AIGC content ownership confirmation and dynamic revenue sharing system according to claim 1, characterized in that: The construction of the multimodal causal embedding topology includes: creating model nodes, operation nodes, and data nodes and extracting local feature vectors; establishing directed edges based on the data flow, and configuring a dynamic causal propagation operator Φ for each edge. ij =MLP ij (e j ;θ ij )⊙g ij A unified mapping for multimodal node embeddings is achieved by comparing and learning a pre-trained cross-modal alignment tensor A; the temporal causal chain index τ of each node is calculated. i =(t i ,l i ,s i ); Based on the causal attention mechanism, recursively compute the causal embedding of nodes e i .

3. The AIGC content ownership confirmation and dynamic revenue sharing system according to claim 1, characterized in that: The contribution factors include model-class contribution factor F. model LoRA model contribution factor F lora , Prompt word class contribution factor F prompt Reference material contribution factor F ref and operational contribution factor F op The set of contribution factors is represented as F = {f1, f2, ..., f...} n } 4. The AIGC content ownership confirmation and dynamic revenue sharing system according to claim 1, characterized in that: The counterfactual regeneration experiment includes: configuring baseline values ​​for various contribution factors, generating an experimental matrix containing single-factor replacement and combined-factor replacement; copying the original MCEGT and replacing the corresponding factors with the baseline configuration to obtain the counterfactual MCEGT; performing regeneration in an environment consistent with the original generation, and recording the task ID, model version, and random seed metadata.

5. The AIGC content ownership confirmation and dynamic revenue sharing system according to claim 1, characterized in that: The contribution calculation includes: mapping the output content into a feature vector using a feature encoding model corresponding to the modality, and calculating the difference value Δ using cosine distance, Euclidean distance, or perceptual similarity. i Normalizing the difference values ​​yields the counterfactual contribution c. i The gradient contribution is obtained by calculating the gradient of the contribution with respect to the causal embedding of the node. ;pass The fusion yields the overall contribution, where λ∈[0,1].

6. The AIGC content ownership confirmation and dynamic revenue sharing system according to claim 1, characterized in that: The serialized record of the hypergraph of interests is a quadruple R. k =(WorkID,FactorID_k,ActorAddr k ,c k Merkle leaf node hash h k =H(WorkID||FactorID k ||ActorAddr k ||c k The root hash Root_H is calculated from the bottom up and used as the commitment value of the hypergraph of equity.

7. The AIGC content ownership confirmation and dynamic revenue sharing system according to claim 1, characterized in that: The on-chain authorized state machine maintains a state set S={S0(Original),S1(PlatformUse),S2(TrainingUse),S3(SubLicense),…}. Through the state transition function δ:S×Event,S' responds to events such as use, train, and sublicense, triggering state transitions and recording them in the on-chain log.

8. The AIGC content ownership confirmation and dynamic revenue sharing system according to claim 1, characterized in that: The formula for calculating the profit split and settlement is r. i =R×c i ×α i Where R is the total revenue, c i As the contribution weight, α i The profit coefficient is Σ i (c i ×α i =1; the final benefit of the rights holder The on-chain settlement is automatically calculated and completed by the smart contract.

9. The AIGC content ownership confirmation and dynamic revenue sharing system according to claim 1, characterized in that: The verifiable regeneration scheduling and proof includes: sampling a portion of the regeneration tasks according to a sampling rate p and allocating them to the verification nodes, with the sampling quantity... Verification nodes verify the validity of tasks by comparing the hash results; aggregate the verified task signatures to form a lightweight proof π, which is stored in the blockchain in association with the stake hypergraph commitment value.

Images