Tibetan type cultural creative graph automatic generation method and system based on artificial intelligence

Abstract

The invention discloses a Tibetan type cultural creative graph automatic generation method and system based on artificial intelligence, and relates to the field of Tibetan type cultural creative graphs, and the method comprises the steps: carrying out the semantic analysis of a natural language description, and extracting a feature vector and a core keyword of a user intention; constructing a Tibetan culture symbol gene pool; based on the vector similarity, an initial culture symbol set conforming to the design intention of the user is screened out; eliminating symbol combinations with culture conflicts, and recommending symbol combination schemes conforming to culture specifications; extracting the shape feature of each symbol as an invariant constraint, and generating a symbol graph set after shape self-adaptive adjustment; and based on the diffusion model framework, generating a final Tibetan cultural and creative pattern. And outputting a structure report of the current Tibetan cultural and creative pattern according to the generated Tibetan cultural and creative pattern. The method has the advantages that natural language description can be automatically converted into Tibetan cultural and creative patterns conforming to culture specifications and aesthetic requirements, and deep fusion of culture inheritance and intelligent generation is achieved.

Description

A method and system for automatically generating Tibetan-style cultural and creative images based on artificial intelligence Technical Field

[0001] This invention relates to the field of Tibetan cultural and creative designs, specifically to an automatic generation method and system for Tibetan cultural and creative designs based on artificial intelligence. Background Technology

[0002] With the booming development of the cultural and creative industries, the market demand for cultural and creative products with profound ethnic cultural heritage is growing. Tibetan culture, as an important part of the treasure of Chinese culture, features intricate and gorgeous patterns, strong color contrasts, and profound symbolic meanings. It is widely used in Thangka paintings, clothing, architectural decorations, and handicrafts. However, the current field of Tibetan cultural and creative design faces challenges: First, the design process relies heavily on the personal experience and artistic inspiration of senior designers. These designers need to undergo long-term study to deeply understand the complex meanings, strict combination rules, and taboos of Tibetan cultural symbols, resulting in extremely high design thresholds and a shortage of talent. Second, traditional design models are inefficient, with long cycles from conception and sketching to final design, making it difficult to meet the needs of rapid iteration and personalized customization in the modern market.

[0003] In recent years, generative artificial intelligence (AIGC) technology has made groundbreaking progress in the field of graphic generation. Among them, diffusion models (such as Stable Diffusion) can generate high-quality graphics based on text prompts. However, general AIGC tools have inherent limitations when applied to the creation of specific ethnic cultures: they cannot understand the deep symbolic semantics and structural rules of Tibetan culture, and the generated results cannot guarantee cultural accuracy. They often suffer from problems such as misuse of symbols, chaotic combinations, and even violations of cultural taboos. Their output has great randomness and uncontrollability. Therefore, there is an urgent need in this field for a technical solution that can combine the deep logic of Tibetan culture with the powerful generative capabilities of AIGC. It needs to be not a simple "pattern generator", but an intelligent design tool that incorporates cultural knowledge. It should be able to automatically follow ethnic cultural norms and efficiently and accurately generate Tibetan cultural and creative patterns that are both culturally orthodox and creative, based on understanding the user's abstract intentions.

[0004] Existing technologies rely solely on the prompting word engineering of text-based image models, making it difficult to accurately capture the abstract, culturally metaphorical intentions in the user's mind. This leads to deviations between the generated results and the user's expectations. Secondly, existing general AIGC models (such as Midjourney and Stable Diffusion) have massive training data and lack a structured understanding of the Tibetan cultural symbol system. They generate patterns based on pixel-level statistical rules rather than an understanding of the cultural symbol's meaning. This results in their output failing to guarantee cultural correctness and seriousness, easily leading to problems such as misattribution of symbols, improper combinations, or even violations of cultural taboos, greatly limiting their direct application in the professional cultural and creative fields. Summary of the Invention

[0005] To address the aforementioned technical issues, this paper provides an AI-based method and system for automatically generating Tibetan cultural and creative images. This solution overcomes the limitations of the prompt word engineering method mentioned in the background, which relies on the image model and struggles to accurately capture the abstract, culturally metaphorical intentions of users, leading to deviations between the generated results and user expectations. Furthermore, the lack of a structured understanding of the Tibetan cultural symbol system results in image generation based on pixel-level statistical patterns rather than an understanding of the cultural symbol's meaning. This compromises the cultural accuracy and seriousness of the output, easily leading to misattribution of symbols, improper combinations, and even violations of cultural taboos, severely limiting its direct application in the professional cultural and creative field.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] An AI-based method for automatically generating Tibetan-style cultural and creative images, comprising:

[0008] It receives natural language descriptions input by users, uses a pre-trained BERT model to perform semantic parsing of the natural language descriptions, and extracts feature vectors of user intent and core keywords;

[0009] Acquire each Tibetan cultural symbol unit and its multidimensional attributes to construct a Tibetan cultural symbol gene bank;

[0010] Based on the feature vectors and core keywords of user intent, and using vector similarity, an initial set of cultural symbols that conforms to the user's design intent is selected.

[0011] Based on the combination rules of cultural symbol units, logical reasoning is performed on the initial set of cultural symbols to eliminate symbol combinations that have cultural conflicts, and to recommend symbol combination schemes that conform to cultural norms.

[0012] Based on the symbol combination scheme, the shape features of each symbol are extracted as invariable constraints, and the graphic parameters are automatically optimized to generate a set of symbol graphics with adaptive shape adjustment.

[0013] A training sample set of Tibetan-style cultural and creative patterns is constructed, and combined with an adaptively adjusted set of symbolic graphics, the final Tibetan-style cultural and creative patterns are generated based on the StableDiffusion diffusion model framework.

[0014] Based on the generated Tibetan cultural and creative patterns, and using the Tibetan cultural symbol gene library and Tibetan cultural and creative pattern training sample set, a structural report of the Tibetan cultural and creative patterns is output.

[0015] Preferably, the natural language description received from the user is semantically parsed using a pre-trained BERT model to extract the feature vector of the user's intent and core keywords, specifically including:

[0016] The user-input natural language description is preprocessed, including text cleaning, word segmentation, and the addition of [CLS] tags to form a sequence format that conforms to the model input.

[0017] We use a BERT model pre-trained on a general large-scale corpus as the basis, input the processed input sequence into the BERT model, and extract the output vector corresponding to the [CLS] marker in the final hidden layer of the model as the semantic feature vector of the entire user input sentence.

[0018] By using the context vector generated by the BERT model for each word in the sequence, and connecting it to a simple token-level classification layer, core cultural symbol keywords can be identified and extracted from user input.

[0019] Preferably, the step of selecting an initial set of cultural symbols that conforms to the user's design intent based on the feature vector and core keywords of the user's intent, and on the basis of vector similarity, specifically includes:

[0020] The cosine similarity between the sentence semantic feature vector and the pre-calculated semantic vector of each cultural symbol unit in the gene pool is used to obtain the first matching score.

[0021] The core keywords of the user's intent are matched with the attributes of each cultural symbol unit in the gene pool, and the second matching score is obtained based on the proportion of the number of matched keywords.

[0022] The first matching score and the second matching score are weighted and summed, and the cultural symbols are sorted according to the total weighted sum. Symbols with scores higher than a preset threshold are selected to form an initial set of cultural symbols.

[0023] Preferably, the step of extracting the shape features of each symbol as an immutable constraint based on the symbol combination scheme, automatically optimizing the graphic parameters, and generating a set of symbol graphics with adaptive shape adjustment specifically includes:

[0024] Each symbol in the symbol combination scheme is regarded as a deformable object, whose deformation is controlled by a set of graphic parameters. The graphic parameters of all symbols are combined into a parameter vector to be optimized.

[0025] Based on the intersection-union ratio of the symbol bounding boxes or the minimum distance between symbol skeletons, a collision loss term is constructed to penalize the overlap between symbols.

[0026] Based on the key point coordinates of symbols, we penalize misaligned relationships between symbols and encourage alignment methods that conform to the composition template, thus constructing an alignment loss term;

[0027] Based on the degree to which each symbol's graphic parameters deviate from their initial values, excessive deformation is penalized to maintain the symbol's cultural recognizability, and a deformation loss term is constructed.

[0028] Based on the collision loss term, alignment loss term, and deformation loss term, a graphical parameter deformation loss function is constructed using a weighted summation formula.

[0029] Based on the gradient descent algorithm, with the goal of minimizing the loss function, the graph parameter vector is iteratively updated. In each iteration, a temporary graph of the symbol is generated according to the current graph parameters, the loss value is calculated, and the parameter vector is adjusted according to the optimization algorithm.

[0030] When the loss function value converges to the preset threshold or reaches the maximum number of iterations, the optimization stops, and the set of symbolic graphics corresponding to the graphic parameter vector at this time is used as the adaptively adjusted set of symbolic graphics.

[0031] Preferably, the step of constructing a training sample set of Tibetan-style cultural and creative designs, combined with an adaptively adjusted set of symbolic graphics, and generating the final Tibetan-style cultural and creative designs based on the Stable Diffusion model framework, specifically includes:

[0032] Collect recognized, high-quality complete Tibetan cultural and creative images to form a Tibetan cultural and creative image training sample set. The training sample set includes complete images and their corresponding symbolic composition, compositional paradigm, and overall symbolic annotation.

[0033] A design concept encoder model is constructed, which consists of a pre-trained convolutional neural network as the feature extraction backbone, connected to a global pooling layer and a fully connected layer. The output of the fully connected layer is the design concept vector.

[0034] The complete graphics in the Tibetan cultural and creative image training sample set are used as input, and the design concept vector output by the model is used as the representation. The model is trained on a supervised task based on composition paradigm classification. After training, the parameters of the encoder model are fixed.

[0035] The T2I-Adapter spatial control network is used to convert the symbolic graphics in the adaptively adjusted symbolic graphic set into semantic segmentation maps;

[0036] Based on the Stable Diffusion model, semantic segmentation graphs are used as spatial layout control conditions, design concept vectors are used as style conditions, and training and learning are carried out on a Tibetan cultural and creative image training sample set.

[0037] Based on the natural language description input by the user, and combining the above content, the final Tibetan-style cultural and creative patterns are generated using a trained Stable Diffusion model.

[0038] Furthermore, this solution proposes an AI-based automatic generation system for Tibetan cultural and creative images, used to implement the aforementioned AI-based automatic generation method for Tibetan cultural and creative images, including:

[0039] The data processing module receives natural language descriptions input by the user, performs semantic parsing on the natural language descriptions using a pre-trained BERT model, and extracts feature vectors and core keywords of the user's intent. It then acquires each Tibetan cultural symbol unit and its multi-dimensional attributes, constructing a Tibetan cultural symbol gene library. Based on the feature vectors and core keywords of the user's intent, and using vector similarity, it selects an initial set of cultural symbols that conforms to the user's design intent. Based on the cultural symbol unit combination rule unit, it performs logical reasoning on the initial set of cultural symbols, excluding symbol combinations with cultural conflicts and recommending symbol combination schemes that conform to cultural norms. Based on the symbol combination schemes, it extracts the shape features of each symbol as an immutable constraint, automatically optimizes the graphic parameters, and generates a set of symbol graphics with adaptive shape adjustments.

[0040] The generation and reporting module is used to construct a training sample set of Tibetan cultural and creative patterns, combine it with an adaptively adjusted set of symbolic graphics, and generate the final Tibetan cultural and creative pattern based on the Stable Diffusion model framework. Based on the generated Tibetan cultural and creative pattern, and using the Tibetan cultural symbol gene library and the Tibetan cultural and creative pattern training sample set, the module outputs a structural report of the Tibetan cultural and creative pattern.

[0041] Preferably, the data processing module includes:

[0042] The user intent extraction unit is used to receive natural language descriptions input by users, and uses a pre-trained BERT model to perform semantic parsing on the natural language descriptions to extract feature vectors and core keywords of user intent.

[0043] A gene bank unit, which is used to acquire each Tibetan cultural symbol unit and its multidimensional attributes, and construct a Tibetan cultural symbol gene bank;

[0044] The filtering unit is used to filter out an initial set of cultural symbols that conform to the user's design intent based on the feature vector and core keywords of the user's intent and on vector similarity.

[0045] The scheme recommendation unit is used to perform logical reasoning on the initial set of cultural symbols based on the cultural symbol unit combination rule unit, exclude symbol combinations that have cultural conflicts, and recommend symbol combination schemes that conform to cultural norms.

[0046] A graphic adjustment unit is used to extract the shape features of each symbol as an invariable constraint based on the symbol combination scheme, automatically optimize the graphic parameters, and generate a set of symbol graphics with adaptive shape adjustment.

[0047] The generation and reporting module includes:

[0048] The pattern generation unit is used to construct a Tibetan cultural and creative pattern training sample set, combine it with an adaptively adjusted symbol graphic set, and generate the final Tibetan cultural and creative pattern based on the Stable Diffusion model framework.

[0049] The report generation unit is used to output a structural report of the Tibetan cultural and creative pattern based on the generated Tibetan cultural symbol gene library and the Tibetan cultural and creative pattern training sample set.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] This invention provides an AI-based method and system for automatically generating Tibetan cultural and creative images. The solution proposes a fully automated creation strategy encompassing semantic parsing, symbol matching, compliant combination, intelligent generation, and cultural tracing. It accurately extracts user intent features and core keywords using a pre-trained BERT model, combines this with a Tibetan cultural symbol gene library for precise symbol selection, and eliminates cultural conflicts and ensures standardized combination through combination rule units. Simultaneously, using symbol shape features as immutable constraints, it achieves adaptive form adjustment through multi-loss term optimization. Finally, based on the Stable Diffusion framework and dual control of space and style, it generates high-quality patterns and outputs a structural report containing symbol composition and meaning interpretation. This solution deeply integrates AI technology with Tibetan cultural rules, breaking through the reliance on professional skills in traditional design, avoiding cultural distortion and offense, and significantly improving the efficiency of cultural and creative image generation. Even non-professional users can quickly obtain Tibetan cultural and creative works that conform to cultural norms and meet design needs. Furthermore, the structural report clearly conveys cultural connotations, contributing to the innovative inheritance and dissemination of Tibetan culture. Attached Figure Description

[0052] Figure 1 is a flowchart of an automatic generation method for Tibetan-style cultural and creative images based on artificial intelligence according to the present invention;

[0053] Figure 2 is a flowchart of the automatic optimization of graphic parameters by extracting the shape features of each symbol as an invariable constraint according to the present invention.

[0054] Figure 3 is a flowchart of the process of generating the final Tibetan cultural and creative patterns based on the Stable Diffusion diffusion model framework of the present invention. Detailed Implementation

[0055] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0056] Referring to Figure 1, an artificial intelligence-based method for automatically generating Tibetan-style cultural and creative images includes:

[0057] It receives natural language descriptions input by users, uses a pre-trained BERT model to perform semantic parsing of the natural language descriptions, and extracts feature vectors of user intent and core keywords;

[0058] Acquire each Tibetan cultural symbol unit and its multidimensional attributes to construct a Tibetan cultural symbol gene bank;

[0059] Based on the feature vectors and core keywords of user intent, and using vector similarity, an initial set of cultural symbols that conforms to the user's design intent is selected.

[0060] Based on the combination rules of cultural symbol units, logical reasoning is performed on the initial set of cultural symbols to eliminate symbol combinations that have cultural conflicts, and to recommend symbol combination schemes that conform to cultural norms.

[0061] Based on the symbol combination scheme, the shape features of each symbol are extracted as invariable constraints, and the graphic parameters are automatically optimized to generate a set of symbol graphics with adaptive shape adjustment.

[0062] A training sample set of Tibetan-style cultural and creative patterns is constructed, and combined with an adaptively adjusted set of symbolic graphics, the final Tibetan-style cultural and creative patterns are generated based on the StableDiffusion diffusion model framework.

[0063] Based on the generated Tibetan cultural and creative patterns, and using the Tibetan cultural symbol gene library and Tibetan cultural and creative pattern training sample set, a structural report of the Tibetan cultural and creative patterns is output.

[0064] Understandably, the core of Tibetan cultural and creative image design lies in accurately matching user needs with Tibetan cultural norms. It aims to both restore the connotation and form of cultural symbols and satisfy personalized creative demands. Therefore, this solution employs a fully automated creation strategy encompassing "semantic parsing—symbol matching—compliant combination—intelligent generation—cultural origin tracing" as its overall process logic. Specifically, the user-input natural language description is parsed into sentence semantic feature vectors and a set of core keywords using a pre-trained BERT model. The Tibetan cultural symbol gene bank stores symbol units with multi-dimensional attributes and combination rules. Based on vector cosine similarity calculations, an initial set of cultural symbols with a matching degree exceeding a preset threshold is selected. Logical reasoning is applied to this initial set of cultural symbols through combination rule units, eliminating symbol pairs with taboo conflicts to obtain a compliant combination scheme. Using symbol shape features as constraints, a graphic parameter deformation loss function constructed through collision, alignment, and deformation loss terms optimizes the graphic parameters, generating an adaptively adjusted symbol set. The Tibetan cultural and creative image training sample set is then used to refine the Stable... The Diffusion model is fine-tuned, and the final pattern is generated by fusing semantic segmentation graphs with design concept vectors. Finally, a structural report containing symbol composition and meaning is output based on the gene bank and sample set, ensuring that the generated result not only meets user needs but also conforms to Tibetan cultural norms.

[0065] The process of receiving natural language descriptions input by the user, and using a pre-trained BERT model to perform semantic parsing on the natural language descriptions to extract feature vectors of user intent and core keywords specifically includes:

[0066] The user-input natural language description is preprocessed, including text cleaning, word segmentation, and the addition of [CLS] tags to form a sequence format that conforms to the model input.

[0067] We use a BERT model pre-trained on a general large-scale corpus as the basis, input the processed input sequence into the BERT model, and extract the output vector corresponding to the [CLS] marker in the final hidden layer of the model as the semantic feature vector of the entire user input sentence.

[0068] By using the context vector generated by the BERT model for each word in the sequence, and connecting it to a simple token-level classification layer, core cultural symbol keywords can be identified and extracted from user input.

[0069] One explanation is that language and input habits lead to a mismatch between the input text and the interpretations of cultural symbols in the Tibetan cultural symbol gene database. Directly matching the input content with the corresponding textual interpretations might fail to accurately pinpoint core keywords, resulting in inaccurate location. Secondly, comprehensive matching from the user's natural language description could increase the overall computational load. Therefore, it's necessary to extract the feature vectors and core keywords from the user's natural language description. This includes text cleaning: removing special characters, spaces, and meaningless function words; using Unicode encoding if the input is "zangyu"; and word segmentation: using the jieba word segmentation tool if the input is Chinese, and using "zan" if the input is "zangyu". Gyu uses the Zangyu-specific word segmenter THULAC-zangyu to split the text into words. Sequence formatting: [CLS] and [SEP] markers are added to the beginning and end of the text respectively, and the text is converted into a fixed-length input sequence according to the requirements of the BERT model. The BERT model pre-trained on a general large-scale corpus as the basis refers to using a variant of the ZangyuBERT model pre-trained on general corpora and Zangyu corpora (such as Tibetan Wikipedia and Tibetan cultural literature corpus) as the base model. The processed input sequence is input into this model, and the output vector corresponding to the [CLS] marker in the final hidden layer of the model is extracted as the semantic feature vector of the entire user input sentence.

[0070] The connection to a simple token-level classification layer, which identifies and extracts core cultural symbol keywords from user input, specifically includes:

[0071] Obtain descriptive texts of several user design intentions, and have experts manually annotate the cultural symbols and keywords within them;

[0072] The BIO (Begin, Inside, Outside) annotation system is adopted to annotate characters or word units, and the annotated cultural symbol keywords are converted into sequence annotations that can be processed by computers.

[0073] On top of the pre-trained BERT model, a simple token-level classification layer is connected, which is typically a linear fully connected layer with the same number of neurons as the number of BIO label categories;

[0074] With the goal of minimizing the difference between the predicted probability distribution of each word and the true labeled label, the cross-entropy formula is used to construct the loss function of the BERT model;

[0075] Based on the acquired sequence labeling data, the BERT model is fine-tuned end-to-end using a supervised approach.

[0076] The supervised end-to-end fine-tuning of the BERT model specifically includes:

[0077] The sequence labeled data were divided into training set, validation set and test set in a ratio of 8:1:1. During the training process, the F1 score of the validation set was used as the basis for selecting the optimal parameters of the model.

[0078] Forward propagation: The training set is input into the BERT model in batches. After the data passes through the BERT model, the context semantic vector of each word is obtained. The semantic vector is then fed into a simple token-level classification layer to output unnormalized scores. The Softmax function is used to convert these scores into probability distributions for each BIO tag.

[0079] Backpropagation and parameter update: Based on the predicted probability and the real BIO label obtained from forward propagation, the loss value of the current batch is calculated using the BERT model loss function. The gradient of all trainable parameters (including the parameters of the BERT model itself and the weights of the classification layer) is calculated using the backpropagation algorithm. All parameters are updated using a gradient descent optimization algorithm (such as AdamW).

[0080] The training process is repeated multiple times (e.g., 3 to 5 times) on the training set based on forward propagation and back propagation and parameter update. When the F1 value on the validation set no longer improves for several consecutive times (e.g., 2 times), the training stops and the optimal model parameters are saved. This allows the model to accurately map natural language descriptions to BIO tag sequences, thereby accurately extracting cultural symbol keywords.

[0081] The process of acquiring each Tibetan cultural symbol unit and its multidimensional attributes to construct a Tibetan cultural symbol gene bank specifically includes:

[0082] Tibetan studies experts collaborated to deconstruct traditional Tibetan patterns and extract several basic units of Tibetan culture as cultural symbol units.

[0083] Based on the experience and rules of Tibetan studies experts, pre-set rules for the combination of cultural symbol units and taboos for their use, and construct a unit for the combination of cultural symbol units;

[0084] The combination rule unit includes at least:

[0085] Taboo rule table: Used to store prohibited combinations. Each record contains at least: symbol A that triggers the taboo, symbol B that triggers the taboo, and taboo level of the combination. The taboo level field is used to quantify the severity of the taboo.

[0086] Recommended Combination Table: Used to store encouraging combination relationships. Each record contains at least: primary symbol ID, recommended combination symbol ID, and combination type label; wherein, the combination type label includes at least: classic, recommended, or high frequency, used to identify and set priority scores;

[0087] All cultural symbol units and cultural symbol unit combination rules are stored according to a preset data structure to construct a structured Tibetan cultural symbol gene library;

[0088] The attributes of each cultural symbol unit in the gene pool are combined into a complete text description, and the same BERT model used to parse user intent is used to generate a standardized semantic vector for each cultural symbol unit, which is then stored in the gene pool.

[0089] It can be understood that the Tibetan cultural symbol unit refers to the smallest basic unit of Tibetan cultural symbols. All complete Tibetan cultural and creative images are composed of multiple smallest basic units, such as: Tibetan meander pattern unit, lotus petal basic unit, single treasure unit from the Eight Auspicious Symbols, etc. Its multidimensional attributes include morphological characteristics, cultural attributes, combination relationships, and cultural boundary conditions. The morphological characteristics include vector data of the basic graphic composition, standard color palette, and color taboos. The cultural attributes include the core symbolic keyword set, the source of the symbol prototype, and hierarchical attributes. The combination relationships include the possibility of independent use, combination interface characteristics, and common combination sequences. The cultural boundary conditions include combination rules and usage taboos. The combination rules define the recommended combination methods of this symbol with other symbols that conform to cultural conventions. The usage taboos define the combinations, scenarios, or applications in which this symbol is prohibited.

[0090] The preset data structure includes symbol ID, symbol name, symbol vector graphic data, core meaning keyword set, cultural boundary conditions, and style feature vector; the morphological features include basic graphic composition, color reference, and pattern texture; the basic graphic composition includes line type, geometric shape, and simplified form of natural imagery, with line type including straight line, curve, and broken line, and geometric shape including circle, triangle, and rhombus; the color reference is the standard color and color combination paradigm of the symbol unit in the traditional context; the pattern texture is the recurring texture element; the cultural attributes include symbol prototype source, core meaning, and hierarchical attributes, with symbol prototype source including Buddhist scriptures, folk legends, and nature worship, and hierarchical attributes including those specific to Buddhist implements, those specific to commoners' clothing, and general-purpose.

[0091] The process of selecting an initial set of cultural symbols that aligns with the user's design intent based on the feature vector and core keywords of the user's intent, and on the basis of vector similarity, specifically includes:

[0092] The cosine similarity between the sentence semantic feature vector and the pre-calculated semantic vector of each cultural symbol unit in the gene pool is used to obtain the first matching score.

[0093] The core keywords of the user's intent are matched with the attributes of each cultural symbol unit in the gene pool, and the second matching score is obtained based on the proportion of the number of matched keywords.

[0094] The first matching score and the second matching score are weighted and summed, and the cultural symbols are sorted according to the total weighted sum. Symbols with scores higher than a preset threshold are selected to form an initial set of cultural symbols.

[0095] This solution aims to accurately map users' ambiguous natural language intentions to a structured cultural symbol gene library, achieving precise matching through a dual-path collaborative strategy: First, deep semantic matching, which captures the deep connection between abstract intentions and symbolic cultural connotations by calculating the cosine similarity between the user's sentence meaning feature vector and the symbol semantic vector in the gene library (for example, the user input "divine protection" can highly match the semantic vector of the flame pattern that carries the meaning of "divine protection"); Second, precise keyword matching, which establishes an interpretable direct association by comparing the parsed core keywords with standardized keywords in the symbol attribute library. After weighted fusion and sorting of the matching scores output by the two paths, an initial set of cultural symbols that both fits the user's intentions and has cultural correctness can be efficiently selected, laying a solid foundation for subsequent creation.

[0096] The rule unit based on the combination of cultural symbols performs logical reasoning on the initial set of cultural symbols, eliminates symbol combinations that have cultural conflicts, and recommends symbol combination schemes that conform to cultural norms. Specifically, this includes:

[0097] Traverse the initial set of cultural symbols, query the taboo rule table, and determine whether there are any symbol pairs or groups that are explicitly prohibited from coexisting in the set. If so, remove one or more symbols that trigger the taboo from the set according to the taboo level, or directly discard the current entire combination scheme.

[0098] After conflict resolution, the recommended combination table is queried to determine whether the symbol IDs in the current symbol set match the records in the recommended combination table. If a matching recommended combination exists, a priority score is assigned to the matching combination based on its combination type label, and the multiple possible symbol combination schemes are sorted and output based on this score.

[0099] This can be understood as follows: in the process of generating Tibetan cultural and creative patterns using artificial intelligence, a step based on cultural rules for logical reasoning is specifically established. Its core purpose is to address the fundamental deficiency of artificial intelligence in understanding deep cultural norms, ensuring that the output possesses both artistic beauty and cultural correctness. This solution introduces a taboo rule table and a recommended combination table from the cultural symbol unit combination rule unit, transforming the experience of Tibetan studies experts into computer-executable judgment logic. Specifically, the system first traverses the initial set of cultural symbols, queries the taboo rule table, and automatically performs symbol removal or scheme abandonment operations based on the taboo level, thereby proactively avoiding cultural misuse and offense risks, establishing a cultural safety baseline for the generation process. After conflict resolution, it further queries the recommended combination table, assigning higher priority to classic symbol combinations that conform to cultural conventions. Essentially, this injects traditional cultural wisdom into the generation logic, guiding the system to produce design schemes that are not only aesthetically pleasing but also possess cultural connotations and authority, effectively improving the efficiency and quality of subsequent generation. Simultaneously, its traceable reasoning process provides a clear cultural basis for the final output, elevating the system from a mere form generator to a trustworthy cultural design consultant.

[0100] Referring to Figure 2, the step of extracting the shape features of each symbol as an invariant constraint and automatically optimizing the graphic parameters specifically includes:

[0101] Each symbol in the symbol combination scheme is regarded as a deformable object, whose deformation is controlled by a set of graphic parameters. The graphic parameters of all symbols are combined into a parameter vector to be optimized.

[0102] Based on the intersection-union ratio of the symbol bounding boxes or the minimum distance between symbol skeletons, a collision loss term is constructed to penalize the overlap between symbols.

[0103] Based on the key point coordinates of symbols, we penalize misaligned relationships between symbols and encourage alignment methods that conform to the composition template, thus constructing an alignment loss term;

[0104] Based on the degree to which each symbol's graphic parameters deviate from their initial values, excessive deformation is penalized to maintain the symbol's cultural recognizability, and a deformation loss term is constructed.

[0105] Based on the collision loss term, alignment loss term, and deformation loss term, a graphical parameter deformation loss function is constructed using a weighted summation formula.

[0106] Based on the gradient descent algorithm, with the goal of minimizing the loss function, the graph parameter vector is iteratively updated. In each iteration, a temporary graph of the symbol is generated according to the current graph parameters, the loss value is calculated, and the parameter vector is adjusted according to the optimization algorithm.

[0107] When the loss function value converges to the preset threshold or reaches the maximum number of iterations, the optimization stops, and the set of symbolic graphics corresponding to the graphic parameter vector at this time is used as the adaptively adjusted set of symbolic graphics.

[0108] It is understandable that basic cultural symbols may require slight adjustments in actual design and cannot be completely copied from fixed cultural symbol graphic forms. For example, based on the aesthetics of the overall pattern, designers may need to stretch, rotate, or compress a symbol graphic to adjust its appearance without changing its graphic form, so as to better adapt to the overall design and avoid a lack of smoothness in the overall design. This solution predefines the parameter types and boundaries that allow deformation for each symbol type during system initialization. When a set of specific symbols is selected to form a scheme, the system will randomly initialize a set of graphic parameter values ​​that conform to its boundaries for each symbol instance in the scheme. The parameter values ​​of all these instances together constitute the parameter vector to be optimized. The goal of the optimization algorithm is to find the specific combination of parameter values ​​that makes the overall composition most aesthetically pleasing.

[0109] The specific implementation of constructing the graphic parameter deformation loss function can be as follows:

[0110] Collision Loss Term: This term ensures a balanced layout between symbols. It calculates the cross-union ratio (CURPR) of the axial bounding boxes of all pairs of symbols and uses the sum of all CURPRs as the collision loss. Ideally, this value should be 0. Its quantification method, for example, is the CURPR method. The expression for the collision loss term is:

[0111]

[0112] In the formula, The sum of the intersection-union ratios among all bounding box pairs. and They are symbols and symbols Axial enclosure box, and Let these represent the intersection and union of the bounding boxes, respectively. This is a function for calculating area. Total number of bounding boxes It is a very small positive number, used to prevent the denominator from being zero;

[0113] Alignment loss term: This term guides the formation of an ordered composition of symbols. It is calculated based on mathematical distance metrics (such as Euclidean distance) and coordinate transformations. For example, under a centrally symmetric composition template, a loss term can be defined to calculate the distance from the key points (such as the center point) of surrounding symbols to the radial lines of the center symbol. The smaller the distance, the smaller the loss. The expression for the alignment loss term is:

[0114]

[0115] In the formula, A set of points Geometric alignment error with the centrally symmetric template For the first The coordinates of the points to be aligned This is a rotation matrix, representing the angle of rotation of the template around the center. , For the template and The corresponding reference point coordinates, Let be the number of points to be aligned. Let be the translation vector, representing the displacement from the template center to the input data center. It is the square of the L2 norm (Euclidean distance);

[0116] The above is an example of alignment loss calculation under the centrally symmetric composition template. It can be understood that the alignment loss term is not limited to this. Those skilled in the art can define corresponding reference templates and transformation relationships (such as rotation matrix and translation vector) according to different composition requirements (such as axisymmetric, grid composition, golden ratio composition, etc.). The core is to guide the layout alignment by calculating the geometric error between the symbol key points and the target template.

[0117] Deformation loss term: This term is the mathematical embodiment of cultural constraints. It sets a tolerance range for the deformation parameters (such as scaling ratio and stretching amplitude) of each symbol. The expression for the deformation loss term is:

[0118]

[0119] In the formula, L2 regularization for graphic deformation parameters, This represents the total number of graphic elements that need to be constrained (e.g., symbols, key points, control points). For the first The current parameter vector of each graphic element. For the first An initial parameter vector for each graphic element.

[0120] Referring to Figure 3, the process of generating the final Tibetan-style cultural and creative patterns based on the Stable Diffusion model framework specifically includes:

[0121] Collect recognized, high-quality complete Tibetan cultural and creative images to form a Tibetan cultural and creative image training sample set. The training sample set includes complete images and their corresponding symbolic composition, compositional paradigm, and overall symbolic annotation.

[0122] A design concept encoder model is constructed, which consists of a pre-trained convolutional neural network as the feature extraction backbone, connected to a global pooling layer and a fully connected layer. The output of the fully connected layer is the design concept vector.

[0123] The complete graphics in the Tibetan cultural and creative image training sample set are used as input, and the design concept vector output by the model is used as the representation. The model is trained on a supervised task based on composition paradigm classification. After training, the parameters of the encoder model are fixed.

[0124] The T2I-Adapter spatial control network is used to convert the symbolic graphics in the adaptively adjusted symbolic graphic set into semantic segmentation maps;

[0125] Based on the Stable Diffusion model, semantic segmentation graphs are used as spatial layout control conditions, design concept vectors are used as style conditions, and training and learning are carried out on a Tibetan cultural and creative image training sample set.

[0126] Based on the natural language description input by the user, and combining the above content, the final Tibetan-style cultural and creative patterns are generated using a trained Stable Diffusion model.

[0127] It can be understood that the Tibetan cultural symbol gene bank and the Tibetan cultural and creative image training sample set together constitute the knowledge foundation of the system. The gene bank provides the grammar and vocabulary of cultural design, ensuring the accuracy of generation, while the training sample set provides written examples and styles, ensuring the artistry of the generated results. The two are linked through symbolic information, jointly guiding AI to create works that are both culturally orthodox and aesthetically pleasing. For any excellent design sample input, a fixed-dimensional, digital design concept vector can be extracted through the trained design concept encoder model. The Stable Diffusion model includes a denoising UNet network, which inputs spatial layout control conditions into the spatial control network and concatenates style conditions with a general text prompt word encoding to form a conditional embedding, which is then input into the denoising UNet network of the Stable Diffusion model. Randomly sampled Gaussian noise and the conditional embedding are input together into the Stable Diffusion model. The Diffusion model, guided by a spatial control network, performs an iterative denoising process. After a preset number of denoising steps, it decodes the output in the latent space into high-resolution Tibetan cultural and creative patterns in the pixel space. An example of implementing the StableDiffusion model is as follows: using the LoRA lightweight fine-tuning mode, freezing the Stable Diffusion backbone network, training only the low-rank matrix in UNet, using the AdamW optimizer, learning rate 5e-5, batch size 16, training epochs 80, and connecting the T2I-Adapter spatial control network to the 3rd and 5th layers of the ResNet blocks in UNet. The compositional paradigms include: religious sacred type, life realistic type, and natural fusion type.

[0128] The process of outputting a structural report of the generated Tibetan cultural and creative patterns, based on the Tibetan cultural symbol gene library and the Tibetan cultural and creative pattern training sample set, specifically includes:

[0129] Based on the generated Tibetan cultural and creative patterns, we traverse all the cultural symbol units that make up the Tibetan cultural and creative patterns, as well as their symbol composition, compositional paradigm, and overall symbolic meaning.

[0130] The structural report of this Tibetan-style cultural and creative pattern specifically includes:

[0131] Symbolic composition: List all identified cultural symbols and their IDs;

[0132] Interpretive meaning: The key cultural symbols and semantic feature vectors for each cultural symbol;

[0133] Combinatorial logic: Referencing compositional rules to explain the cultural basis for combining cultural symbols;

[0134] Application Recommendations: Based on taboo rules, we provide recommendations for applicable scenarios.

[0135] It can be understood that generating a structural report after generating Tibetan cultural and creative patterns is a key step in realizing the transformation of AI creation from technical implementation to design credibility. It is a cultural compliance certification and design traceability explanation of the generated results based on the Tibetan cultural symbol gene library and training sample set. Its value lies in listing the IDs of all cultural symbols in the pattern one by one, analyzing the traditional symbolic keywords and semantic vector features of each symbol, and clearly citing specific clauses in the cultural symbol unit combination rule library to explain the cultural basis of symbol combination. This provides verifiable cultural origins for every detail of the generation. This not only gives the AI ​​creation results cultural credibility, but also forms a complete technical closed loop from symbol deconstruction to pattern generation to cultural verification. The final output of the structural report is not only the pattern itself, but a complete cultural design solution that includes symbol composition analysis, cultural meaning interpretation, combination logic explanation, and application scenario suggestions based on taboo rules. This transforms the generated results from simple graphic files into standardized digital assets with clear cultural genes that can be directly applied to cultural and creative product development, digital collection issuance, and cultural dissemination scenarios.

[0136] Furthermore, based on the same inventive concept as the aforementioned AI-based automatic generation method for Tibetan cultural and creative designs, this solution proposes an AI-based automatic generation system for Tibetan cultural and creative designs, comprising:

[0137] The data processing module receives natural language descriptions input by the user, performs semantic parsing on the natural language descriptions using a pre-trained BERT model, and extracts feature vectors and core keywords of the user's intent. It then acquires each Tibetan cultural symbol unit and its multi-dimensional attributes, constructing a Tibetan cultural symbol gene library. Based on the feature vectors and core keywords of the user's intent, and using vector similarity, it selects an initial set of cultural symbols that conforms to the user's design intent. Based on the cultural symbol unit combination rule unit, it performs logical reasoning on the initial set of cultural symbols, excluding symbol combinations with cultural conflicts and recommending symbol combination schemes that conform to cultural norms. Based on the symbol combination schemes, it extracts the shape features of each symbol as an immutable constraint, automatically optimizes the graphic parameters, and generates a set of symbol graphics with adaptive shape adjustments.

[0138] The generation and reporting module is used to construct a Tibetan cultural and creative pattern training sample set, combine it with an adaptively adjusted symbol graphic set, and generate the final Tibetan cultural and creative pattern based on the Stable Diffusion model framework; based on the generated Tibetan cultural and creative pattern, and based on the Tibetan cultural symbol gene library and the Tibetan cultural and creative pattern training sample set, output a structural report of the Tibetan cultural and creative pattern.

[0139] The data processing module includes:

[0140] The user intent extraction unit is used to receive natural language descriptions input by users, and uses a pre-trained BERT model to perform semantic parsing on the natural language descriptions to extract feature vectors and core keywords of user intent.

[0141] A gene bank unit, which is used to acquire each Tibetan cultural symbol unit and its multidimensional attributes, and construct a Tibetan cultural symbol gene bank;

[0142] The filtering unit is used to filter out an initial set of cultural symbols that conform to the user's design intent based on the feature vector and core keywords of the user's intent and on vector similarity.

[0143] The scheme recommendation unit is used to perform logical reasoning on the initial set of cultural symbols based on the cultural symbol unit combination rule unit, exclude symbol combinations that have cultural conflicts, and recommend symbol combination schemes that conform to cultural norms.

[0144] A graphic adjustment unit is used to extract the shape features of each symbol as an invariable constraint based on the symbol combination scheme, automatically optimize the graphic parameters, and generate a set of symbol graphics with adaptive shape adjustment.

[0145] The generation and reporting module includes:

[0146] The pattern generation unit is used to construct a Tibetan cultural and creative pattern training sample set, combine it with an adaptively adjusted symbol graphic set, and generate the final Tibetan cultural and creative pattern based on the Stable Diffusion model framework.

[0147] The report generation unit is used to output a structural report of the Tibetan cultural and creative pattern based on the generated Tibetan cultural symbol gene library and the Tibetan cultural and creative pattern training sample set.

[0148] In summary, the advantages of this invention are: by constructing a gene library of Tibetan cultural symbols that integrates expert knowledge and intelligent reasoning rules, natural language descriptions are automatically transformed into Tibetan cultural and creative patterns that conform to cultural norms and aesthetic requirements, thus achieving a deep integration of cultural inheritance and intelligent generation.

[0149] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for automatically generating Tibetan-style cultural and creative images based on artificial intelligence, characterized in that, include: The system receives natural language descriptions input by users and uses a pre-trained BERT model to perform semantic parsing, extracting feature vectors and core keywords representing user intent. It then acquires each Tibetan cultural symbol unit and its multidimensional attributes, constructing a Tibetan cultural symbol gene library. Based on the feature vectors and core keywords representing user intent, and using vector similarity, it selects an initial set of cultural symbols that aligns with the user's design intent. Based on the cultural symbol unit combination rule unit, it performs logical reasoning on the initial set of cultural symbols, excluding symbol combinations with cultural conflicts and recommending symbol combination schemes that conform to cultural norms. Based on the symbol combination schemes, it extracts the shape features of each symbol as an immutable constraint, automatically optimizing the graphic parameters to generate a set of symbol graphics with adaptive shape adjustment. It constructs a Tibetan cultural and creative image training sample set, and combines it with the adaptively adjusted symbol graphic set, generating the final Tibetan cultural and creative pattern based on the StableDiffusion diffusion model framework. Finally, based on the generated Tibetan cultural and creative pattern, and using the Tibetan cultural symbol gene library and the Tibetan cultural and creative image training sample set, it outputs a structure report of the Tibetan cultural and creative pattern.

2. The method for automatically generating Tibetan-style cultural and creative images based on artificial intelligence according to claim 1, characterized in that, The process of receiving natural language descriptions input by users and using a pre-trained BERT model to perform semantic parsing of the natural language descriptions to extract feature vectors of user intent and core keywords specifically includes: preprocessing the natural language descriptions input by users, including text cleaning, word segmentation, and adding [CLS] tags to form a sequence format that conforms to the model input; using a BERT model pre-trained on a general large-scale corpus as a basis, inputting the processed input sequence into the BERT model, extracting the output vector corresponding to the [CLS] tags in the final hidden layer of the model as the semantic feature vector of the entire user input sentence; and using the context vector generated by the BERT model for each word in the sequence, connecting it to a simple token-level classification layer to identify and extract core cultural symbol keywords from the user input.

3. The method for automatically generating Tibetan-style cultural and creative images based on artificial intelligence according to claim 2, characterized in that, The acquisition of each Tibetan cultural symbol unit and its multidimensional attributes, and the construction of a Tibetan cultural symbol gene bank, specifically includes: collaborating with Tibetan studies experts to deconstruct traditional Tibetan patterns and extract multiple basic units of Tibetan culture as cultural symbol units; based on the experience and rules of Tibetan studies experts, pre-setting rules for combining cultural symbol units and taboos, and constructing a combination rule unit for cultural symbol units; wherein, the combination rule unit includes at least: a taboo rule table: used to store prohibited combination relationships, each record containing at least: symbol A that triggers the taboo, symbol B that triggers the taboo, and the taboo level of the combination, the taboo level field being used to quantify the severity of the taboo. Degree; Recommended Combination Table: Used to store encouraging combination relationships, each record contains at least: main symbol ID, recommended pairing symbol ID, and combination type label; wherein, the combination type label includes at least: classic, recommended, or high frequency, used to identify and set priority scores; All cultural symbol units and cultural symbol unit combination rule units are stored according to a preset data structure to construct a structured Tibetan cultural symbol gene library; The attributes of each cultural symbol unit in the gene library are combined into a complete text description, and the same BERT model used to parse user intent is used to generate a standardized semantic vector for each cultural symbol unit, which is then stored in the gene library.

4. The method for automatically generating Tibetan-style cultural and creative images based on artificial intelligence according to claim 3, characterized in that, The step of selecting an initial set of cultural symbols that conforms to the user's design intent based on the feature vector and core keywords of the user's intent, and on the basis of vector similarity, specifically includes: calculating the cosine similarity between the sentence meaning feature vector and the pre-calculated semantic vector of each cultural symbol unit in the gene pool to obtain a first matching score; matching the core keywords of the user's intent with the attributes of each cultural symbol unit in the gene pool, and obtaining a second matching score based on the proportion of matched keywords; performing a weighted summation of the first matching score and the second matching score, and sorting the cultural symbols according to the total weighted summation score, and selecting symbols with scores higher than a preset threshold to form an initial set of cultural symbols.

5. The method for automatically generating Tibetan-style cultural and creative images based on artificial intelligence according to claim 4, characterized in that, The rule unit based on cultural symbol units performs logical reasoning on the initial set of cultural symbols, eliminates symbol combinations with cultural conflicts, and recommends symbol combination schemes that conform to cultural norms. Specifically, this includes: traversing the initial set of cultural symbols, querying the taboo rule table, determining whether there are symbol pairs or groups in the set that are explicitly prohibited from coexisting, and if so, removing one or more symbols that trigger taboos from the set according to the taboo level, or directly discarding the entire current combination scheme; after conflict elimination, querying the recommended combination table, determining whether the symbol IDs in the current symbol set match the records in the recommended combination table, and if a matching recommended combination exists, assigning a priority score to the matching combination according to its combination type label, and sorting and outputting multiple possible symbol combination schemes based on this score.

6. The method for automatically generating Tibetan-style cultural and creative images based on artificial intelligence according to claim 5, characterized in that, The symbol combination scheme, which extracts the shape features of each symbol as an immutable constraint and automatically optimizes the graphic parameters to generate a set of symbol graphics with adaptive shape adjustment, specifically includes: treating each symbol in the symbol combination scheme as a deformable object whose deformation is controlled by a set of graphic parameters; combining the graphic parameters of all symbols into a parameter vector to be optimized; penalizing the overlap between symbols based on the intersection-union ratio of the symbol bounding boxes or the minimum distance between symbol skeletons, and constructing a collision loss term; penalizing the misalignment between symbols based on the key point coordinates of the symbols, and encouraging alignment methods that conform to the composition template, and constructing an alignment loss term; and based on the deviation of each symbol's graphic parameters... The initial value is used to penalize excessive deformation to maintain the cultural recognizability of the symbol, and a deformation loss term is constructed. Based on the collision loss term, alignment loss term, and deformation loss term, a graphic parameter deformation loss function is constructed using a weighted summation formula. Based on the gradient descent algorithm, the graphic parameter vector is iteratively updated with the goal of minimizing the loss function. In each iteration, a temporary graphic of the symbol is generated based on the current graphic parameters, the loss value is calculated, and the parameter vector is adjusted according to the optimization algorithm. When the loss function value converges to a preset threshold or reaches the maximum number of iterations, the optimization stops, and the set of symbol graphics corresponding to the graphic parameter vector at this time is used as the adaptively adjusted set of symbol graphics.

7. The method for automatically generating Tibetan-style cultural and creative images based on artificial intelligence according to claim 6, characterized in that, The construction of the Tibetan cultural and creative image training sample set, combined with the adaptively adjusted symbolic graphic set, and the generation of the final Tibetan cultural and creative pattern based on the StableDiffusion diffusion model framework, specifically includes: collecting recognized, high-quality complete Tibetan cultural and creative images to form the Tibetan cultural and creative image training sample set, wherein the training sample set includes complete images and their corresponding symbolic composition, compositional paradigm, and overall meaning annotation; building a design concept encoder model, which consists of a pre-trained convolutional neural network as the feature extraction backbone, connected to a global pooling layer and a fully connected layer, the output of which is a design concept vector; using the complete images in the Tibetan cultural and creative image training sample set as input, and the design concept vector output by the model as representation, training is performed on a supervised task based on compositional paradigm classification, and after training, the parameters of the encoder model are fixed; the symbolic graphics in the adaptively adjusted symbolic graphic set are converted into semantic segmentation maps through the T2I-Adapter spatial control network; and the StableDiffusion diffusion model is used to further refine the design concept image. Based on the Diffusion model, the system uses semantic segmentation graphs as spatial layout control conditions and design concept vectors as style conditions. It is trained and learned based on a training sample set of Tibetan cultural and creative images. According to the natural language description input by the user, the system integrates the above content and generates the final Tibetan cultural and creative pattern based on the trained Stable Diffusion model.

8. The method for automatically generating Tibetan-style cultural and creative images based on artificial intelligence according to claim 7, characterized in that, The structure report of the generated Tibetan cultural and creative patterns, based on the Tibetan cultural symbol gene library and the Tibetan cultural and creative pattern training sample set, specifically includes: traversing all cultural symbol units that make up the generated Tibetan cultural and creative patterns, their symbol composition, compositional paradigm, and overall meaning annotation; the structure report of the generated Tibetan cultural and creative patterns specifically includes: symbol composition: listing all identified cultural symbols and their IDs; meaning interpretation: cultural symbol keywords and semantic feature vectors for each cultural symbol; combination logic: citing combination rules to explain the cultural basis for the combination of cultural symbols; application suggestions: providing recommendations for applicable scenarios based on taboo rules.

9. An AI-based automatic generation system for Tibetan-style cultural and creative images, characterized in that, The method for automatically generating Tibetan cultural and creative images based on artificial intelligence as described in any one of claims 1-8 includes: a data processing module, which receives natural language descriptions input by users, performs semantic parsing on the natural language descriptions using a pre-trained BERT model, extracts feature vectors and core keywords of user intent; obtains each Tibetan cultural symbol unit and its multi-dimensional attributes, and constructs a Tibetan cultural symbol gene library; based on the feature vectors and core keywords of user intent, and based on vector similarity, selects an initial set of cultural symbols that conforms to the user's design intent; based on the cultural symbol unit combination rule unit, performs logical reasoning on the initial set of cultural symbols, excludes symbol combinations with cultural conflicts, and recommends symbol combination schemes that conform to cultural norms; based on the symbol combination scheme, extracts the shape features of each symbol as an immutable constraint, automatically optimizes the graphic parameters, and generates a set of symbol graphics with adaptive shape adjustment; and a generation and reporting module, which constructs a training sample set of Tibetan cultural and creative images, combines the adaptively adjusted symbol graphic set, and based on Stable... The Diffusion model framework is used to generate the final Tibetan cultural and creative patterns. Based on the generated Tibetan cultural and creative patterns, and using the Tibetan cultural symbol gene library and the Tibetan cultural and creative pattern training sample set, a structural report of the Tibetan cultural and creative patterns is output.

10. The automatic generation system for Tibetan-style cultural and creative images based on artificial intelligence according to claim 9, characterized in that, The data processing module includes: a user intent extraction unit, which receives natural language descriptions input by users, performs semantic parsing on the natural language descriptions using a pre-trained BERT model, and extracts feature vectors and core keywords of user intent; a gene bank unit, which acquires each Tibetan cultural symbol unit and its multidimensional attributes to construct a Tibetan cultural symbol gene bank; a screening unit, which selects an initial set of cultural symbols that conforms to the user's design intent based on the feature vectors and core keywords of the user intent and vector similarity; a scheme recommendation unit, which performs logical reasoning on the initial set of cultural symbols based on the cultural symbol unit combination rule unit, excludes symbol combinations with cultural conflicts, and recommends symbol combination schemes that conform to cultural norms; and a graphic adjustment unit, which extracts the shape features of each symbol as an immutable constraint based on the symbol combination scheme, automatically optimizes the graphic parameters, and generates a set of symbol graphics with adaptive shape adjustment. The generation and reporting module includes: a pattern generation unit, which constructs a Tibetan cultural and creative image training sample set, combines it with the adaptively adjusted symbol graphic set, and uses a Stable... The Diffusion model framework generates the final Tibetan cultural and creative pattern; the report generation unit is used to output a structural report of the generated Tibetan cultural and creative pattern based on the Tibetan cultural symbol gene library and the Tibetan cultural and creative pattern training sample set.

Images