Multi-scene art AI generation and customization method and system based on LoRA

Abstract

The invention discloses a multi-scene art AI generation and customization method and system based on LoRA, and relates to the technical field of AI art generation and LoRA model training. The core aims at four core scenes of children beauty, an art museum / museum, artists and enterprises, each scene and each subject exclusive LoRA sub-model are directionally trained, core technologies of scene function adaptation, style calibration, multi-modal input and output and copyright traceability are matched, the core focuses on artistic style and visual system customization, character image modification is not involved, and the method is simple and convenient. Double-feature fusion training logic is adopted, material compliance is clear, and a whole-process closed loop is formed; the technical scheme is highly in accordance with the requirements of various scenes, can accurately meet the core demands of children teaching assistance, visual unification and cultural tonality adaptation of art museums / museums, artist style replication, enterprise visual customization and the like, is high in style accuracy, convenient to operate and safe in copyright, does not conflict with previous patents, has complete novelty and creativity, and is suitable for popularization and application. The method is suitable for commercialized landing and large-scale application and is extremely high in innovativeness and practicability.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence art generation, low-rank adaptation (LoRA) model training, and multi-scenario visual content customization. Specifically, it refers to an AI generation and customization method and system based on the LoRA model, developed for four core scenarios: children's art education, art gallery / museum visual customization, artist creation assistance, and enterprise-specific visual system customization. The core achieves precise satisfaction of the specific needs of each scenario by training a dedicated LoRA sub-model and optimizing the adaptation logic. At the same time, it supports multimodal input and output, style consistency assurance, and copyright traceability, adapting to the commercialization and large-scale application needs of each scenario. Background Technology

[0002] Current AI art generation technology faces a critical pain point in its application across various core art and commercial scenarios: the technology fails to meet the specific style requirements of each scenario, exhibiting weak targeting and low adaptation efficiency. Children's art scenarios struggle to balance age-appropriate aesthetics with practical pedagogical applications; art galleries / museums struggle to ensure stylistic consistency and cultural alignment across all exhibitions, cultural and creative products, and signage materials; artists struggle to accurately replicate their personal styles; and businesses struggle to balance industry visual conventions with their own brand aesthetics. Furthermore, existing technologies are mostly single-function designs, failing to form a scenario-based closed loop of "dedicated model training - accurate content generation - style calibration - copyright protection," resulting in poor practicality and an inability to truly address the actual needs of each scenario. This invention addresses this by designing technical solutions around the real needs of four major scenarios, ensuring the technology fully serves the application scenarios and filling the core gap in technology-scenario adaptation. Summary of the Invention

[0003] 1. Purpose of the invention The core objective of this invention is to create a multi-scenario art AI generation and customization method and system that highly aligns with the technical solutions and scenario requirements. Targeting four core scenarios—children's art education, art gallery / museum visual customization, artist creation assistance, and enterprise-specific visual system customization—it conducts dedicated LoRA model-oriented training and builds scenario-based functional adaptation modules, ensuring the technology fully serves the practical applications of each scenario. This achieves efficient assistance in children's art education, unified generation of style and precise cultural tone adaptation for all categories of materials in art galleries / museums, accurate replication of artists' personal styles and inspiration assistance, and visual customization for enterprises that aligns with industry attributes and exclusive aesthetics. Simultaneously, it simplifies the operation process, improves generation efficiency, ensures style consistency and copyright security, ultimately achieving deep integration of technology and scenarios to support the commercialization of various scenarios. 2. Technical Solution: This invention relies on open-source foundational models such as Stable Diffusion to build its core framework. It focuses on four specific scenarios and features three core technologies: a dedicated LoRA model training system, a scenario-based functional adaptation module, and a style calibration guarantee module. These are complemented by two general guarantee modules: multimodal input / output and copyright traceability, forming a complete technical closed loop. All technical designs directly serve the application scenarios, as detailed below: ① Dedicated LoRA model training module for multiple subjects and multiple scenarios This module is the core technical support for the implementation of all scene functions. Its core logic involves targeted material input, exclusive feature extraction, precise parameter adaptation, and model training. It trains a dedicated LoRA sub-model for each scene and each customized subject, ensuring that the generated content perfectly matches the requirements. This invention does not involve face-swapping, clone models, or other character image modification technologies. Its core focus is on the customized generation of artistic styles and visual systems, and it has no technical overlap with face-optimized artistic photo generation technologies. Specific scene-specific technical designs are as follows: • Targeting children's art education scenarios: Supports inputting aesthetically suitable materials for children of corresponding age groups (rounded and playful materials for younger children, and rich and layered materials for older children), extracting core features such as "age-specific color preferences, shape characteristics, and brushstroke norms," ​​setting adaptation parameters (Rank=10-15, α to r ratio=0.8:1), and training three age-specific LoRA sub-models to ensure that the generated works conform to the cognitive and aesthetic needs of the corresponding age group, directly serving the core needs of teachers in generating model paintings and optimizing student works; additionally, regional cultural materials (such as Xi'an shadow puppetry and paper cutting) can be imported, and the core visual features of regional elements can be extracted simultaneously and integrated into the children's exclusive model to achieve a natural combination of regional culture and children's teaching, making regional elements suitable for teaching scenarios. • For art museums / museums: Supports inputting exclusive visual materials for art museums / museums (logo, standard colors, past exhibition materials, visual elements of collections, style reference cases), accurately extracting core style features such as "exclusive color matching, font characteristics, composition logic, line style, core visual elements, and cultural tone" of art museums / museums, setting adaptation parameters (Rank=12-18, α to r ratio=0.8:1), training an exclusive unified style LoRA sub-model for art museums / museums, ensuring that all content such as posters, exhibition materials, cultural and creative products, wayfinding systems, and digital display materials of collections fully conforms to the visual pursuit, aesthetic orientation, and cultural tone of art museums / museums, guaranteeing style unity and cultural connotation consistency across all categories of materials. • For artist creation assistance scenarios: Supports inputting past original works of artists, accurately extracts the artist's core personal characteristics such as "personal unique brushstrokes, color matching, composition habits, picture texture, and artistic expression style", sets adaptation parameters (Rank=14-18, α to r ratio=0.8:1), trains the artist's personal LoRA sub-model, replicates the artist's artistic appearance 1:1, and provides core technical support for subsequent inspiration assistance, draft optimization, and creative expansion, ensuring that AI assistance does not deviate from the personal style. • For customized visual scenarios for specific enterprises: The first step is to extract the corresponding industry visual features (children's institutions with playful and saturated colors, beauty industry with gentle and soft focus, technology industry with minimalist geometric features, etc.). The second step is to input the enterprise's exclusive materials (LOGO, VI standard colors, brand patterns, past promotional materials), extract the enterprise's "exclusive aesthetic preferences and core brand visual genes", and perform dual-dimensional feature fusion training based on the industry visual feature base library and the enterprise's exclusive visual genes. This is different from the general LoRA training logic of single material input, and achieves precise adaptation between industry attributes and enterprise aesthetics. After fusing the two features, the adaptation parameters are set (Rank=11-19, α to r ratio=0.8:1) to train the enterprise's exclusive LoRA sub-model. This ensures that the generated content not only conforms to the industry attribute perception, but also accurately matches the enterprise's own aesthetics and brand style, serving the generation of commercial materials for all categories of enterprises. • General technical design: All dedicated LoRA sub-models support independent fine-tuning and superimposed calls, and reserve interfaces for iterative upgrades. They can be optimized and adjusted according to the main needs to adapt to subsequent changes in scenario requirements. At the same time, they support lightweight processing and are compatible with cloud / local dual deployment to meet the usage scenarios of different main entities. ② Module for Precise Adaptation of Functions Based on Different Scenarios Based on scenario-specific LoRA sub-models, functional logic is designed to directly translate technology into usable capabilities for each scenario. Each function corresponds to specific application requirements, with no redundant design. • Features adapted for children's art education scenarios: 1. Teacher lesson preparation sample painting generation function: Calls the corresponding age-specific LoRA model, supports the generation of multiple versions of sample paintings according to teaching theme, style, and difficulty, and allows adjustment of size and color, directly adapting to teaching use; 2. Teacher-student interactive creation optimization function: Supports student draft uploads, extracts draft outline features based on the child-specific model, generates targeted modification suggestions, and optimizes coloring and adds elements while retaining the original core; supports text / voice input to generate age-appropriate and fun works, and supports real-time fine-tuning; 3. Children's independent creation support function: Doodle recognition extension, combined with fun coloring and simple dynamic effects, fits children's creative habits, stimulates interest, and all functions are implemented based on the child-specific LoRA model to ensure that the output meets the needs of teaching and children. • Art Museum / Museum Scene Adaptation Functions: 1. Exhibition Poster Generation Function: Utilizes the art museum / museum's exclusive LoRA model, inputting the exhibition theme, size, and application scenario to generate multiple versions of posters with a unified style, automatically incorporating core exhibition information and cultural elements, and supporting fine-tuning of details; 2. Full Exhibition Visual System Generation Function: Matches the art museum / museum's exclusive style to generate all exhibition materials, including exhibition wall backgrounds, wayfinding systems, exhibition labels, and digital display backgrounds for museum collections, adapting to the exhibition space and lighting environment, and simultaneously generating reference images; 3. Cultural and Creative Products and Supporting Materials Generation Function: Links with the art museum / museum's exclusive model to generate cultural and creative products, tickets, invitations, commemorative albums, and other materials, ensuring style consistency and cultural tone consistency; 4. Visual Expansion Function for Museum Collections: Based on an exclusive model trained with museum collection materials, generates materials for artistic display, digital restoration, and cultural and creative derivative designs of cultural relics, adapting to the needs of cultural relic protection and cultural dissemination. All function outputs are centered on the art museum / museum's exclusive LoRA model, ensuring visual consistency and cultural tone consistency. • Artist Creative Assistance Features: 1. Personal Style Replication: Utilizes the artist's exclusive LoRA model, supports uploading drafts and inputting inspirational keywords, generating inspirational solutions and optimized draft versions that match the artist's personal style, reducing trial-and-error time; 2. Creative Expansion: Based on personal style, pushes suitable creative materials and composition ideas to assist in the generation of inspiration; 3. Creative Tracing: Records the creative adjustment process in real time, preserving the creative trajectory. All functions are based on the artist's exclusive LoRA model, ensuring that the artist's personal artistic style is not diluted, truly achieving inspiration assistance. • Enterprise-specific visual customization and scene adaptation functions: 1. Full-category commercial material generation function: calls the enterprise's exclusive LoRA model to generate full-category content such as posters, brochures, packaging, and store materials, supporting batch generation and serial development; 2. Style unification verification function: sets verification standards based on the enterprise's exclusive model and automatically calibrates the content style; 3. Flexible adjustment function: supports fine-tuning material details according to enterprise needs to adapt to different promotional scenarios. All functions are based on the enterprise's exclusive LoRA model, balancing industry attributes and enterprise personalized needs. ③ Style calibration and optimization module To address the core need for style consistency across various scenarios, we offer dedicated style calibration technology: 1. Based on each subject's proprietary LoRA model, we input corresponding style standard thresholds (color range, element specifications, layout requirements, cultural tone, etc.) to establish a dedicated verification system; 2. During content generation, we compare the output results with the standard thresholds in real time, automatically identifying and calibrating any deviations; 3. We support manual fine-tuning, balancing style consistency with personalized creativity, ensuring that children's artworks conform to age-appropriate aesthetics, art gallery / museum materials align with the museum's tone and cultural connotations, artists' works retain their individual characteristics, and corporate materials match brand standards, ensuring that the technology output accurately meets the needs. ④ Multimodal fusion generation module It supports multimodal input including text, images, voice, and hand-drawn doodles (matching diverse input scenarios such as teacher text instructions, student hand-drawn sketches, artist drafts, museum / galleries curatorial requirements, and enterprise requirements), and outputs high-definition images, vector graphics, CMYK format for printing, short video materials, 3D display materials, and other formats (matching diverse output needs such as teaching printing, museum / galleries exhibitions and digital dissemination, artist creation, and enterprise commercial use), ensuring convenient use in various scenarios and unimpeded technology implementation. ⑤ Copyright traceability and security module Based on the core copyright protection requirements of various scenarios, an invisible digital watermark is automatically embedded in all generated content. This watermark includes information such as generation parameters, copyright ownership, and scope of authorization. The watermark is tamper-proof and can be quickly traced, protecting both the technical rights of applicants and the content rights of collaborating entities such as art galleries / museums, artists, and enterprises. This avoids copyright disputes in commercial applications across various scenarios and provides security for the commercialization of the technology. Model training materials require legal authorization certificates from the customizing entity. The system only provides technical training services and does not participate in material collection, complying with AI training copyright compliance requirements. The corresponding supporting system consists of six core units: a dedicated LoRA model training unit, a scenario-specific function adaptation unit, a style calibration and optimization unit, a multimodal generation and output unit, a copyright traceability and security unit, and a human-computer interaction unit. These units operate in tandem, fully corresponding to the aforementioned technical solutions. The system supports cloud, local, and hybrid deployment modes, offering convenient operation. All unit functions are designed around the four major application scenarios, ensuring a high degree of integration between technology and application. 3. Beneficial effects ① Highly aligned with the actual needs of the four core scenarios: All technical designs are based on the actual needs of the four core scenarios. Dedicated LoRA model training and scenario-specific function adaptation directly solve the core pain points of each scenario, making it highly practical. ② High style accuracy: Each subject's exclusive LoRA model ensures that the generated content perfectly matches its own style requirements. Combined with style calibration technology, the style consistency is maximized, matching the core requirements of art galleries / museums, enterprises, and artists for style unity, and also meeting the age-appropriate aesthetic requirements of children's art. ③ Significantly improved efficiency: One-click generation of multiple versions of content and support for batch output greatly reduces the preparation time for children's art teachers, the design and curation cycle of materials for art galleries / museums, the trial and error cost for artists, and the production cycle of corporate promotional materials, thus improving work efficiency in various scenarios. ④ Highly convenient to use: Multimodal input + visual operation, no professional AI technology required, users in various scenarios can quickly get started, reducing the technical threshold for use; ⑤ Copyright guaranteed: Full-process copyright traceability avoids risks in commercial applications in various scenarios. At the same time, the material compliance requirements are clear, conforming to industry standards, supporting multiple deployment modes, meeting the privacy and usage needs of different entities, and adapting to long-term commercial operation. ⑥ Excellent scalability: The model supports iterative upgrades and can be optimized and adjusted according to the needs of various scenarios, adapting to subsequent scenario expansions and changes in needs, and has high long-term practical value. ⑦ Strong cultural adaptability: The exclusive design for art gallery / museum scenes ensures that the generated content not only conforms to the visual style but also accurately adapts to the cultural tone, providing technical support for cultural dissemination and cultural relic protection. Detailed Implementation Example 1: Children's Art Education Scenario 1. Preliminary preparation: In the dedicated LoRA model training module, import aesthetically suitable materials for children aged 3-6, and combine them with childlike transformation materials of Xi'an shadow puppetry. Extract the core features of rounded shapes and highly saturated colors for young children, set Rank=12 and α to r ratio=0.8:1, and train the dedicated LoRA sub-model for children aged 3-6. 2. Teacher Preparation Application: Log in to the system, select the children's art education scenario, call up the LoRA model exclusive to 3-6 year olds, input "shadow puppet rabbit, simple line drawing style, suitable for classroom teaching", and the scene-specific function adaptation module will start the sample painting generation function, generating 3 different sample paintings within 10 seconds. Teachers can fine-tune the colors and lines, and export the print format for teaching. The whole process takes no more than 3 minutes, greatly improving the efficiency of lesson preparation. 3. Student Interactive Application: After students hand-draw a rabbit sketch, they take a photo and upload it. The system then activates the interactive creation optimization function, extracts the outline of the sketch based on a unique model, and generates modification suggestions such as "making the lines more rounded and adding details to the ears". After the student confirms, the system optimizes the coloring, retains the original core, and can also generate a colored version, which meets the creative needs of young students. 4. Copyright Protection: The generated works are automatically embedded with watermarks to clearly indicate copyright ownership, are compatible with teaching scenarios and usage guidelines, and all training materials are legally authorized by teaching institutions and comply with regulations. Example 2: Art Gallery / Museum Scene 1. Preliminary preparation: In the dedicated LoRA model training module, import the logo of a museum in Xi'an, standard color chart (antique bronze + off-white), past exhibition posters, and visual elements of the museum's collection of Tang Dynasty murals. Extract its core style characteristics and cultural tone, which are antique simplicity, smooth lines, and profound cultural heritage. Set Rank=16 and α to r ratio=0.8:1, train the museum's dedicated LoRA sub-model, and generate test posters and cultural and creative design drafts to confirm that the style is completely consistent with the museum's aesthetic orientation and cultural tone. 2. Exhibition Poster Generation: Log in to the system, select the art gallery / museum scene, call the exclusive LoRA model, and input "Special Exhibition of Tang Dynasty Murals, poster size 79cm×179cm, for outdoor promotion, incorporating flying apsara elements from the murals". The scene-specific function adaptation module will start the poster generation function, generating 4 different posters, automatically incorporating the exhibition time, location, core cultural relic information, etc. The style is completely consistent with the art gallery, the cultural tone is accurately adapted, and fine-tuning of details is supported. 3. Exhibition and Cultural and Creative Product Generation: Based on the same exclusive LoRA model, activate the exhibition visual generation function to generate exhibition wall backgrounds, signage design drafts, and exhibition label templates, adapting to the exhibition hall space size and lighting environment, and incorporating elements of Tang Dynasty murals; switch to the cultural and creative product generation function to generate design drafts for notebooks, refrigerator magnets, scarves, and other cultural and creative products, with all materials having a highly unified style and consistent cultural tone; 4. Style Guarantee: The style calibration module performs real-time verification to ensure that all outputs are free from style deviation and cultural inclination. Watermarks are embedded simultaneously to protect the museum's copyright rights. Training materials are fully authorized by the museum, ensuring compliance and no risk. Example 3: Artist's Creative Assistance Scene 1. Preliminary preparation: In the dedicated LoRA model training module, import 5 original works by a certain ink painting artist, extract the core personal artistic features of dry brushstrokes, light ink wash, and blank composition, set Rank=17, α to r ratio=0.8:1, train the artist's personal dedicated LoRA sub-model, generate a test draft to confirm that it completely replicates the personal artistic style. All training materials are provided by the artist himself, and the ownership is clear. 2. Creative Assistance Application: When an artist uploads a sketch of the Big Wild Goose Pagoda and enters "Preserve the ink painting style and add the artistic conception of clouds and mist", the system will activate the creative assistance function, call up the artist's personal LoRA model, generate 3 optimized solutions that match the artist's personal style, provide different cloud and mist composition ideas, and assist the artist in making inspirational decisions. 3. Creative Expansion: Based on the artist's personal style, the system recommends suitable ink painting creation materials. The artist can further adjust the materials based on the materials. The system records the creative process in real time, which is convenient for subsequent tracing. 4. Rights Protection: The generated solution automatically embeds a watermark containing the artist's information to protect their creative copyright. This process only replicates the style and provides inspiration, and does not involve any risky content such as modification of the artist's image. Example 4: Customized Visual Scenarios for Enterprises 1. Preliminary Preparation: In the dedicated LoRA model training module, the industry visual features of the beauty industry, such as gentle soft focus, low saturation warm color tone, and floral elements, are extracted first. Then, the logo, standard colors (nude pink + off-white), and past promotional materials of a beauty salon are imported to extract its simple and fresh exclusive aesthetic features. Based on the industry visual feature base library and the enterprise's exclusive visual genes, a two-dimensional feature fusion training is carried out. This is different from the general LoRA training logic of single material input, so as to achieve accurate matching between industry attributes and enterprise aesthetics. After fusing the two features, Rank=15 and α to r ratio=0.8:1 are set to train the exclusive LoRA sub-model of the beauty salon. 2. Commercial Material Generation: Log in to the system, select the enterprise scenario, call the exclusive LoRA model, and input "Spring skincare project poster, suitable for in-store display + online promotion". The scenario-specific function adaptation module starts the material generation function to generate multiple versions of posters. The style is in line with the gentle tone of the beauty industry and fits the exclusive aesthetics of the beauty salon. At the same time, a series of materials such as promotional leaflets and store window design drafts are generated, which can be exported in batches. 3. Style Guarantee: The style calibration module verifies in real time to ensure that all materials comply with the beauty salon's VI specifications, avoiding style deviations. Watermarks are embedded to clearly indicate copyright, making them suitable for commercial promotion. Training materials are legally authorized by the enterprise, ensuring compliance and no disputes. Other implementations (school-based aesthetic education courses in primary and secondary schools, cultural and creative product development, digital display of museum artifacts, etc.) are all based on the above core technical solutions, linked with corresponding dedicated LoRA models and functional modules to ensure that the technology and scenario requirements are highly aligned and that the entire process complies with compliance requirements.

Claims

1. A method for multi-scene art AI generation and customization based on LoRA, characterized in that, Includes the following steps: (1). Build a core framework based on the LoRA model, and carry out targeted training of exclusive LoRA sub-models for four core scenarios: children's art education, art gallery / museum visual customization, artist creation assistance, and enterprise exclusive visual system customization. This invention does not involve face swapping, clone model and other character image modification technologies. The core focus is on the customized generation of art style and visual system, and there is no technical overlap with the face-optimized art photo generation technology. Input exclusive materials for each scene, extract core styles / features, set adaptation parameters (Rank value 10-19, α to r ratio = 0.8:1), and generate exclusive LoRA sub-models for each scene and subject. Among them, the enterprise-specific LoRA sub-model is trained by dual-dimensional feature fusion based on the industry visual feature base library and the enterprise's exclusive visual genes. It is different from the general LoRA training logic of single material input, and achieves accurate adaptation of industry attributes and enterprise aesthetics. The art gallery / museum-specific LoRA sub-model extracts the museum's exclusive visual features and cultural tone, and achieves style unity and cultural connotation consistency for all categories of materials. It supports fine-tuning, iterative upgrades and dual-end deployment of all models. (2). Based on each exclusive LoRA sub-model, a scene-specific function adaptation system is built to realize the implementation of the core requirements of each scene: the children's art education scene realizes the generation of teacher's model paintings and the optimization of student creative interaction; the art gallery / museum scene realizes the unified generation and cultural tone adaptation of all categories of materials such as posters, exhibitions, cultural and creative products, and wayfinding systems; the artist scene realizes the replication of personal style and the assistance of creative inspiration; and the enterprise scene realizes the generation of commercial materials that fit the industry attributes and exclusive aesthetics. (3). Establish a style calibration and optimization system, set style standard thresholds based on each exclusive LoRA sub-model, monitor and calibrate the deviation of generated content in real time, support manual fine-tuning, and ensure style consistency; (4). Configure multimodal fusion generation capability, support text, image, voice and hand-drawn doodle input, output multi-format commercial materials, and adapt to the needs of various scenarios; (5) Configure copyright traceability capabilities, automatically embed invisible digital watermarks into generated content, realize full life cycle copyright protection, model training materials must be provided with legal authorization certificates by the customized subject, the system only provides technical training services and does not participate in material collection, which complies with the AI ​​training copyright compliance requirements, and completes the whole process closed loop from model training, function adaptation, generation output to copyright protection.

2. The method according to claim 1, characterized in that, The LoRA sub-models specifically designed for children's art education scenarios are age-appropriate for training in three age groups: 3-6 years, 7-9 years, and 10-12 years. They can incorporate regional cultural elements for a more child-friendly approach and include core functions such as generating multiple versions of teacher's model paintings, optimizing student drafts, and expanding graffiti.

3. The method according to claim 1, characterized in that, The LoRA sub-model for art galleries / museums extracts the museum's exclusive color scheme, composition logic, and core visual elements, and provides supporting functions for generating exhibition posters, building exhibition systems, developing cultural and creative products, and visualizing cultural relics, while simultaneously outputting reference images for implementation.

4. The method according to claim 1, characterized in that, The LoRA sub-model for artist creation assistance extracts personal brushstrokes, color schemes, and composition habits, and is equipped with draft optimization, inspiration scheme generation, and creation trajectory recording functions to accurately replicate artistic styles.

5. The method according to claim 1, characterized in that, The Rank values ​​of the LoRA sub-models for each scenario are set differently according to the scenario: 10-15 for children's art education scenario, 12-18 for art gallery / museum scenario, 14-18 for artist scenario, and 11-19 for enterprise scenario, to adapt to the requirements of style accuracy.

6. The method according to claim 1, characterized in that, The invisible digital watermark has the characteristics of being tamper-proof and having rapid traceability. The multimodal output also includes short video materials and 3D display materials. Each exclusive LoRA sub-model supports independent calling and superimposed use.

7. A multi-scene art AI generation and customization system based on LoRA, characterized in that, It includes a dedicated LoRA model training unit, a scene-specific function adaptation unit, a style calibration and optimization unit, a multimodal generation and output unit, a copyright traceability and security unit, and a human-computer interaction unit. These units work together to implement the method described in any one of claims 1-6.