Intelligent image-text design method and system
By employing intelligent graphic design methods and utilizing AI-powered dynamic retrieval and adaptive layout, the problems of time-consuming and labor-intensive traditional graphic design and insufficient automation tools are solved, enabling efficient and personalized graphic design generation.
Patent Information
- Application Number
- CN202511413941.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-29
- Publication Date
- 2026-01-06
AI Technical Summary
Traditional graphic design processes are time-consuming and labor-intensive, difficult for non-professionals to implement, and user-defined needs are difficult to directly correspond to template parameters. Automated tools have limited feedback and response, requiring modifications to be made from scratch, and often result in problems such as color conflicts, element stacking, and poor readability.
The system employs an intelligent graphic design approach. It identifies design requirements by receiving fuzzy language descriptions, uses AI for dynamic retrieval and adaptive layout to generate preliminary design schemes, optimizes them based on user feedback, and finally generates graphic designs that meet user needs.
It achieves intelligent transformation from vague language to precise design, enabling non-professional users to quickly generate high-quality designs and professional users to obtain editable drafts. The more the system is used, the better it understands users, avoiding cluttered layouts and color imbalances, and improving the quality of the design.
Smart Images
Figure CN121279110A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of graphic design technology, specifically to an intelligent graphic design method and system. Background Technology
[0002] In today's e-commerce and digital media industry, designers often need to quickly create visual design works containing images and text based on vague language descriptions, such as promotional posters and web pages.
[0003] In traditional design processes, designers need to start from scratch, conceiving layouts, selecting materials, and adjusting arrangements based on experience. This is not only time-consuming and labor-intensive but also difficult for non-professionals to achieve. Furthermore, users' verbal requests are difficult to directly correspond to template parameters, leading to distorted communication. With the development of artificial intelligence technology, especially the emergence of deep learning and generative models, new possibilities have been offered for the automation and intelligence of graphic design. However, many automated tools have limited responsiveness to user feedback, requiring modifications to often start over or be manually adjusted. Moreover, automated layouts frequently suffer from color clashes, element stacking, and poor readability.
[0004] Therefore, in order to address the above problems, there is an urgent need for an intelligent graphic design method and system. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides an intelligent graphic design method and system that solves the problem of users quickly generating personalized graphic designs through fuzzy language descriptions. By utilizing AI dynamic retrieval, adaptive layout, and multimodal fusion technologies, it achieves intelligent transformation from unstructured requirements to precise design solutions.
[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent graphic design method, comprising the following steps: Step S1, by receiving the fuzzy language description input by the user, identifying the number of sections required for the graphic design and the corresponding design tags for each section, and retrieving the corresponding number of port data groups based on the number of sections and design tags; Step S2, performing multi-dimensional retrieval on the multi-level content branches of each port data group, locating the optimal branch combination in each port data group through dynamic weights, and using adaptive layout to fuse and display the optimal branch combination in each port data group to form a preliminary design scheme; Step S3, receiving user feedback on the preliminary design scheme, mapping the feedback information back to the corresponding port data group, updating the weight distribution and correlation of the port data group, re-executing the branch combination retrieval based on the updated port data group, and performing multimodal fusion processing using the re-retrieved branch combination to generate the final graphic design scheme that conforms to the user's fuzzy language description.
[0007] Furthermore, the design tags include existing tags and abstract tags; for the existing tags, a preset port data group is retrieved, which is a preset dynamic associated data cluster containing multiple branch contents; for the abstract tags, a corresponding port data group is generated by performing multi-level fine-grained partitioning.
[0008] Furthermore, the design tags include existing tags and abstract tags; for the existing tags, a preset port data group is retrieved, which is a preset dynamic associated data cluster containing multiple branch contents; for the abstract tags, a corresponding port data group is generated by performing multi-level fine-grained partitioning.
[0009] Furthermore, the specific analysis of generating corresponding port data clusters by performing multi-level fine-grained partitioning is as follows: Abstract labels are decomposed into first-level dimensions, including visual feature dimension, emotional attribute dimension, and layout logic dimension; second-level partitioning is performed on the first-level dimensions, specifically including: for the visual feature dimension, color tone, line style, and graphic complexity; for the emotional attribute dimension, emotional tendency and atmosphere intensity; and for the layout logic dimension, element density, hierarchical structure, and spatial layout; third-level feature values are set for each second-level dimension, and a mapping rule between the third-level feature values and visual elements is established through a semantic association network to generate port data clusters.
[0010] Furthermore, the specific analysis of locating the optimal branch combination in each port data group through dynamic weighting is as follows: initialize the multi-dimensional weight coefficients of the multi-level content branches, and score the multi-dimensional content branches of each port data group in multiple dimensions, including semantic matching score, visual coordination score, and historical preference score; dynamically adjust the weight ratio of each dimension according to the user's current design scenario requirements, determine the comprehensive score of each multi-level content branch based on the dynamic weights, and select the branch combination with the highest comprehensive score as the optimal branch combination.
[0011] Furthermore, the specific analysis of using adaptive layout to fuse and display the optimal branch combination in each port data group is as follows: extract the visual attribute parameters of each element in the optimal branch combination, the visual attribute parameters including size ratio, color saturation and shape complexity; identify the theme of the section based on the fuzzy language input by the user, and determine the layout priority rules based on the theme of the section; and make coordination adjustments to the layout using dynamic grid layout and color balance detection.
[0012] Furthermore, the specific analysis of re-executing the branch combination retrieval based on the updated port data group is as follows: determine the type and target element of the user feedback information, the feedback information including deletion instructions, modification instructions, and layout change instructions; map deletion instructions to the weight decay of the corresponding branch, modification instructions to the weight enhancement of the associated branches of the corresponding branch, and layout change instructions to the multi-dimensional weight coefficient reset; retain the core feature parameters of the user-confirmed element, and establish an association index between the core feature parameters and the multi-level content branches of the port data group; re-execute the retrieval determination of the optimal branch combination using the updated weight distribution, and retrieve the multi-level content branches whose core feature parameters match the user-confirmed element with a matching degree exceeding the matching degree threshold.
[0013] Furthermore, the multimodal fusion processing specifically includes semantic consistency detection, color balance evaluation, and layout balance evaluation of the adjusted modules.
[0014] An intelligent graphic design system, applying the aforementioned intelligent graphic design method, includes: a fuzzy requirement analysis module, used to identify the number of sections required for the user's graphic design and the corresponding design tags for each section by receiving the fuzzy language description input by the user, and to retrieve the corresponding number of port data groups based on the number of sections and design tags; an intelligent retrieval and fusion module, used to perform multi-dimensional retrieval of the multi-level content branches of each port data group, to locate the optimal branch combination in each port data group through dynamic weights, and to fuse and display the optimal branch combination in each port data group using adaptive layout to form a preliminary design scheme; and an interactive feedback optimization module, used to receive user feedback information on the preliminary design scheme, to map the feedback information back to the corresponding port data group, to update the weight distribution and correlation of the port data group, to re-execute the branch combination retrieval based on the updated port data group, and to perform multimodal fusion processing using the re-retrieved branch combination to generate a final graphic design scheme that conforms to the user's fuzzy language description.
[0015] The present invention has the following beneficial effects:
[0016] This intelligent graphic design method and system, through three stages—fuzzy requirement analysis, intelligent retrieval and adaptive layout integration, and interactive feedback optimization—can automatically generate high-quality graphic design solutions starting from the user's vague language descriptions, and gradually optimize them to a satisfactory level with the participation of user feedback. It introduces the multimodal understanding and generation capabilities of deep learning into the design process, enabling non-professional users to quickly complete poster and promotional image design work with the help of AI. Non-professional users only need to provide vague descriptions to obtain professional-level finished products, while professional designers can also use the system as an "inspiration amplifier" to obtain editable drafts. Dynamic weighted multi-dimensional scoring ensures that the content highly matches the user's intent from multiple perspectives—semantic, visual, and aesthetic—and feedback iteration continuously improves the customization accuracy with interaction. The adaptive layout algorithm, combined with color balance, visual contrast, and other detection rules, automatically avoids problems such as cluttered layouts, color imbalances, and difficult-to-read text, improving the overall design quality. The feedback mapping mechanism translates user operations into weight adjustments, allowing the system to better understand users the more it is used, achieving personalized recommendations and material reuse.
[0017] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description
[0018] Figure 1 This is a flowchart of an intelligent graphic design method according to the present invention.
[0019] Figure 2 This is a flowchart of step S1 in the intelligent graphic design method of the present invention.
[0020] Figure 3 This is a flowchart of step S2 in the intelligent graphic design method of the present invention.
[0021] Figure 4 This is a flowchart of step S3 in the intelligent graphic design method of the present invention.
[0022] Figure 5 This is a structural diagram of an intelligent graphic design system according to the present invention. Detailed Implementation
[0023] This application provides an intelligent graphic design method and system that combines natural language processing, multimodal content generation, and intelligent optimization algorithms to offer an innovative solution for graphic design.
[0024] The problem addressed in this application's embodiments can be summarized as follows:
[0025] The user's natural language description is decomposed into executable structured parameters using a large language model: number of sections and design tags, including existing tags and abstract tags. Pre-set material data groups are directly retrieved for existing tags. Abstract tags are decomposed into corresponding data groups using multi-level fine-grained semantic decomposition. Then, the multi-level content branches of each data group are scored in multiple dimensions, including semantics, visuals, and preferences. The optimal content combination is selected under a dynamic weighting mechanism. Based on the visual attributes of elements and section priority, algorithms such as dynamic grids and color balance detection are used to automatically lay out the layout and quickly generate a preliminary design. User feedback such as deletion, modification, and layout adjustment is mapped to weight increases, decreases, or resets, and confirmed elements are locked. The layout is re-searched and laid out on the updated weight distribution until a final solution that meets the requirements is generated.
[0026] Please see Figure 1 This invention provides a technical solution: an intelligent graphic design method, comprising the following steps: Step S1, identifying the number of sections required for the graphic design and the corresponding design tags for each section by receiving the fuzzy language description input by the user, and retrieving the corresponding number of port data groups based on the number of sections and design tags; Step S2, performing multi-dimensional retrieval on the multi-level content branches of each port data group, locating the optimal branch combination in each port data group through dynamic weights, and using adaptive layout to fuse and display the optimal branch combination in each port data group to form a preliminary design scheme; Step S3, receiving user feedback on the preliminary design scheme, mapping the feedback information back to the corresponding port data group, updating the weight distribution and correlation of the port data group, re-executing the branch combination retrieval based on the updated port data group, and performing multimodal fusion processing using the re-retrieved branch combination to generate a final graphic design scheme that conforms to the user's fuzzy language description.
[0027] Specifically, please refer to Figure 2 Step S1 is responsible for converting the user's vague language description into structured design requirement parameters, including the required number of sections and the design tags corresponding to each section. Design tags refer to a theme or style identifier for the content of each section. Using natural language processing technology, keywords and semantic cues are extracted from the user's description. For example, if a user inputs: "I want an e-commerce poster, including three sections: section A is natural scenery, section B is product display, and section C is a holiday promotion atmosphere," the system identifies that the user needs three sections and extracts a preliminary theme description for each section: section A is "natural scenery," section B is "product display," and section C is "holiday promotion atmosphere." These theme descriptions are the corresponding design tags. Step S1 includes the following steps:
[0028] Language understanding and intent recognition: Using pre-trained large-scale language models or semantic analysis algorithms, the system parses the fuzzy language input by the user, distinguishing descriptions of quantity and content within the fuzzy language, and identifying information fragments such as "several parts or sections," "the theme is...", and "contains..." elements. For example, in the above example, keywords such as "poster" and "three sections" are identified to determine the number of sections as 3, and each section description paragraph appearing in the sentence is located.
[0029] Keyword extraction and tag mapping: For each section description, core noun phrases are extracted as candidate design tags (e.g., "natural scenery," "product display," "promotional atmosphere"). Using a predefined tag knowledge base or classification model, the extracted phrases are mapped to standardized design tags. If the extracted phrase matches a known tag in the knowledge base with a match score higher than a threshold, it is marked as an existing tag; otherwise, it is considered an abstract tag. For example, "natural scenery" directly matches an existing tag in the knowledge base, while "promotional atmosphere" is not in the existing tag list and is considered an abstract concept tag.
[0030] Segment Attribute Identification: This involves analyzing the intended role or stylistic tendency of each segment within the overall design. For example, the user or context might suggest which segment is the background, which is the main content, and which is decorative or accental. This information helps determine the priority and size proportions of segments during subsequent layout planning.
[0031] In this implementation scheme, the type of design tag determines the construction method of the port data group, and the design tags are divided into two categories: existing tags and abstract tags.
[0032] Existing tags: These refer to common theme or style tags predefined and stored by the system, with corresponding structured data that can be directly utilized. For existing tags, the system directly retrieves the preset port data cluster. A preset port data cluster is a dynamically related data cluster pre-built for a known theme, containing various material content related to that theme and their branching structures. For example, for the existing tag "natural scenery," the system may pre-store a port data cluster covering a large amount of material related to natural scenery, including multiple landscape images, landscape-related text descriptions or titles, and even corresponding color schemes. This content is organized into a hierarchical branching structure, facilitating subsequent algorithmic combination and selection. The preset port data cluster can be viewed as a theme material library, but compared to a static material library, it also includes the relationships between materials and a hierarchical organization method for dynamic combination.
[0033] Abstract tags: These refer to data not directly pre-set by the system and require further analysis and retrieval to generate tags for port data clusters. These tags are typically general or have complex meanings, such as expressing an abstract style, atmosphere, or concept, like "holiday promotion atmosphere," "tech feel," or "retro fashion." For abstract tags, a multi-level, fine-grained partitioning process is performed, breaking down abstract concepts into more specific multi-dimensional feature combinations to construct the corresponding port data clusters. The specific process is as follows:
[0034] Dimensional decomposition: The meaning of abstract labels is decomposed into several primary dimensions, including three major categories: visual feature dimension, emotional attribute dimension, and layout logic dimension. For example, "holiday promotion atmosphere" can be considered in terms of color and graphic elements in terms of visual features, atmosphere and mood in terms of emotional attributes, and the arrangement of elements in terms of layout logic.
[0035] Refine the features: For each primary dimension, further subdivide typical secondary features. For example: the visual feature dimension includes color tone, line style, and graphic complexity; the emotional attribute dimension includes emotional tendency and atmosphere intensity; the layout logic dimension includes element density, hierarchical structure, and spatial layout.
[0036] Feature value settings: Several selectable feature values or levels are preset for each secondary dimension. For example, color tone includes specific options such as "warm tone", "cool tone", and "high saturation color"; emotional tendency includes "joyful" and "tense and exciting"; element density under layout logic includes "more white space" and "more elements". Each feature value corresponds to a certain visual presentation effect.
[0037] Semantic association and material mapping: A semantic association network establishes a mapping relationship between the aforementioned third-level feature values and specific visual material elements. Specifically, by defining rules or models, abstract feature descriptions are transformed into selectable or generated graphic and textual materials. For example, the semantic association network maps "warm colors" to a set of color parameters or background materials of a corresponding style, maps "joyful atmosphere" to image materials or illustrations with elements such as balloons and ribbons, and associates "high element density" with a compact layout template structure, and so on. The semantic association network can be trained by a deep learning model, enabling it to understand the correspondence between abstract semantics and visual elements.
[0038] Generate Port Data Cluster: Combining the above steps, a port data cluster is generated for this abstract tag. This cluster contains several branches, each corresponding to different feature combinations and associated materials. For example, the "Holiday Promotion Atmosphere" port data cluster generates the following branches: Branch 1: Warm and cheerful tone, dense elements, corresponding to red festive balloons and large "SALE" text; Branch 2: Cool tone, tense and exciting tone, dense elements, corresponding to cool neon lights and countdown discount text; Branch 3: Warm and cheerful tone, moderate elements, corresponding to gold ribbons and "Limited Time Offer" text, etc. These branches are organized into a multi-level structure, splitting progressively according to visual features, emotion, and layout dimensions, providing a foundation for subsequent retrieval and combination.
[0039] Specifically, please refer to Figure 3 Step S2 is responsible for retrieving the best content combination for each section's port data group and merging and arranging the content of multiple sections to form a complete preliminary design scheme. Step S2 includes two core parts: First, multi-dimensional retrieval and optimal branch combination selection, which evaluates the candidate combinations of multi-level content branches in each data group through dynamic weight evaluation and selects the optimal branch content; Second, adaptive layout fusion display, which integrates and arranges the selected content of each section according to the design rules to generate a draft of the text and graphics.
[0040] In this implementation plan, the multi-dimensional retrieval and optimal branch combination selection are specifically analyzed as follows: Each port data group contains multi-level content branches. Within each data group, the "content branch" or combination that best matches the current design requirements is found. This branch contains specific material elements and will be presented as the actual content for that section. The retrieval algorithm employs a multi-dimensional scoring mechanism and introduces dynamic weights to balance the importance of different scoring dimensions. The specific process is as follows:
[0041] The scoring criteria are defined by considering multiple dimensions and their initial weights. These dimensions include: Semantic Matching Score: This measures the degree to which the content of a branch matches the design tags and overall semantic requirements of the section. It is obtained by calculating the similarity between text tags and image content using an image-text matching model. For example, does the content of a landscape section match the semantic meaning of "natural scenery"? Visual Harmony Score: This measures the visual appeal of the content of a branch and its harmony with other sections. It is obtained by analyzing the color histogram and style feature vector of the image, and by referencing the content already selected in other sections to evaluate the harmony of the combination. For example, does the tone and style of a certain image material match the expected overall style, and is it easy to integrate with other sections? Historical Preference Score: If the system has user historical preferences or globally popular design preferences, this dimension can be introduced. It is obtained by training a preference prediction model on past feedback data. For example, if the system records the characteristics of a user's previously satisfactory design schemes, then branch content with similar characteristics will be given a higher score.
[0042] Initially, the weights of each dimension are set on average or according to a ratio based on experience. For each complete content branch combination within each port data group, its score on each of the above dimensions is calculated. Based on the initial weights, the weights are dynamically adjusted according to the user's current design scenario requirements. If the user description focuses more on style and visual effects, such as emphasizing color and atmosphere, the weight of visual harmony is increased. If the description specifically mentions the theme concept, such as highlighting a certain concept or text information, the importance of semantic matching increases. In addition, the role of the section in the overall design also affects the weight. For example, as the main visual section, visual harmony may be particularly important, while the auxiliary information section considers the accurate communication of semantics more. Dynamic weight adjustment is based on rules or a simple linear model, that is, the initial weights are proportionally amplified or reduced by judging the scenario conditions, so that the sum is still 1. For each candidate content branch, a comprehensive score is calculated according to the adjusted weights. Then, among all candidates in the port data group, the branch combination with the highest comprehensive score is selected as the optimal branch combination for the section. The specific material elements contained in this branch, such as a certain image or a certain piece of text, are determined as the content adopted for this section in the preliminary plan.
[0043] Once the content elements for each section are selected, the adaptive layout integration phase begins. This phase combines all section content based on the section theme and visual parameters to generate a complete layout design. While maintaining aesthetic balance, the design fully reflects the themes and hierarchical relationships of each section.
[0044] The adaptive layout in step S2 includes the following key points:
[0045] Visual attribute parameter extraction: Extract the visual attribute parameters of the main elements in the optimal branch of each section. These parameters include, but are not limited to, size proportion, color saturation, and shape complexity. Identify the section theme based on the user's fuzzy language input, and determine layout priority rules based on the section theme. For example, if a section plays a major display role, it is given higher priority, making it larger and more prominent in the layout. Employ a dynamic grid layout algorithm to divide the canvas into areas based on section priority and element size requirements. The dynamic grid layout is similar to a grid system in web design, automatically adjusting the grid size according to the content. First, determine the main body on the canvas. The system first divides the space into blocks, then the remaining space is divided according to the needs of the secondary blocks. The dynamic grid considers the ideal size ratio parameters of each block element to ensure that the allocated area can accommodate the content. The content of each block is then filled into the planned area, and necessary alignment adjustments are made. After the initial placement, a color balance test is performed to analyze the color distribution of the entire image, paying particular attention to whether the distribution of high-saturation colors is balanced and whether the primary and secondary colors are distinct. The main colors of each block are counted, and color differences are judged by color wheel distance or HSV space distance. The system also assesses whether the visual center of gravity has shifted. For example, if a corner is found to be bright red while the rest of the image is pale, this may lead to an imbalance.
[0046] Specifically, please refer to Figure 4 Step S3 allows users to provide feedback on the initial draft. Based on the type of user feedback, the corresponding section's port data is updated, and content retrieval and fusion are re-executed to generate the optimized final solution. This makes the design process a human-machine collaborative cycle: AI provides solutions, humans provide feedback, and AI further optimizes until the requirements are met. Interactive feedback is mainly divided into several categories: delete commands, modify commands, and layout change commands.
[0047] In this implementation, a deletion instruction indicates that the user explicitly dislikes a certain element in the initial draft and wishes to remove or replace it. For example, if the user points out that they "don't like this beach picture" as the background image, or deletes a certain image element, the system will locate which section and branch of the content the element belongs to, and then reduce the weight of that branch in the corresponding port data group.
[0048] A modification instruction indicates that the user requests to modify a certain element, such as replacing it with a different but similar element, adjusting the color or text content, etc. It shows that the user agrees with the general direction of the section, but hopes for local improvements. The modification instruction is mapped to increase the weight of related branches, while possibly slightly reducing the weight of the original branches or conflicting styles, so that the new solution is more in line with the user's intention when re-searching.
[0049] Layout change instructions indicate that users have requested adjustments to the layout itself, such as moving sections, changing their size ratio, or selecting different template layouts. This feedback is not directed at the content itself, but at the arrangement relationships. This is seen as a signal that the layout logic needs to be reconsidered, so multi-dimensional weight coefficients are reset or adjusted. For example, if a user drags an element to a more prominent position, it implies that they want that element or section to stand out more. In the recalculation, the priority of that section's content in the overall score is increased, or its area proportion in the layout algorithm is increased. Or, if the user selects a different layout template, the previous layout-related optimization constraints are cleared, and the layout is rearranged according to the requirements of the new template. Therefore, the processing of layout change instructions can be summarized as: resetting the weight of layout-related parameters and preparing for reintegration under the new layout rules.
[0050] After processing the above three types of feedback, the elements confirmed by the user are protected. Confirmed elements refer to those parts that the user has not objected to and is satisfied with. Their core characteristic parameters should be continued in the new solution. For example, if the user is satisfied with the product photos in section B, then the updated search should ensure that the same photo or at least a similar photo continues to be used; if the user approves of the wording and font of a title, then these textual features should be retained in the final solution. To this end, the system establishes an association index between core characteristic parameters and the content branches of the port data group: marking the branches to which these confirmed elements belong and locking their key attributes, such as image ID or category, text content, color, etc. In the next round of retrieval, hard constraints or forced matching are set for these features to ensure that the optimization process does not unintentionally change the parts that the user is already satisfied with, and only adjustments are made to the remaining unsatisfactory parts.
[0051] After completing feedback parsing and data updates, the optimal branch combination retrieval is re-executed based on the updated port data clusters. At this point, the content score distribution and weights of each data cluster have changed, requiring the selection of new content combinations. For example, in the background section data cluster, the previously rejected "beach" branch has been downgraded, while the "forest" branch has been upgraded. Therefore, the new optimal combination might be a forest graph. After reselecting the content, the system performs multimodal fusion processing again, repeating the layout fusion steps to generate an improved design scheme. The following points should be considered when performing multimodal fusion processing using the re-retrieved branch combinations:
[0052] Maintain semantic consistency: Ensure that the modified design still meets the user's initial vague semantic requirements. For example, after the background is changed to a forest, the overall design should still reflect "natural scenery". The system performs semantic consistency checks on the final combination of images and text. For example, it uses an image description model to generate a description for each image and compares it with the text to check if there are any parts that violate the user's intention. If so, manual intervention or readjustment may be required.
[0053] Color balance re-check: Due to changes in the colors of some elements, it is necessary to re-evaluate the overall color. For example, if a user changes the bright red promotional text to green, the new scheme may have a green forest background and green text, which may lack contrast and be less eye-catching. Therefore, a color balance assessment should be performed again.
[0054] Layout balance recheck: If the user adjusts the layout, the new layout structure needs to be verified for balance again. For example, if the user enlarges the product image, the system checks whether it has squeezed other elements, causing imbalance. If necessary, the spacing between surrounding elements is readjusted, and the layout quality is reviewed again using the same rules as the initial draft.
[0055] Please see Figure 5 An intelligent graphic design system, applying the aforementioned intelligent graphic design method, includes: a fuzzy requirement analysis module, used to identify the number of sections required for the user's graphic design and the corresponding design tags for each section by receiving the fuzzy language description input by the user, and to retrieve the corresponding number of port data groups based on the number of sections and design tags; an intelligent retrieval and fusion module, used to perform multi-dimensional retrieval of the multi-level content branches of each port data group, to locate the optimal branch combination in each port data group through dynamic weights, and to fuse and display the optimal branch combination in each port data group using adaptive layout to form a preliminary design scheme; and an interactive feedback optimization module, used to receive user feedback information on the preliminary design scheme, to map the feedback information back to the corresponding port data group, to update the weight distribution and correlation of the port data group, to re-execute the branch combination retrieval based on the updated port data group, and to perform multimodal fusion processing using the re-retrieved branch combination to generate a final graphic design scheme that conforms to the user's fuzzy language description.
[0056] In summary, this application has at least the following effects:
[0057] Through three stages—fuzzy requirement analysis, intelligent retrieval and adaptive layout integration, and interactive feedback optimization—this method can automatically generate high-quality graphic design solutions starting from the user's vague verbal descriptions, and gradually optimize them to a satisfactory level with user feedback. This approach introduces the multimodal understanding and generative capabilities of deep learning into the design process, enabling non-professional users to quickly complete design work such as posters and promotional images with the help of AI. Simultaneously, for design professionals, it can serve as an intelligent assistant, accelerating the generation of initial drafts and providing creative references.
[0058] Those skilled in the art will understand that embodiments of the present invention can be provided as methods or systems. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0059] This invention is described with reference to flowchart illustrations and structural diagrams of methods and systems according to embodiments of the invention. It should be understood that the combination of each process and module in the flowchart and structural diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 One or more processes and structures Figure 1 A device for a function specified in one or more modules.
[0060] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and structures Figure 1 The function specified in one or more modules.
[0061] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and structures Figure 1 The steps of a specified function in one or more modules.
[0062] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0063] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. An intelligent graphic design method, characterized in that, The method comprises the following steps: Step S1, identifying the number of blocks required by the user graphic design and the design labels corresponding to each block by receiving the user's input of the ambiguous language description, and calling the corresponding number of port data groups according to the number of blocks and the design labels; Step S2, performing multi-dimensional retrieval on the multi-level content branches of each port data group, respectively locating the optimal branch combination in each port data group through dynamic weights, and using adaptive layout to display the optimal branch combination in each port data group, forming a preliminary design scheme; Step S3, receiving the feedback information of the user on the preliminary design scheme, and mapping the feedback information back to the corresponding port data group, updating the weight distribution and association relationship of the port data group, and re-executing the branch combination retrieval based on the updated port data group, performing multi-modal fusion processing using the re-searched branch combination, and generating a final graphic design scheme that meets the user's ambiguous language description.
2. The intelligent graphic design method according to claim 1, wherein, The design label includes an existing label and an abstract label; For the existing label, a preset port data group is called, and the preset port data group is a dynamic association data cluster containing multiple branch contents; For the abstract label, the corresponding port data group is generated by performing multi-level fine-grained division.
3. The intelligent graphic design method according to claim 2, wherein, The specific analysis of generating the corresponding port data group by performing multi-level fine-grained division is as follows: The abstract label is disassembled into a one-level dimension, which includes a visual feature dimension, an emotional attribute dimension, and a layout logic dimension; The one-level dimension is divided into two levels, and the specific two-level division includes: for the visual feature dimension, including color keynote, line style and graphic complexity; for the emotional attribute dimension, including emotional tendency and atmosphere intensity; for the layout logic dimension, including element density, hierarchical structure and space layout; For each two-level dimension, set three characteristic values, establish a mapping rule between the three characteristic values and the visual elements through a semantic association network, and generate a port data group.
4. The intelligent graphic design method according to claim 1, wherein, The specific analysis of locating the optimal branch combination in each port data group through dynamic weights is as follows: Initialize the multi-dimensional weight coefficients of the multi-level content branches, and perform multi-dimensional scoring on the multi-level content branches of each port data group, including semantic matching degree score, visual coordination score and historical preference score; Dynamically adjust the weight proportion of each dimension according to the current design scene demand of the user, determine the comprehensive score of each multi-level content branch based on the dynamic weight, and select the branch combination with the highest comprehensive score as the optimal branch combination.
5. The intelligent graphic design method according to claim 1, wherein, The specific analysis of using adaptive layout to display the optimal branch combination in each port data group is as follows: Extract the visual attribute parameters of each element in the optimal branch combination, including size proportion, color saturation and shape complexity; Determine the layout priority rules based on the block theme identified by the user's input of the ambiguous language; Use dynamic grid layout and color balance detection to adjust the layout coordination.
6. The intelligent graphic design method according to claim 1, wherein, The specific analysis of re-executing branch combination retrieval based on the updated port data group is as follows: Determine the type and target element of user feedback information, including deletion instructions, modification instructions and layout change instructions, map the deletion instructions to the weight attenuation of the corresponding branch, the modification instructions to the weight enhancement of the associated branch of the corresponding branch, and the layout change instructions to the multi-dimensional weight coefficient reset; Preserve the core feature parameters of the user confirmation element, and establish the association index of the core feature parameters and the multi-level content branch of the port data group; Re-execute the retrieval determination of the optimal branch combination using the updated weight distribution, and retrieve the multi-level content branch whose matching degree with the core feature parameters of the user determined element exceeds the matching degree threshold.
7. The intelligent graphic design method according to claim 1, wherein, The multi-modal fusion processing specifically includes semantic consistency detection, color balance evaluation and layout balance evaluation on the adjusted plate.
8. An intelligent graphic design system, which applies the intelligent graphic design method of any one of claims 1-7, characterized in that, It includes: The fuzzy requirement analysis module is used to identify the number of plate blocks and the design labels required for each plate block by receiving the user's input of fuzzy language description, and to retrieve a corresponding number of port data groups according to the number of plate blocks and design labels; The intelligent retrieval and fusion module is used to perform multi-dimensional retrieval on the multi-level content branch of each port data group, dynamically locate the optimal branch combination in each port data group through dynamic weight, and display the optimal branch combination in each port data group through adaptive layout, forming a preliminary design scheme; The interactive feedback optimization module is used to receive user feedback information on the preliminary design scheme, map the feedback information back to the corresponding port data group, update the weight distribution and association relationship of the port data group, re-execute the branch combination retrieval based on the updated port data group, and generate a final graphic design scheme that meets the user's fuzzy language description by using the multi-modal fusion processing of the re-searched branch combination.