Character image generation method and system, computer equipment and storage medium

Abstract

The invention provides a figure image generation method and system, computer equipment and a storage medium, relates to the technical field of text generation images, and realizes artificial low dependency of cue words in the text generation image technology, semantic traceability of the cue words, structured analysis of documents and closed-loop integration of tool chains. The scheme comprises the following steps: acquiring literatures; using a preset large language model to perform text segment splitting on the literature to obtain a plurality of text segments, and performing character feature extraction on the text segments to obtain character feature information; performing structured analysis on the character feature information to obtain a structured feature information set; generating cue words based on the structured feature information set; inputting the cue word into a preset text graph model, and outputting a target figure image; and tracing literatures and related text segments corresponding to the cue words by utilizing the cue words.

Description

Technical Field

[0001] This application relates to the field of text-to-image technology, and in particular to a method, system, computer device, and storage medium for generating images of people. Background Technology

[0002] With the rapid development of large model technology, natural language processing technology and text-to-image (T2I) technology, the demand for using artificial intelligence to extract information about people from documents and generate corresponding images is becoming increasingly prominent.

[0003] Currently, large-scale language models have strong semantic understanding and text generation capabilities, and T2I models are also able to generate high-quality visual content, making it possible to reconstruct human figures based on documents. However, existing document parsing and image generation processes still have significant technical limitations.

[0004] First, existing text-based image generation processes heavily rely on manually generated prompts, requiring users to possess strong historical, cultural, and narrative knowledge to obtain accurate prompts. Simultaneously, the parsing results of large-scale language models on the same document vary across different runs, lacking structured constraints and making it difficult to maintain consistency in generated prompts. This results in images prone to content deviation or distortion. Second, current NLP tools primarily rely on keyword extraction, failing to automatically extract visible features of individuals from documents, such as appearance, clothing, hairstyle, and physique, as well as invisible features like temperament, style, and narrative tendencies. They also lack the ability to map content to fixed feature dimensions. When processing multiple documents, content from different sources is easily confused, and features are difficult to unify, necessitating secondary analysis and classification by manual intervention. Third, traditional text-based image platforms only provide a one-way generation mechanism of "prompts → image," failing to establish a source-tracing relationship between prompts and documents. Generated images are often unverifiable, making it impossible to check whether their feature information truly originates from the literature. This lack of consistency and interpretability makes them unsuitable for high-credibility scenarios such as academic research and cultural reconstruction. Furthermore, literature analysis, feature extraction, prompt word generation, and image generation are often scattered across different tools, lacking a unified, structured workflow. Since prompt words cannot be directly generated based on stable multidimensional feature information, and there is a lack of standardization mechanisms across different platforms, users often need to conduct extensive trial and error to obtain usable results, leading to low efficiency, high repetitive work, and unstable generated results. Summary of the Invention

[0005] Based on this, it is necessary to address the shortcomings of existing text-to-image generation technologies, such as high reliance on manual intervention for prompts, lack of semantic source tracing for prompts, inability to perform structured analysis of documents, and fragmented toolchains that make it difficult to form a closed loop. This paper proposes a method, system, computer equipment, and storage medium for generating portrait images, which achieves low manual dependence on prompts, semantic source tracing for prompts, structured analysis of documents, and closed-loop integration of the toolchain.

[0006] To achieve the above-mentioned technical effects, the technical solution of the present invention is as follows: A method for generating a person image, comprising: S1. Obtain the documents; S2. The document is decomposed into several segments using a pre-defined large language model, and the personal characteristics of the segments are extracted to obtain personal characteristic information. S3. Perform structured parsing on the character feature information to obtain a set of structured feature information; S4. Generate prompt words based on the structured feature information set; S5. Input the prompt words into the preset text-to-image model and output the target person image; S6. Use the prompt words to trace back to the corresponding documents and related texts.

[0007] Preferably, the step of performing structured parsing on the character feature information to obtain a set of structured feature information includes: S31. Map the character feature information to a fixed dimension according to a preset feature information structure table to obtain a high-dimensional semantic feature information set; S32. Calculate the semantic comparison degree between the high-dimensional semantic feature information based on the dimension of the high-dimensional semantic feature information set and the type of the document in which it is located; S33. If the semantic comparison degree is greater than the preset semantic comparison threshold, the corresponding high-dimensional semantic feature information is determined to be similar high-dimensional semantic feature information. After comparison, a set of high-dimensional semantic feature information containing similar high-dimensional semantic feature information is obtained. S34. Manually screen the set of high-dimensional semantic feature information containing similar high-dimensional semantic feature information to obtain a set of structured feature information.

[0008] Preferably, it further includes: S35. Based on the feature information structure table, perform field integrity detection on each dimension field in the structured feature information set, and determine whether a missing field is detected in each dimension field. If yes, proceed to S36; otherwise, proceed to S4. S36. Use the large language model to determine whether there is valid information in the document corresponding to the missing field. If so, complete the content of the missing field and write it into the structured feature information set. If not, execute S4.

[0009] Preferably, generating prompt words based on the structured feature information set includes: S41. Map each dimension field in the structured feature information set to a text description; S42. Adjust the text description using a preset platform feature adaptation engine to obtain preliminary prompt words; S43. Perform pre-validation on the initial prompts to obtain the final prompts.

[0010] Preferably, the prompt words include positive prompt words and negative prompt words.

[0011] Preferably, the step of using the prompt word to trace back to the corresponding literature and related text passages includes: S61. Bind and store the identifier of the target person image with the prompt word and the corresponding document to generate binding information; S62. Based on the binding information, the prompt words are retrieved using the large language model to trace back to the corresponding documents and related texts.

[0012] Preferably, it further includes: S71. Enter the new document; S72. Use a large language model to extract features from the newly added documents to obtain the feature information of the newly added individuals; S73. Calculate the semantic relevance between the newly added documents and the existing documents to obtain the corpus relevance weight, and combine it with the preset innovation absorption preference coefficient to calculate the feature compatibility weight between the newly added character feature information and the existing character feature information. S74. The corpus relevance weight and the feature compatibility weight are fused to obtain the new feature weight; S75. Adjust the newly added character feature information and the existing character feature information based on the newly added feature weights to obtain the modified character feature information.

[0013] The present invention also provides a human image generation system, comprising: The document retrieval module is used to retrieve documents; The feature information extraction module is used to extract personal features from the document using a preset large language model to obtain personal feature information. The feature information processing module is used to perform structured parsing of the person's feature information to obtain a set of structured feature information; The prompt word generation module is used to generate prompt words based on the structured feature information set; The text image module is used to input the prompt words into a preset text image model and output the target person image; The source chain construction module is used to trace the source of the prompt word to the corresponding literature and related texts.

[0014] The present invention also provides a computer device, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction that causes the processor to perform the operation of the person image generation method.

[0015] The present invention also provides a computer-readable storage medium storing at least one executable instruction, which, when executed on a computer device, causes the computer device to perform the operation of the person image generation method.

[0016] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention proposes a method, system, computer device, and storage medium for generating portrait images. First, it automatically extracts features from uploaded documents using a large language model, and then performs structured parsing of the extracted portrait features to generate a multi-dimensional set of structured feature information. Subsequently, based on this set of structured feature information, it automatically generates prompts for a text-to-image model, which are then input into the model to generate the target portrait image. Finally, through an image-based reverse lookup mechanism, it establishes a mapping relationship between the prompts corresponding to the generated image and the original document, achieving traceability of feature information. This invention effectively integrates document parsing, feature extraction, structured processing, prompt generation, and image generation into a closed-loop mechanism. Users do not need a professional historical or cultural background to efficiently obtain standardized structured prompts from documents and generate highly consistent portrait images. Simultaneously, the generated results are traceable to the original document and related text passages, ensuring information credibility and academic verifiability. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method for generating human images in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the method for generating human images in an embodiment of the present invention; Figure 3 This is a first interface diagram of the prompts and story in an embodiment of the present invention; Figure 4 This is a second interface diagram of prompts and stories in an embodiment of the present invention; Figure 5 This is a third interface diagram of the prompts and story in an embodiment of the present invention; Figure 6 This is a schematic diagram of the parameter setting interface in an embodiment of the present invention; Figure 7 This is a schematic diagram of the interface for image generation in an embodiment of the present invention; Figure 8 The image shown is a person image in the reverse image lookup example of this invention. Figure 9 This is the first interface diagram for reverse lookup in this embodiment of the invention; Figure 10 This is the second interface diagram for reverse lookup in this embodiment of the invention; Figure 11 This is a third interface diagram for reverse lookup in this embodiment of the invention; Figure 12 This is the first interface diagram generated using an image of Su Shi as the subject in this embodiment of the invention; Figure 13 This is a second interface diagram generated using an image of Su Shi as the subject in an embodiment of the present invention; Figure 14 This is a third interface diagram generated using an image of Su Shi as the subject in an embodiment of the present invention; Figure 15 This is the fourth interface diagram generated using an image of Su Shi as the subject in this embodiment of the invention; Figure 16 This is a character image generated from an image of Su Shi in an embodiment of the present invention; Figure 17 This is a structural block diagram of the human image generation system in an embodiment of the present invention; Figure 18 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention.

[0018] 181. Processor; 182. Memory; 183. Communication interface; 184. Communication bus; 185. Executable instructions. Detailed Implementation

[0019] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. It is understandable to those skilled in the art that some well-known details may be omitted from the accompanying drawings; The positional relationships depicted in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent. To better illustrate this embodiment, some parts of the accompanying drawings may be omitted, enlarged, or reduced, and do not represent actual dimensions. The descriptions of directions such as "up" and "down" are not intended to limit this patent. To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0020] Example 1 like Figure 1 and Figure 2 As shown, this embodiment provides a method for generating a person image, including: S1. Obtain the documents; In S1, specifically, the user first enters the name of the person to establish the parsing object, and then proceeds to the document upload step; the user submits the relevant PDF file on the document upload interface, and then the system automatically parses the document content.

[0021] S2. The document is decomposed into several segments using a pre-defined large language model, and the personal characteristics of the segments are extracted to obtain personal characteristic information. In S2, the system provides the content and extraction method to the large model, such as Qwen. Qwen is used to decompose the document into several segments, and the system extracts the personal features of the segments to obtain personal feature information in JSON format, which contains 18-dimensional feature information.

[0022] In this embodiment, the feature information is: {"filename":{"role":["Li Qingzhao"],"gender":["female"],"age":["adult"],"Facial features":["fair skin",...],"hairstyle":["long hair down to the shoulders",...],"headdress":["simple hairpin",...],"build":["slender and well-proportioned",...],"clothing":["plain long dress",...],"footwear":["embroidered shoes",...],"accessories":["jade pendant",...],"activity":["reading books",...],"visual background":["classical courtyard",...],"context":["Northern Song Dynasty",...],"vibe":["tranquil and elegant",...],"temperament":["gentle and resilient",...],"language":["Chinese",...],"voice":["gentle",...],"style":["Chinese classical meticulous brushwork style",...]} S3. Perform structured parsing on the character feature information to obtain a set of structured feature information; In S3, the system maps the character feature information to a fixed dimension according to a preset feature information structure table to obtain a high-dimensional semantic feature information set; based on the dimension of the high-dimensional semantic feature information set and the type of the document, it calculates the semantic comparison degree between the high-dimensional semantic feature information; if the semantic comparison degree is greater than a preset semantic comparison threshold, it determines that the corresponding high-dimensional semantic feature information is similar high-dimensional semantic feature information, and after comparing them one by one, it obtains a high-dimensional semantic feature information set containing similar high-dimensional semantic feature information; the high-dimensional semantic feature information set containing similar high-dimensional semantic feature information is fed back to the user, who can then screen it to obtain a structured feature information set.

[0023] Subsequently, the system performs field integrity checks on each dimension field in the structured feature information set based on the feature information structure table, and determines whether any missing fields are detected in each dimension field. If so, the system uses the large language model to determine whether there is valid information in the document corresponding to the missing field. If not, the system executes step S4. If there is valid information in the document corresponding to the missing field, the system completes the content of the missing field and writes it into the structured feature information set. If not, the system executes step S4.

[0024] S4. Generate prompt words based on the structured feature information set; In S4, such as Figures 3 to 5 As shown, after mapping each dimension field in the structured feature information set to text description, the system adjusts the text description using a preset platform feature adaptation engine to obtain preliminary prompt words. Finally, the preliminary prompt words are pre-validated to obtain the final positive and negative prompt words, such as... Figure 6 As shown, users can modify the prompts and add descriptions of the story background in the prompt interface.

[0025] S5. Input the prompt words into the preset text-to-image model and output the target person image; In S5, the user configures parameters such as the generation method and the number of images to be generated, and then submits a request. The system inputs the prompt words into the preset text-to-image model and outputs the target person image.

[0026] S6. Use the prompt words to trace back to the corresponding documents and related texts.

[0027] In S6, such as Figure 7 As shown, when a user clicks on the target person's image, the suggested phrases will appear in the "Text" field; clicking "Re-Fine" will take the user to a different page, such as... Figure 8 As shown, taking Li Qingzhao as an example, clicking on Li Qingzhao's image will... Figures 9 to 11As shown, the page displays the corresponding text segment and the corresponding document containing the text segment, as well as the prompt words generated from these documents. Different documents are displayed with a unique color, and prompt words of different colors are displayed at the bottom of the page, allowing users to distinguish which document the keywords were extracted from based on the color. Users can manually input / remove keywords to extract features from newly added documents using a large language model, obtaining new character feature information, or remove existing documents to remove all related keywords. Users can update or upload new documents; the system will calculate the semantic relevance between the new and existing documents, obtaining a corpus relevance weight, and combine it with a preset innovation absorption preference coefficient to calculate the feature compatibility weight between the new and existing character feature information. The corpus relevance weight and the feature compatibility weight are then merged to obtain the new feature weight. Subsequently, the new and existing character feature information are adjusted based on the new feature weight to obtain the modified character feature information. Finally, new prompt words are generated based on the modified character feature information. Users can give a new file name, save it to generate a new version, and use the new version parameters to generate new images, while the original version remains unaffected. Users can compare the completed graphs of different versions to see if they meet their needs, and can follow the prompts, corresponding literature, and corresponding texts according to different versions to complete the iteration.

[0028] Take Su Shi as an example, such as Figures 12 to 16 As shown in the figure, the various interfaces for generating the image of Su Shi are displayed, and the results are very good.

[0029] This embodiment first uses a large language model to automatically extract features from uploaded documents, and then performs structured parsing of the extracted personal feature information to generate a multi-dimensional structured feature information set. Subsequently, based on this structured feature information set, it automatically generates prompt words for a text-to-image model, which are then input into the model to generate the target person's image. Finally, through an image-based reverse lookup mechanism, it establishes a mapping relationship between the prompt words corresponding to the generated image and the original document, achieving traceability of feature information. Compared with existing technologies, this embodiment has the following advantages: 1. The multi-dimensional structured output of character feature information ensures stable, consistent, and easily reproducible extraction results. This invention represents personnel information in a unified multi-dimensional JSON format, avoiding inconsistencies in output across different runs of a large model. Feature information is automatically mapped to fixed dimensions, and filtering, normalization, and sorting are performed in the background, ensuring structural stability and consistency across multiple documents and versions.

[0030] 2. Establish a source tracing link of "document - key information - prompt words" to significantly improve content auditability. The system uses color-coding to establish a hierarchical correspondence between feature information and document sources, enabling the traceability of the origin of each feature. This mechanism avoids the problem of "multiple documents being unable to distinguish their sources" in traditional processes, making the content verifiable, comparable, and interpretable.

[0031] 3. Semantic-level deduplication mechanism reduces redundant descriptions and improves the quality of prompt words. This invention performs semantic merging and deduplication on feature information of the same dimension, reducing the length of prompt words by approximately 50%. This mechanism avoids style drift caused by keyword stacking and significantly improves the accuracy, compactness, and stability of generated prompt words.

[0032] 4. Unified prompt word structure and support for multi-platform image generation, lowering the barrier to cross-platform use. The system can automatically convert feature information into prompts suitable for different text and image generation platforms to generate 2D images, and can even generate 3D images. Users do not need to understand the grammar of prompts on different platforms to obtain images of people with a consistent style and stable performance.

[0033] 5. Supports version management for additional literature and keyword revisions, enabling iterative model updates. This invention allows users to add new documents, manually supplement words, and automatically generate a new version of the feature information set. When documents are added or deleted, the system only updates the relevant feature information dimensions without requiring manual reorganization, enabling the person model to continuously iterate and flexibly expand.

[0034] Example 2 like Figure 17 As shown, this embodiment provides a human image generation system, including: The document retrieval module is used to retrieve documents; The feature information extraction module is used to extract personal features from the document using a preset large language model to obtain personal feature information. The feature information processing module is used to perform structured parsing of the person's feature information to obtain a set of structured feature information; The prompt word generation module is used to generate prompt words based on the structured feature information set; The text image module is used to input the prompt words into a preset text image model and output the target person image; The source chain construction module is used to trace the source of the prompt word to the corresponding literature and related texts.

[0035] This embodiment first uses a large language model to automatically extract features from uploaded documents, and then performs structured parsing of the extracted character feature information to generate a multi-dimensional structured feature information set. Subsequently, based on this structured feature information set, it automatically generates prompt words for a text-to-image model, which are then input into the model to generate the target character image. Finally, through an image-based reverse lookup mechanism, it establishes a mapping relationship between the prompt words corresponding to the generated image and the original document, achieving traceability of feature information. A stable document element extraction mechanism is constructed, maintaining output consistency through a preset multi-dimensional feature information structure, ensuring a unified feature information framework for the same character even after multiple processing sessions or document changes. An automated prompt word generation system is established, directly transing the structured feature information into usable prompt words, reducing manual trial and error and avoiding reliance on the user's cultural background or language expression ability. Cross-document feature information tracing capability is achieved, allowing each keyword to be bound to its document source and original sentence, thus avoiding content confusion when merging multiple documents. Document parsing, feature information processing, prompt word generation, and image creation are integrated into a single process, reducing the disconnect caused by cross-tool operations and ensuring the continuity and consistency of the character generation process.

[0036] Example 3 This embodiment also proposes a computer device, such as... Figure 18 As shown, it includes: processor 181, memory 182, communication interface 183 and communication bus 184, wherein the processor 181, the memory 182 and the communication interface 183 communicate with each other through the communication bus 184; The processor 181, memory 182, and communication interface 183 communicate with each other via communication bus 184. Communication interface 183 is used to communicate with other network elements such as clients or other servers. The processor 181 executes executable instructions 185, specifically performing the operations of the described character image generation method. Specifically, executable instructions 185 may include program code. The processor 181 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0037] Memory 182 is used to store executable instructions 185. Memory 182 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0038] Executable instruction 185 can be invoked by processor 181 to cause the computer device to perform the following operations: S1. Obtain the documents; S2. The document is decomposed into several segments using a pre-defined large language model, and the personal characteristics of the segments are extracted to obtain personal characteristic information. S3. Perform structured parsing on the character feature information to obtain a set of structured feature information; S4. Generate prompt words based on the structured feature information set; S5. Input the prompt words into the preset text-to-image model and output the target person image; S6. Use the prompt words to trace back to the corresponding documents and related texts.

[0039] Example 4 This embodiment proposes a computer-readable storage medium storing at least one executable instruction. When the executable instruction is executed on a computer device, it causes the computer device to perform the operation of the person image generation method, specifically including the following steps: S1. Obtain the documents; S2. The document is decomposed into several segments using a pre-defined large language model, and the personal characteristics of the segments are extracted to obtain personal characteristic information. S3. Perform structured parsing on the character feature information to obtain a set of structured feature information; S4. Generate prompt words based on the structured feature information set; S5. Input the prompt words into the preset text-to-image model and output the target person image; S6. Use the prompt word to trace back to the document corresponding to the prompt word.

[0040] This embodiment first uses a large language model to automatically extract features from uploaded documents, and then performs structured parsing of the extracted personal feature information to generate a multi-dimensional structured feature information set. Subsequently, based on the structured feature information set, prompt words for the text-to-image model are automatically generated and input into the text-to-image model to generate the target person image. Finally, through an image-based reverse lookup mechanism, a mapping relationship is established between the prompt words corresponding to the generated image and the original document to achieve traceability of feature information.

[0041] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. A method for generating a person image, characterized in that, include: S1. Obtain the documents; S2. The document is decomposed into several segments using a pre-defined large language model, and the personal characteristics of the segments are extracted to obtain personal characteristic information. S3. Perform structured parsing on the character feature information to obtain a set of structured feature information; S4. Generate prompt words based on the structured feature information set; S5. Input the prompt words into the preset text-to-image model and output the target person image; S6. Use the prompt words to trace back to the corresponding documents and related texts.

2. The method for generating a person image according to claim 1, characterized in that, The step of performing structured parsing on the character feature information to obtain a set of structured feature information includes: S31. Map the character feature information to a fixed dimension according to a preset feature information structure table to obtain a high-dimensional semantic feature information set; S32. Calculate the semantic comparison degree between the high-dimensional semantic feature information based on the dimension of the high-dimensional semantic feature information set and the type of the document in which it is located; S33. If the semantic comparison degree is greater than the preset semantic comparison threshold, the corresponding high-dimensional semantic feature information is determined to be similar high-dimensional semantic feature information. After comparison, a set of high-dimensional semantic feature information containing similar high-dimensional semantic feature information is obtained. S34. Manually screen the set of high-dimensional semantic feature information containing similar high-dimensional semantic feature information to obtain a set of structured feature information.

3. The method for generating a person image according to claim 2, characterized in that, Also includes: S35. Based on the feature information structure table, perform field integrity detection on each dimension field in the structured feature information set, and determine whether a missing field is detected in each dimension field. If yes, proceed to S36; otherwise, proceed to S4. S36. Use the large language model to determine whether there is valid information in the document corresponding to the missing field. If so, complete the content of the missing field and write it into the structured feature information set. If not, execute S4.

4. The method for generating a person image according to claim 1, characterized in that, The generation of prompt words based on the structured feature information set includes: S41. Map each dimension field in the structured feature information set to a text description; S42. Adjust the text description using a preset platform feature adaptation engine to obtain preliminary prompt words; S43. Perform pre-validation on the initial prompts to obtain the final prompts.

5. The method for generating a person image according to claim 1, characterized in that, The prompts include positive prompts and negative prompts.

6. The method for generating a person image according to claim 1, characterized in that, The process of tracing the source of the prompt word to the corresponding literature and related text passages includes: S61. Bind and store the identifier of the target person image with the prompt word and the corresponding document to generate binding information; S62. Based on the binding information, the prompt words are retrieved using the large language model to trace back to the corresponding documents and related texts.

7. The method for generating a person image according to claim 1, characterized in that, Also includes: S71. Enter the new document; S72. Use a large language model to extract features from the newly added documents to obtain the feature information of the newly added individuals; S73. Calculate the semantic relevance between the newly added documents and the existing documents to obtain the corpus relevance weight, and combine it with the preset innovation absorption preference coefficient to calculate the feature compatibility weight between the newly added character feature information and the existing character feature information. S74. The corpus relevance weight and the feature compatibility weight are fused to obtain the new feature weight; S75. Adjust the newly added character feature information and the existing character feature information based on the newly added feature weights to obtain the modified character feature information.

8. A system for generating human images, characterized in that, include: The document retrieval module is used to retrieve documents; The feature information extraction module is used to extract personal features from the document using a preset large language model to obtain personal feature information. The feature information processing module is used to perform structured parsing of the person's feature information to obtain a set of structured feature information; The prompt word generation module is used to generate prompt words based on the structured feature information set; The text image module is used to input the prompt words into a preset text image model and output the target person image; The source chain construction module is used to trace the source of the prompt word to the corresponding literature and related texts.

9. A computer device, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction that causes the processor to perform the operation of the person image generation method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one executable instruction, which, when executed on a computer device, causes the computer device to perform the operation of the person image generation method as described in any one of claims 1 to 7.

Images