A method and system for prompt word identification and parsing towards multi-platforms

Abstract

This application relates to the field of artificial intelligence technology and discloses a method for recognizing and parsing prompt words across multiple platforms. The method includes: receiving prompt words input by a user; performing platform feature matching on the prompt words, and calling the parser of the platform corresponding to the prompt words based on the matching results to parse the prompt words and extract the structured data corresponding to the prompt words; and converting the structured data into a preset data format for standardization of the prompt words. This method can automatically identify the prompt word formats of different platforms and uniformly parse them into a standard data structure, thereby achieving cross-platform prompt word management.

Description

1. Description of a Prompt Recognition and Parsing Method and System for Multiple Platforms Technical Field This application relates to the field of artificial intelligence technology, and more particularly to a prompt recognition and parsing method and system for multiple platforms. Background Technology Existing AI platforms (such as Midjourney, Stable Diffusion, ComfyUI, DALL-E, GPT, etc.) use completely different prompt formats and parameter syntaxes. Users cannot uniformly manage, search, and reuse prompts across platforms. Currently, there is no tool that can automatically identify the platform to which a prompt belongs and parse it into unified structured data. Specifically, in the field of AI-generated content (AIGC), users typically need to use multiple different AI platforms to complete creative tasks, such as using Midjourney to generate images, Stable Diffusion for fine-tuning parameters, ComfyUI to build workflows, and using large language models such as Claude and GPT to assist in text processing. Each platform relies on user-input prompts to guide the generation process; however, the prompt formats and parameter syntaxes of different platforms are completely heterogeneous. In existing technologies, users mainly manage prompts in the following ways: First, they use general clipboard tools (such as Ditto) to save the original text, but these tools can only save the text as is and cannot understand the semantic structure of the prompts; second, they use note-taking applications (such as Notion) for manual categorization and labeling, but users need to maintain the label system themselves, which is inefficient and prone to omissions; third, they rely on the history function built into each AI platform, but the platforms are isolated from each other and cannot communicate with each other. None of the above-mentioned existing technologies can meet the unified management needs of cross-platform prompts, and there is an urgent need for a new method that can automatically identify the platform to which the prompts belong and extract structured information. The present application addresses the technical problem of providing a multi-platform prompt recognition and parsing method and system that can automatically identify the prompt formats of different platforms and manage prompts across platforms. HK 30135843 A 2 To solve the above-mentioned technical problems, the embodiments of this application adopt the following technical solutions: In a first aspect, the embodiments of this application provide a prompt word recognition and parsing method for multiple platforms, including: receiving prompt words input by a user; performing platform feature matching on the prompt words, and calling the parser of the platform corresponding to the prompt words to parse the prompt words according to the matching results, and extracting the structured data corresponding to the prompt words; converting the structured data into a preset data format to standardize the prompt words. In some embodiments, performing platform feature matching on the prompt words includes: performing feature matching on the prompt words through a pre-stored feature signature library, and identifying the platform corresponding to the prompt words according to the matching results, wherein the feature signature library stores feature rules corresponding to each platform.In some embodiments, feature matching of prompt words is performed using a pre-stored feature signature library, including: matching prompt words using regular expressions, and / or matching prompt words using a keyword set. In some embodiments, structured data is converted into a preset data format, including: mapping parameters used to control the same generation features in each platform to the same field. In some embodiments, each feature rule contains multiple feature items, each feature item having a preset weight value; the prompt word is parsed by calling the parser of the platform corresponding to the prompt word according to the matching result, including: calculating a weighted confidence score based on each feature item; if the weighted confidence score is higher than a preset score threshold, the platform corresponding to the weighted confidence score is taken as the matching result, and the prompt word is parsed by calling the parser of the platform corresponding to the prompt word according to the matching result. In some embodiments, the method further includes: if the weighted confidence score is lower than a preset score threshold, the preset parser is called to parse the prompt word, and the platform corresponding to the prompt word is marked as an unidentified platform. In some embodiments, the preset data format includes the platform identifier, the platform model version, the main text of the prompt word, the name and corresponding parameter value of the generation parameter, and the prompt word input by the user. In some embodiments, the input methods for receiving user input prompts include at least one of the following (HK 30135843 A 3): manual input, automatic capture by browser extensions, clipboard detection, file import, and application programming interface (API) reception. In some embodiments, the method further includes: storing structured data in a database, and classifying the structured data according to one or more dimensions selected from platform identifier, the purpose of the prompt, and keywords extracted from the main text of the prompt according to preset rules. Secondly, embodiments of this application also provide a prompt recognition and parsing system for multiple platforms, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method provided in the first aspect. The beneficial effects of this application's embodiments are as follows: Unlike existing technologies, this application provides a multi-platform prompt word recognition and parsing method, comprising: receiving prompt words input by a user; performing platform feature matching on the prompt words, and based on the matching results, calling the parser of the platform corresponding to the prompt words to parse the prompt words and extract the structured data corresponding to the prompt words; and converting the structured data into a preset data format for standardization of the prompt words. This method can automatically identify the prompt word formats of different platforms and uniformly parse them into a standard data structure, thereby achieving cross-platform management of prompt words.The accompanying drawings illustrate one or more embodiments by way of example through the corresponding images. These illustrative descriptions do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings represent similar elements. Unless otherwise stated, the figures in the drawings do not constitute a limitation on scale. Figure 1 is a flowchart of a multi-platform prompt word recognition and parsing method provided by an embodiment of this application; Figure 2 is a detailed flowchart of a multi-platform prompt word recognition and parsing method provided by an embodiment of this application. Figure 3 is a structural schematic diagram of a multi-platform prompt word recognition and parsing device HK 30135843 A 4 provided by an embodiment of this application. Detailed Description The present application will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application. To make the purpose, technical solution, and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. In addition, the terms "first," "second," and "third" used herein do not limit the data or execution order, but only distinguish identical or similar items with substantially the same function and effect. Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items. Furthermore, the technical features involved in the various embodiments of this application described below can be combined with each other as long as they do not conflict with each other. In the field of AI-generated content (AIGC), users typically need to use multiple different AI platforms to complete creative tasks, such as using Midjourney to generate images, Stable Diffusion for fine-tuning parameters, ComfyUI to build workflows, and large language models such as Claude and GPT to assist in text processing.Each platform relies on user-input prompts to guide the generation process; however, the prompt formats and parameter syntax are completely heterogeneous across platforms. For example, Midjourney uses natural language with double hyphen parameters (e.g., --ar, HK 30135843 A 5 --v); Stable Diffusion uses key-value pair labels (e.g., Steps:, CFG scale:); ComfyUI uses JSON-formatted node workflow descriptions; and prompts for large language models are usually pure natural language text, lacking a unified structural identifier. Currently, prompt processing typically treats all prompts as ordinary text strings. Clipboard tools can only save the text as is, note-taking applications require users to manually categorize and tag them, and the history records built into each platform are isolated and cannot be shared. In view of this, referring to Figure 1, this application embodiment provides a prompt recognition and parsing method S100 for multiple platforms, specifically including: S10: Receiving user-input prompts. When recognizing and parsing prompts from multiple platforms, the user-input prompts are first received. The aforementioned prompts refer to the instruction text input by the user to the AI ​​platform to generate content (such as images or text). These prompts can include natural language descriptions, generation parameters, model identifiers, or workflow configurations. Based on these user-input prompts, further recognition and analysis can be performed. In some embodiments, receiving user-input prompts can be achieved through various input methods, such as manual input (user directly pastes or types), browser extension automatic capture (extracting from web pages via browser plugins), clipboard monitoring (detecting text in the system clipboard), file import (reading JSON or text files, such as ComfyUI workflows), and application programming interface (API) reception. These input methods can be flexibly configured by those skilled in the art according to actual application scenarios, and this application does not impose any limitations. By employing diverse input methods to receive user-input prompts, the convenience and automation of data collection can be improved. This not only allows users to flexibly choose according to their usage habits or application scenarios, providing convenience for users, but also facilitates integration with other tools, broadening the application scope of the recognition and analysis method and improving its applicability. S20: Perform platform feature matching on the prompt words, and based on the matching results, call the parser of the platform corresponding to the prompt words to parse the prompt words and extract the structured data corresponding to the prompt words. After receiving the prompt words input by the user, the corresponding platform is identified and matched based on the input prompt words, that is, platform feature matching is performed on the received prompt words.Platform feature matching (HK 30135843 A 6) refers to analyzing specific grammatical patterns, keywords, or structural features in prompt words to determine which AI platform the prompt word originates from, such as Midjourney, Stable Diffusion, or ComfyUI. The result of platform feature matching is called the matching result, which represents the most likely corresponding platform for the prompt word. Then, based on the matching result, the parser corresponding to that platform is invoked to parse the input prompt word and extract the structured data corresponding to it. The parser is an extraction program specifically designed for the prompt word format of a particular platform; it can extract structured information from the user-inputted raw prompt word. For example, if the recognition result is from the Midjourney platform, the Midjourney parser is invoked to extract the main descriptive text (prompt_text) from the prompt words, i.e., all the text before the first parameter marker "--"; the parameters starting with "--" and their values ​​are extracted, including but not limited to the aspect ratio parameter ar, version parameter v, stylization parameter s, disorder parameter c, style preset parameter style, Niji mode parameter niji, etc.; and the model version is determined based on the version parameter. If the recognition result is from the Stable Diffusion platform, the Stable Diffusion parser is invoked to separate the positive prompt words (Positive Prompt) and negative prompt words (Negative Prompt); the generation parameters are extracted, including the number of sampling steps (Steps), prompt word compliance (CFG Scale), random seed (Seed), sampler name (Sampler), image size (Size), model name and version (Model), etc. If the recognition result is the ComfyUI platform, the ComfyUI parser is invoked to read the workflow file (workflow.json) and verify that it is in valid JSON format. Each node in the node graph (nodes array) is traversed to extract its type (class_type), input parameters (inputs), and node title (_meta.title). Simultaneously, the connections between nodes (links array) are parsed to establish dependencies, thus preserving the complete workflow topology. The parsed structured data refers to data with fixed fields formed by extracting and reorganizing the original unstructured prompt text according to a preset data format. The parsed structured data includes at least all the key semantic information representing the original prompt: it indicates the source of the prompt, its content intent, generation control conditions, and complete source information.By parsing the prompt words using a dedicated parser for the corresponding platform, the corresponding structured data can be extracted. This process breaks down text, which mixes natural language and specific grammatical symbols, into standardized expressions that are clearly understandable by machines. In some embodiments, the structured data includes at least the text content of the prompt words, generation parameters, and model version. By calling the corresponding parser based on the platform matching result to parse the input prompt words, the structured information in the prompt words from different platforms can be accurately extracted, providing a data foundation for subsequent unified standardization processing. In some embodiments, the "platform feature matching of prompt words" in step S20 specifically includes: S21: performing feature matching of prompt words using a pre-stored feature signature library, and identifying the platform corresponding to the prompt words based on the matching result. The feature signature library stores feature rules corresponding to each platform. Optionally, when performing platform feature matching on the prompt words input by the user, feature matching can be performed using a pre-stored feature signature library. The feature signature library is an extensible data structure that stores a set of corresponding feature rules for each AI platform. Feature rules can include regular expression patterns (e.g., the `--ar` parameter pattern for Midjourney), keyword sets (e.g., the `Steps:` tag for Stable Diffusion), or structural judgment conditions (e.g., whether it is in JSON format). For example, the feature rules corresponding to each platform include, but are not limited to: when the prompt word contains parameter identifiers in double hyphen format such as "--ar", "--v", "--s", "--style", it is identified as the Midjourney platform; when the prompt word contains generation parameter tags such as "Steps:", "CFG:", "Seed:", "Sampler:", it is identified as the Stable Diffusion platform; when the prompt word contains model identifiers such as "flux-dev" or "flux-schnell", it is identified as the Flux platform; and when the prompt word is in JSON format and contains a node graph structure, it is identified as the ComfyUI platform. Based on the above stored feature rules, the platform to which the prompt word belongs can be identified. Because the signature library is scalable and can dynamically add feature rules for new platforms as needed, the above method of matching prompt words with platform feature rules allows the recognition and parsing method to flexibly adapt to emerging new AI platforms, exhibiting good foresight and maintainability. In some embodiments, the aforementioned step S21 specifically includes: S211: matching prompt words using regular expressions, and / or matching prompt words using the keyword set HK 30135843 A 8.Regular expressions are used to match prompt words, specifically by using a regular expression engine to detect the existence of platform-specific parameter format patterns. Keyword sets are also used to match prompt words, specifically by searching for platform-specific identifiers. Regular expressions are a formal language used to describe string patterns, capable of accurately identifying parameter formats unique to each AI platform. Keyword sets are a set of platform-specific identifiers; these keywords are frequently appearing identifiers in prompt words for a specific AI platform. Keyword matching is a lightweight and fast matching method suitable for detecting words with obvious platform characteristics. For example, if the user inputs the prompt word "a cyberpunk girl --ar 16:9 --v 6.1", regular expression matching detects a parameter pattern containing double hyphens followed by parameter name, whitespace, numbers, colons, and numbers (a width-to-height ratio parameter pattern), and a version parameter pattern containing double hyphens followed by parameter name, whitespace, numbers, periods, and numbers. These patterns belong to the feature rules of the Midjourney platform in the feature signature database; therefore, the platform corresponding to this prompt word is identified as Midjourney. If a user inputs the prompt "Steps: 30, CFG scale: 7, Seed: 12345", keyword matching detects the keywords "Steps:", "CFG scale:", and "Seed:" appearing in the prompt. These keywords belong to the feature rules of the Stable Diffusion platform in the feature signature library. Therefore, the platform corresponding to the prompt is identified as Stable Diffusion. It is understandable that the above two methods can be used individually or in combination to improve the accuracy and robustness of the recognition. Regular expressions can accurately match platform-specific parameter syntax, avoiding false positives; keyword sets can quickly capture platform identifiers. Furthermore, when adding a new platform, only the corresponding regular expression pattern or keyword needs to be added to the feature signature library, enhancing the applicability of the recognition and parsing method. In some embodiments, each feature rule includes multiple feature items, and each feature item has a preset weight value; the aforementioned step S21, "performing feature matching on the prompt word through a pre-stored feature signature library, and identifying the platform corresponding to the prompt word based on the matching result," specifically includes: S22: calculating a weighted confidence score based on each feature item; if the weighted confidence score is higher than a preset score threshold, then the platform corresponding to the weighted confidence score is taken as the matching result, and the parser of the platform corresponding to the prompt word is called to parse the prompt word based on the matching result. It is understood that in some embodiments, the feature rule corresponding to each platform in the feature signature library includes multiple feature items, and each feature item has a preset weight value.In this system, a feature term is the smallest detection unit in the feature signature library, which can be a regular expression, a keyword, or a structural judgment condition. The preset weight value is a numerical representation of the importance of each feature term, reflecting its importance in identifying the corresponding platform. A higher weight value indicates greater recognizability of the feature term. For example, in the Midjourney platform, the "--v" parameter (version number) is crucial for identifying the platform and can be assigned a high weight (e.g., 25), while some auxiliary parameters can be assigned lower weights (e.g., 10). Based on this, when performing feature matching on prompt words using the feature rules in the pre-stored feature signature library, the platform matching result corresponding to the prompt word is determined based on the weight value of each matched feature term. The weight values ​​of all successfully matched feature terms for the same platform are summed to obtain the weighted confidence score for that platform. Combining the final weighted confidence score for each platform with a preset score threshold, the platform matching result corresponding to the prompt word is confirmed. The weighted confidence score quantifies the reliability of the match between the prompt word and the platform. The preset score threshold is a pre-set threshold for the weighted confidence score, used to determine whether the weighted confidence score is sufficient to reliably identify the platform. If the weighted confidence score is higher than the preset score threshold and higher than the weighted confidence scores of all other platforms, then the platform is determined to be a match result, and the corresponding parser for that platform is then called to parse the prompt word. For example, suppose the feature signature library is configured with three feature items for the Midjourney platform, with a preset score threshold of 30. The feature item used to match aspect ratio parameter patterns (such as --ar 16:9) has a weight of 25, the feature item used to match version parameter patterns (such as --v 6.1) has a weight of 25, and the feature item used to match the keyword "--style raw" has a weight of 15. The user inputs the prompt "a cyberpunk girl --ar 16:9 --v 6.1". By matching the aspect ratio and version parameters, the weighted confidence score for this platform is 25 + 25 = 50. Since this platform's weighted confidence score of 50 is higher than the preset threshold of 30 and also higher than the weighted confidence scores of other platforms, the Midjourney platform is considered to be the platform corresponding to this prompt, and the Midjourney parser is invoked to parse the user input. It is understood that the aforementioned preset threshold can be pre-set by those skilled in the art according to actual application scenarios, and this application does not impose any limitations on it.By combining the platform's weighted confidence score and a preset score threshold to determine the final matching result of the prompt word, the reliability of platform identification for the input prompt word can be effectively improved. Important matching features (HK 30135843 A 10) have a greater weight in the decision-making process, avoiding misjudgments caused by interference from secondary features, thus improving the accuracy of platform identification for the input prompt word. S23: If the weighted confidence score is lower than the preset score threshold, the preset parser is called to parse the prompt word, and the platform corresponding to the prompt word is marked as an unidentified platform. It is understandable that if the weighted confidence scores of all platforms are lower than the preset score threshold, the platform to which the prompt word belongs cannot be accurately identified. In this case, the preset parser is called to parse the prompt word, and the platform corresponding to the prompt word is marked as an unidentified platform for subsequent manual confirmation by the user. The aforementioned preset parser is a general parser. The general parser uses heuristic rules, such as detecting the existence of negative prompt word separators, key-value pair patterns, etc., to extract identifiable platform feature information within the prompt word. For example, if the user inputs the prompt "Create a beautiful landscape", the preset score threshold is 30. Since the prompt does not contain any features of any known platform, after feature matching of the prompt using the pre-stored feature signature library, the weighted confidence score of all platforms is 0, which is lower than the preset score threshold. That is, the above recognition and parsing method cannot identify the platform corresponding to the prompt. Based on this, the preset parser is called to parse the prompt and the platform is marked as an unrecognized platform, prompting the user to manually confirm. By setting the preset score threshold, it is possible to confirm that a dedicated parser is called when there is a prompt with a high match with the platform. When the above scenario does not exist, the preset parser is called to parse the prompt and mark the corresponding platform as an unrecognized platform. This ensures the robustness and fault tolerance of the recognition and parsing method, while also allowing the user to manually correct and supplement the feature signature library. This ensures the accuracy of the above method and improves its applicability. S30: Convert the structured data into a preset data format to standardize the prompt. After the parser extracts the structured data corresponding to the prompt words, it converts the extracted structured data into a preset data format. The preset data format is a standardized data format pre-defined. By converting the structured data extracted by the parser into this preset standardized format, standardized processing of the structured data is achieved, enabling structured data from different platforms to be stored, retrieved, and compared using the unified fields and value ranges specified in HK 30135843 A 11. The aforementioned preset data format can be customized by those skilled in the art according to actual needs, and this application does not impose any limitations on it.In some embodiments, the preset data format includes the platform identifier, the platform's model version, the main text of the prompt word, the name and corresponding parameter value of the generation parameter, and the prompt word input by the user. The platform identifier identifies the AI ​​platform from which the prompt word originates, clarifying its source and facilitating subsequent classification, filtering, and statistical analysis by platform. Examples include midjourney, stable_diffusion, comfyui, or unidentified. The platform's model version represents the AI ​​model version that generated the prompt word, preserving model evolution information and helping users choose the appropriate version for reuse, avoiding differences in generation results due to version incompatibility. Examples include "6.1" for Midjourney and "SDXL 1.0" for Stable Diffusion. The main text of the prompt word is the core descriptive text extracted from the original prompt word, removing platform-specific parameter markers and grammatical symbols. It is used to restore the user's creative intent and serves as the primary basis for cross-platform retrieval, tag extraction, and similarity comparison. The names and corresponding values ​​of the generated parameters are stored as a set of key-value pairs, containing all the generation control parameters in the original prompt words. This preserves the user's specific control settings for the generation process and provides a data foundation for subsequent parameter comparison, optimization analysis, and cross-platform parameter mapping. For example, the names and corresponding values ​​of generated parameters on the Midjourney platform are stored as {"ar":"16:9", "s":750}, and on the Stable Diffusion platform, they are stored as {"steps":30, "cfg":7}. The prompt words entered by the user refer to the original string entered by the user, without any modification or deletion. This ensures the traceability of information; when the parsing result is incorrect or needs to be re-parsed, it can be reprocessed based on the original input to avoid information loss. By standardizing the structured data extracted by the parser using a preset data format, unified retrieval and filtering across platforms are achieved. Users can simultaneously search for suggestion words from multiple platforms such as Midjourney and Stable Diffusion using the same dimension, without needing to construct separate queries for each platform. Furthermore, the preset data format includes fields such as platform identifier, model version, main text, generated parameter key-value pairs, and original input text. These fields comprehensively cover the core information of the suggestion words, preserving the traceability of the original input while extracting key structured information, providing a data foundation for subsequent applications such as storage, search, classification, and version management.HK 30135843 A 12 In some embodiments, the aforementioned step S30, "converting structured data into a preset data format," specifically includes: S31: Mapping parameters used to control the same generation features across platforms to the same field. As described above, the preset data format is used to standardize the structured data extracted by parsers from different platforms. This standardization process mainly involves mapping parameters used to control the same generation features across different platforms to the same field. The so-called "same generation features" refer to parameters that have a similar impact on the AI ​​generation results, such as aspect ratio (Midjourney's --ar and Stable Diffusion's Size), prompt word compliance (Midjourney's --s and Stable Diffusion's CFG scale), etc. Mapping refers to uniformly converting the original parameter names and values ​​from different platforms into standard field names and normalized values ​​in the preset data format according to preset conversion rules. This includes the correspondence of field names and the conversion of numerical values. After mapping, the functionally equivalent parameters of different platforms have unified field names in the preset data format, thereby achieving cross-platform semantic alignment, i.e., cross-platform standardization. The above method realizes a unified expression of cross-platform parameters, allowing users to search and compare prompt words from different platforms in the same dimension, providing convenience for the reuse and migration of cross-platform prompt words. In addition, the above standardization process makes the recognition and parsing method highly scalable. When a new platform is added, only the corresponding relationship needs to be added to the mapping table, without affecting existing data and upper-layer application logic, thus improving the applicability of the recognition and parsing method. In some embodiments, the above method S100 further includes: S40: storing the structured data in a database, and classifying the structured data according to preset rules based on one or more dimensions of platform identifier, prompt word purpose, and keywords extracted from the prompt word body text. As can be seen from the foregoing, after standardizing the structured data extracted by the parser, it can also be stored in a database, and the structured data can be automatically classified according to one or more dimensions of platform identifier, prompt word purpose, and keywords extracted from the prompt word body text according to preset rules. The database used to store structured data in a preset data format can be a local database (such as SQLite), a relational database (such as MySQL or PostgreSQL), or a cloud database service, to support subsequent retrieval, statistics, and analysis operations. Simultaneously, the structured data stored in the database is automatically categorized according to preset classification rules.The preset classification rules are a predefined set of classification logic or conditions used to automatically determine the category to which structured data belongs. Optionally, the classification rules can be based on one or more dimensions, such as platform identifiers, the purpose of prompt words, and keywords extracted from the main text of the prompt words. Classification processing refers to dividing the structured data into different categories according to the preset rules. The classification results can be used for the organization, retrieval, and recommendation of structured data. For example, classification by platform identifier means dividing the structured data according to the AI ​​platform from which it originates, such as classifying it into platforms like Midjourney, Stable Diffusion, and ComfyUI. Classifying structured data by platform identifier helps users quickly view and filter all prompt words from a specific platform, facilitating management and comparative analysis by platform. Classification by the purpose of prompt words means determining their application scenario based on the main text content of the prompt words, such as classifying them into categories like image generation, code generation, text analysis, and video generation. Categorizing by purpose helps users retrieve prompts with the same application goals across platforms, facilitating the migration and reuse of creative ideas across different platforms. Categorization by keywords extracted from the main text of the prompts involves extracting keywords from the main text field of the prompts in the structured data and categorizing the structured data based on these extracted keywords. For example, style or theme keywords such as "cyberpunk," "portrait," "landscape," and "anime" can be extracted from the main text of the prompts, thereby dividing the structured data into corresponding style tags or theme categories. Categorization by keywords helps users quickly locate prompts with the same visual style, theme, or creative elements, improving search efficiency and user experience. In some embodiments, categorization can also be based on negative prompts. Negative prompts are keywords specified by the user in the prompts to exclude content or styles that are not desired in the AI-generated results; they are commonly found in prompt formats on platforms such as Stable Diffusion. For example, prompts containing negative keywords such as "bad anatomy", "worst quality", "low quality", "blurry", and "extra limbs" can be categorized as "exclude low-quality content"; prompts containing negative keywords such as "NSFW", "nude", and "violence" can be categorized as "exclude inappropriate content".By classifying based on negative prompts, users can understand the quality control preferences or style exclusion strategies of each version snapshot and quickly filter out prompt versions that meet specific quality requirements. In some embodiments, one of the above classification dimensions is used alone; in other embodiments, multiple of the above classification dimensions are combined. For example, classification can be done first by platform identifier, and then further subdivided within the same platform category by purpose or keywords to form a multi-level classification system; or, positive and negative prompt keywords can be combined simultaneously for classification. Through automatic classification processing, the above method can classify stored prompts without manual organization by the user; keyword extraction and similarity association functions help users discover content-related prompts; and storing the classification results together with structured data provides a foundation for subsequent intelligent retrieval, version tracking, and recommendation systems. In some embodiments, based on the above classification processing, the identification and parsing method can also provide automatic tag suggestions to generate recommended tags for the structured data stored in the database, facilitating subsequent retrieval and classification. Understandably, automatic tag suggestion involves extracting keywords from the main text field of prompts in structured data, generating recommended tags, and storing these tags along with the structured data in a database. These tags are then displayed in the user interface as a tag cloud or list, allowing users to edit, delete, or add tags. This method reduces the time cost of compiling prompts and improves the usability of the data after it has been stored. Furthermore, tag-based classification and retrieval enable users to quickly locate prompts with the same visual style, theme, or purpose, enhancing the user experience. Based on this, when recognizing and parsing user-inputted prompts using the above method, as shown in Figure 2, the system first receives user-inputted prompts (supporting multiple methods such as manual input, browser extensions, clipboard, file import, and API); then, it combines a pre-stored feature signature library (containing regular expressions, keyword sets, and other feature rules for each platform) to perform platform feature matching on the prompts, and confirms the platform feature matching result based on the weighted confidence score and a preset score threshold; after identifying the platform corresponding to the prompt, it calls the platform's dedicated parser or preset parser according to the above matching result to extract structured data containing platform identifier, model version, main text, generation parameter key-value pairs, and original input; then, it standardizes the above structured data, converts it into a unified preset data format, and maps the parameters controlling the same generation feature HK 30135843 A 15 in different platforms to the same field; finally, it stores the standardized structured data in the database and automatically classifies it according to preset rules.Secondly, embodiments of this application also provide a multi-platform prompt word recognition and parsing system, including: at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method provided in any of the first aspects. In some embodiments, the above method S100 can be implemented by a multi-platform prompt word recognition and parsing device, as shown in FIG3, which specifically includes: a receiving module, a matching module, a parsing module, a standardization module, and a storage and classification module. The receiving module is used to receive prompt words input by the user. In some embodiments, the receiving module supports multiple input methods, including manual input, automatic capture by browser extensions, clipboard monitoring, file import, and application programming interface reception, etc. By adopting diverse input methods, the convenience and automation of data collection can be improved. The matching module is used to perform platform feature matching on the prompt words to obtain the matching result of the platform corresponding to the prompt words. Specifically, the matching module performs feature matching on the prompt words through a pre-stored feature signature library and identifies the platform corresponding to the prompt words based on the matching result. The feature signature library stores feature rules corresponding to each platform. Each feature rule contains multiple feature items, and each feature item has a preset weight value. The matching module can use regular expressions to match prompt words and / or use keyword sets to match prompt words, and calculate a weighted confidence score based on each feature item. If the weighted confidence score is higher than a preset score threshold, the platform corresponding to the weighted confidence score is taken as the matching result; if the weighted confidence score is lower than the preset score threshold, the matching module marks the platform as an unidentified platform for subsequent processing by a preset parser. The parsing module is used to parse the prompt words based on the matching results obtained by the matching module, calling the parser for the platform corresponding to the prompt words, and extracting the structured data corresponding to the prompt words. When the matching result is an identified platform, the parsing module calls the dedicated parser corresponding to that platform; when the matching result is an unidentified platform, the parsing module calls the preset parser and uses heuristic rules to extract the identifiable information within the prompt words. The parser extracts the main text, generated parameters, model version, and other information from the original prompts, forming structured data with fixed fields. The standardization module converts the structured data extracted by the parser into a preset data format to standardize the prompts. The preset data format includes the platform identifier, the platform's model version, the main text of the prompts, the names and corresponding values ​​of the generated parameters, and the prompts entered by the user.The storage and classification module stores standardized structured data into a database and classifies the structured data according to preset rules based on one or more dimensions, including platform identifier, the purpose of prompt words, and keywords extracted from the main text of the prompt words. Classification dimensions can be used individually or in combination to form a multi-level classification system. The storage and classification module can also automatically suggest tags, extracting keywords from the main text field of the prompt words to generate recommended tags, facilitating subsequent user retrieval and classification. The functions of the above modules can be implemented by the processor executing computer program instructions stored in the memory. The combination of the functions of each module can achieve the method described in any embodiment of the first aspect. Based on the above, this application provides two specific application embodiments: Embodiment 1: The user copies the complete prompt words returned by Midjourney Bot from Discord and inputs them into the system. The input content is: a cyberpunk warrior girl, intricate detailed armor, neon lights, rain --ar 16:9 --v 6.1 --s 750 --c 25 --style raw. After receiving the prompt words input by the user, the system first performs platform feature matching. The feature signature library pre-stores Midjourney's feature rules, including features for matching aspect ratio parameter patterns (e.g., --ar 16:9), version parameter patterns (e.g., --v 6.1), stylization parameter patterns (e.g., --s 750), disorder parameter patterns (e.g., --c 25), and the keyword "--style raw". The system matches all five features, calculates a weighted confidence score of 100 (the highest weighted confidence score is 100), which is higher than the preset threshold and also higher than the weighted confidence scores of other platforms. Therefore, the platform identification result for this prompt is the Midjourney platform. Subsequently, the system calls the Midjourney dedicated parser. The parser extracts all text before the first "--" as the main text, resulting in "a cyberpunk warrior girl, intricate detailed armor, neon lights, rain"; it extracts each "--" parameter and its value, resulting in {ar:"16:9", HK 30135843 A 17 v:"6.1", s:"750", c:"25", style:"raw"}; and determines the model version as 6.1 based on "--v 6.1". Next, the system unifies the structured data extracted by the parser into a standardized data structure.Specifically, the `ai_tool` field is assigned the value "midjourney"; the `model_version` field is assigned the value "6.1"; the `prompt_text` field is assigned the extracted main text "a cyberpunk warrior girl, intricate detailed armor, neon lights, rain"; the `parameters` field stores all extracted parameters in key-value pairs, i.e., {"ar":"16:9","v":"6.1","s":"750", "c":"25","style":"raw"}; and the `raw_input` field stores the complete raw string input by the user. All fields are converted according to a preset standard data format to ensure structural consistency of the parsing results across different platforms. Simultaneously, a standardized mapping is performed to convert functionally equivalent parameters from different platforms into unified semantic fields. In this embodiment, the specific mapping is as follows: the aspect ratio parameter of Midjourney—ar 16:9—is mapped to the general field aspect_ratio, whose value remains "16:9"; the stylization parameter—s 750—is mapped to the quality_level field, and 750 is categorized as HIGH (high-level stylization) according to the preset numerical range (0~1000); the style preset parameter—style raw—is mapped to the artistic_freedom field, where raw mode represents lower creative freedom, so it is assigned the value MINIMAL (minimal artistic intervention). Through the above standardization process, the prompts on the Midjourney platform are converted into a platform-independent unified representation. Finally, the system stores the structured data in the database, automatically categorizes it as Midjourney by platform, and classifies it as image generation by purpose (detecting keywords such as "cyberpunk" and "warrior"), automatically extracting the keywords "cyberpunk", "warrior", "armor", "neon", and "rain" as recommendation tags.In Example 2, after generating an image from the Stable Diffusion WebUI, the user copies the generation information from the metadata and inputs it into the system. The input includes three parts: positive prompts such as "masterpiece," "best quality," "1girl," "cyberpunk city," "neon signs," "detailed face," and "realistic"; negative prompts such as "bad anatomy," "worst quality," "low quality," "blurry," and "extra limbs"; and generation parameters: Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1847293, Size: 768x1024, and Model: SDXL 1.0. Subsequently, the system performs platform feature matching based on the received prompts using the feature rules of Stable Diffusion stored in the feature signature library. This includes keywords such as "Steps:," "CFG HK 30135843 A 18 scale:," "Seed:," "Sampler:," and "Model:," as well as feature items used to match the negative prompt separator pattern (i.e., the character content following "Negative prompt:"). The system detected all five keywords and the negative prompt word separator pattern, and calculated a weighted confidence score of 100 (the highest weighted confidence score is 100), which is higher than the preset score threshold (30) and higher than the weighted confidence scores of other platforms. Therefore, the platform identification result corresponding to this prompt word is the Stable Diffusion platform. Next, the system calls the Stable Diffusion dedicated parser to parse the received prompt word, using "Negative prompt:" as the separator, taking the part before the separator as the positive prompt word, and the part after the separator to the newline character as the negative prompt word, thus separating the positive and negative prompt words.Simultaneously, the generated parameters were extracted, and the number following "Steps:" was extracted as the sampling step number (30), the string following "Sampler:" and before the newline character was extracted as the sampler name (DPM++ 2M Karras), the number following "CFG scale:" was extracted as the cue word compliance (7), the number following "Seed:" was extracted as the random seed (1847293), the combination of the number, the letter x, and the number following "Size:" was extracted as the image size (768x1024), and the string following "Model:" and before the newline character was extracted as the model name (SDXL 1.0). The system then standardizes the structured data extracted by the parser, assigning `ai_tool` the value "stable_diffusion"; `model_version` the value "SDXL 1.0"; `prompt_text` the complete string of positive prompts; and `negative_prompt` the complete string of negative prompts. `parameters` stores all extracted parameters in key-value pairs, for example, `{"steps":30, "sampler":"DPM++ 2M Karras", "cfg":7, "seed":1847293, "size": "768x1024", "model":"SDXL 1.0"}`. `raw_input` stores the raw input string. Simultaneously, the system performs standardization mapping: calculating the aspect ratio of the 768x1024 size to 3:4 and storing it in the general field `aspect_ratio`; mapping the CFG scale value of 7 to `quality_level=STANDARD`. Finally, the system stores the standardized data in the database, categorizing it by platform as Stable Diffusion and by purpose (keywords "cyberpunk", "city", "neon", "realistic") as image generation. It automatically extracts the keywords "cyberpunk", "city", "neon", "realistic", and "portrait" from the prompt_text field as recommendation tags, and stores these recommendation tags along with the structured data in the database. HK 30135843 A 19 In summary, the technical solution of this application, through feature matching of user-input prompt words, can automatically identify and uniformly structure prompt words from multiple AI platforms, laying a technical foundation for cross-platform prompt word management.Secondly, by constructing a feature signature library and combining it with weighted confidence scores to confirm feature matching results, the accuracy of prompt word platform recognition can be improved, accurately identifying prompt word formats from different AI platforms (including Midjourney, Stable Diffusion, ComfyUI, etc.), effectively avoiding misidentification and missed identification. Furthermore, by mapping functional equivalence parameters of different platforms to a unified field, structured data with different structures can be compared, retrieved, and optimized on the same dimension, improving ease of use. In addition, the scalable feature library and plug-in parser architecture adopted in the above scheme can quickly adapt to newly emerging platforms, improving the applicability of the above method. Finally, the automatic classification and tag suggestion functions reduce the workload of manual organization by users, improving the management efficiency of prompt words and the user experience. It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of this embodiment. Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of this application as described above (HK 30135843 A 20). For the sake of brevity, they are not provided in the details; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.HK 30135843 A 1 Claims 1. A method for identifying and parsing prompt words across multiple platforms, characterized in that it includes: receiving prompt words input by a user; performing platform feature matching on the prompt words, and calling the parser of the platform corresponding to the prompt words to parse the prompt words according to the matching result, and extracting the structured data corresponding to the prompt words; converting the structured data into a preset data format to standardize the prompt words. 2. The method according to claim 1, characterized in that the step of performing platform feature matching on the prompt words includes: performing feature matching on the prompt words through a pre-stored feature signature library, and identifying the platform corresponding to the prompt words according to the matching result, wherein the feature signature library stores feature rules corresponding to each of the platforms. 3. The method according to claim 2, characterized in that the step of performing feature matching on the prompt words through a pre-stored feature signature library includes: matching the prompt words using regular expressions, and / or matching the prompt words using a keyword set. 4. The method according to claim 1, wherein converting the structured data into a preset data format includes: mapping parameters used to control the same generation features in each of the platforms to the same field. 5. The method according to claim 2, wherein each feature rule includes multiple feature items, each feature item has a preset weight value; and calling the parser of the platform corresponding to the prompt word to parse the prompt word according to the matching result includes: calculating a weighted confidence score based on each feature item; if the weighted confidence score is higher than a preset score threshold, then taking the platform corresponding to the weighted confidence score as the matching result, and calling the parser of the platform corresponding to the prompt word to parse the prompt word according to the matching result. 6. The method according to claim 5, further comprising: if the weighted confidence score is lower than a preset score threshold, then calling a preset parser to parse the prompt word, and marking the platform corresponding to the prompt word as an unidentified platform. 7. The method according to claim 1, wherein the preset data format includes the platform identifier, the platform model version, the main text of the prompt word, the name of the generation parameter and the corresponding parameter value, and the prompt word input by the user. 8. The method according to claim 1, wherein the input method for receiving the prompt word input by the user includes at least one of the following: manual input, automatic capture by browser extensions, clipboard detection, file import, and receiving via application programming interface.9. The method according to any one of claims 1-8, characterized in that it further comprises: storing the structured data in a database, and classifying the structured data according to one or more dimensions selected from the platform identifier, the purpose of the prompt word, and keywords extracted from the main text of the prompt word, based on preset rules. 10. A prompt word recognition and parsing system for multiple platforms, characterized in that it comprises: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-9. HK 30135843 A 1 The instruction manual diagram shows the following steps: Receive user-inputted prompt words; perform platform feature matching on the prompt words; and, based on the matching result, call the parser corresponding to the platform of the prompt words to parse the prompt words, extract the structured data corresponding to the prompt words, and convert the structured data into a preset data format for standardization processing of the prompt words. (S100 S10 S20 S30) Figure 1 HK 30135843 A 2 Receive user-inputted prompt words; perform platform feature matching on the prompt words; match successful; call the platform-specific parser to parse the prompt words; call the general parser to parse the prompt words; structured data extraction; cross-platform standardization; storage and automatic classification; (Figure 2 HK 30135843 A 3 Receive module; matching module; parsing module; standardization module; storage and classification module) Figure 3 HK 30135843 A.

Images