A method, system and medium for intelligent adaptation of a scenario-based workflow template

Abstract

The application provides a scenario-based workflow template intelligent adaptation method, system and medium, which is used to solve the problem of low accuracy of matching target workflow templates with actual business scenarios in traditional solutions. The method part includes: obtaining user inputted to-be-processed creation requirement data, the creation requirement data including at least one of requirement text and reference materials; analyzing the creation requirement data to obtain a target business scenario, a creation target and quantifiable generation parameters, and performing feature extraction on the reference materials to obtain a scenario matching feature vector; based on the scenario matching feature vector, performing multi-layer matching in a preset scenario-based template library to determine a target workflow template; based on the generation parameters and the reference material features obtained by the analysis, filling in the execution parameters of each node in the target workflow template; based on the creation target, adjusting the node link of the target workflow template, generating a scenario-based workflow and performing rendering display.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and provides a method, system, and medium for intelligent adaptation of scenario-based workflow templates. Background Technology

[0002] In the fields of artificial intelligence and content creation technology, workflow templates are important tools for improving content production efficiency. However, most workflow templates provided by existing content creation platforms are general-purpose, fixed-topology templates. In practical applications, these general-purpose templates typically only provide basic node links. The system cannot intelligently identify specific business scenarios and creative goals based on user-input natural language descriptions or uploaded reference materials, nor can it dynamically adjust the workflow links according to actual needs. Therefore, users not only find it difficult to accurately match templates suitable for their own business scenarios, but also need to manually configure complex node execution parameters and modify workflow link logic, resulting in poor workflow scenario adaptability and raising the usage threshold for non-technical personnel. Summary of the Invention

[0003] This application provides a method, system, and medium for intelligent adaptation of scenario-based workflow templates, which solves the technical problem of low accuracy in matching target workflow templates with actual business scenarios in traditional solutions.

[0004] To solve the above problems, the following technical solution is provided: A method for intelligent adaptation of scenario-based workflow templates, the method includes: Obtain user-inputted creative requirement data, which includes at least one of requirement text and reference materials; The creation requirement data is analyzed to obtain the target business scenario, creation goal and quantifiable generation parameters, and the reference material features are extracted to construct the scene matching feature vector. Based on scene matching feature vectors, multi-level matching is performed in a pre-set scene-based template library to determine the target workflow template; Based on the generated parameters obtained from the analysis and the features of the reference materials, the execution parameters of each node in the target workflow template are filled in; Based on the creation goals, the node links of the target workflow template are added, removed, or adjusted to generate a scenario-based workflow and render it for display.

[0005] In one implementation, the creation requirement data is parsed to obtain the target business scenario, creation goals, and quantifiable generation parameters, including: The large language model is invoked to parse the requirement text, identify industry tags and core business scenarios in the requirement text, and use them as target business scenarios. Analyze the intent of the requirements in the requirement text, identify the target content type and visual style description, and use them as creative goals; Extract the number of generated units, aspect ratio, resolution, and target publishing platform information from the requirements text as quantifiable generation parameters.

[0006] In one implementation, feature extraction is performed on the reference material to obtain reference material features, including: Perform multimodal understanding on the reference materials to extract visual element features, composition layout features, main color tone features, and copywriting selling points from the reference materials; The extracted features are transformed into feature vectors to construct reference material features.

[0007] In one implementation, based on scene matching feature vectors, multi-level matching is performed in a pre-defined scene-based template library to determine the target workflow template, including: Industry tags, content types, and applicable platforms are extracted from the scene matching feature vectors, and candidate template sets are selected from the scene-based template library; For each template in the candidate template set, calculate the vector similarity between the vector representation of the requirement text and the corresponding functional description and scene tag of the template, and select the first preset number of templates as the selected template set based on the vector similarity. Obtain the node link complexity, historical success rate, and user rating weight for each template in the selected template set. Sort the selected template set by requirement matching degree and determine the template with the first place in the ranking as the target workflow template.

[0008] In one implementation, based on the parsed generation parameters and reference material features, the execution parameters of each node in the target workflow template are populated, including: Configure the output specification parameters of the generated nodes in the target workflow template based on the scale and resolution in the generation parameters. Based on the target business scenario, the corresponding pre-trained generative model is matched in the model mapping library, and the matched pre-trained generative model is loaded into the generation node; Based on the features of the reference materials, optimized prompts are generated, and the optimized prompts are used to replace the default prompts in the prompt nodes of the target workflow template.

[0009] In one implementation, the node links of the target workflow template are added to, removed from, or adjusted based on the creation goal, including: If an audio generation requirement is identified in the creation objective, then an audio generation node and a subtitle overlay node are inserted into the execution chain of the target workflow template; If a batch generation requirement is identified in the generation parameters, a batch export node will be added to the end of the target workflow template.

[0010] In one implementation, after adding a batch export node at the end of the target workflow template's link, it also includes: Identify variable parameters in contextualized workflows; Based on the variable parameters, a corresponding batch data import template is generated so that users can fill in task information in batches using the batch data import template.

[0011] In one implementation, after rendering and displaying, the following steps are also included: In response to the execution command, a scenario-based workflow is executed to generate the target content; Perform quality scoring and compliance testing on the target content; Record the parameter configuration and user feedback data during the execution process, and perform weight iteration on the template parameters and node links in the scenario-based template library based on the recorded data.

[0012] A scenario-based workflow template intelligent adaptation system, the system includes: The scenario-based template library module is used to store a preset template system for core business scenarios; The requirement parsing module is used to parse requirement text and / or reference materials to obtain the target business scenario, creation goal, generation parameters and reference material characteristics; The template matching module is used to perform multi-layer matching logic in the scenario-based template library module based on scenario matching feature vectors to determine the target workflow template; The adaptation and optimization module is used to fill the execution parameters of the target workflow template based on the parsed parameters, and adjust the node links of the target workflow template according to the creation goal, so as to output the scenario-based workflow to be executed.

[0013] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the above-described method.

[0014] The solution provided in this application acquires and analyzes creative demand data to construct scene matching feature vectors. Multi-level matching is performed in a scene-based template library to determine the target workflow template. Based on the analysis results, execution parameters are automatically filled in, and node links are added or removed and adjusted. Multi-modal understanding technology is used to transform unstructured natural language and materials into precise quantitative feature vectors and perform multi-level intelligent retrieval in the feature space. Simultaneously, automated variable injection and graph topology reconstruction technologies are used to achieve dynamic customization of the workflow. This significantly improves the accuracy of matching the target workflow template with actual business scenarios, avoids repeated manual trial and error due to inapplicable template links, and greatly reduces the operational costs of manually adjusting complex parameters and modifying node connection logic. This significantly lowers the barrier to entry for AI creation systems and improves the flexibility and automated production efficiency of content workflows in various complex scenarios. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a scenario-based workflow template intelligent adaptation method according to an embodiment of this application; Figure 2 This is a schematic diagram of a scenario-based workflow template intelligent adaptation system according to one embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device according to one embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] like Figure 1 As shown in the illustration, this application provides a method for intelligent adaptation of scenario-based workflow templates, which can be applied to content production services on AI creation platforms. This is merely an example and not a specific limitation; the method is primarily executed by various intelligent processing modules in the system backend.

[0019] In one embodiment, the method includes the following steps: S101. Obtain the user-inputted creation requirement data, which includes at least one of the requirement text and reference materials. S102. Analyze the creation requirement data to obtain the target business scenario, creation goal and quantifiable generation parameters, and extract the features of the reference materials to construct the scene matching feature vector. S103. Based on the scene matching feature vector, perform multi-level matching in the preset scene-based template library to determine the target workflow template; S104. Based on the generated parameters obtained from the analysis and the features of the reference materials, fill in the execution parameters of each node in the target workflow template; S105. Based on the creation goal, adjust the node links of the target workflow template by adding or removing nodes, generate a scenario-based workflow, and render and display it.

[0020] In the initial stage of workflow adaptation, the system needs to perceive and acquire the user's creative intent. During this process, step S101 is executed to acquire the user's input data on the creative requirements to be processed. This data serves as the input basis for subsequent intelligent processing by the system, representing an information carrier that embodies the user's specific content production expectations and constraints. Regarding acquisition methods, users can directly input their requirements in natural language form as trigger conditions through the system's interactive interface, for example, inputting "Help me generate images of women's clothing e-commerce products, 3:4 aspect ratio, for display." As an example, users can also upload reference materials, which specifically refer to unstructured data files that reflect the desired visual style, sound characteristics, or content structure, such as reference images, reference videos, or script documents. The system preferably supports multiple interaction methods, including only text input, only material upload, or a combination of both, thereby greatly facilitating users lacking technical backgrounds to quickly initiate creative tasks in different business scenarios such as e-commerce product image generation, short video production, or enterprise new media marketing.

[0021] After acquiring the creation request data to be processed, step S102 is executed, where the system performs deep analysis to structure the request data. Specifically, for the request text, the system uses the parsing module to extract the target business scenario, i.e., the user's industry or specific application environment, such as a cross-border e-commerce scenario or a corporate brand marketing scenario; it also extracts the creation goal, i.e., the desired content type and visual or acoustic style effect; and extracts quantifiable generation parameters, which are specific values ​​that can be used as mathematical or logical control variables in subsequent execution stages, such as the aspect ratio representing the screen size, the resolution value determining image clarity, video length, specific applicable platforms, format requirements, and multilingual requirements. In addition, if reference materials are obtained, the system performs multi-dimensional feature understanding, extracting the main color tone, composition layout features, or semantic selling points contained in the reference materials. Subsequently, the system performs mathematical-level dimensional integration and high-dimensional encoding of the scenarios, goals, parameters, and features obtained above, constructing a scene matching feature vector that integrates multi-source information. This feature vector is essentially a precise mathematical expression of the user's vague creative intent, and its role is to provide a reliable quantitative benchmark for subsequent similarity search and matching in the implicit feature space.

[0022] Next, in step S103, the system performs multi-level matching based on the constructed scene matching feature vector within a pre-defined scene-based template library. This scene-based template library is a data center storing a large number of workflow templates deeply optimized for various core business scenarios. For example, the library includes pre-defined scene templates such as e-commerce product image templates covering everything from product copywriting generation to intelligent image retouching, short video production scene templates adapted to various social media platform formats, AI animation creation scene templates, and enterprise new media marketing scene templates. Furthermore, the library preferably supports tenant-customized storage, allowing enterprises to save mature internal workflows as private scene templates to accumulate industry best practices. The core principle of multi-level matching is that the system does not use a single keyword search, but rather employs a multi-level filtering and semantic similarity calculation mechanism to comprehensively measure the fit between the feature vector and each template in the library, thereby accurately determining the target workflow template that best meets the current user's needs from a massive candidate set. The specific matching levels and algorithm feature dimensions are not limited; this matching mechanism effectively solves the problems of low efficiency and the high risk of selecting the wrong business link when manually searching for templates one by one.

[0023] Existing fixed templates often require users to manually adjust configurations at each node. In this application, however, in step S104, the system directly fills the corresponding numerical constraints and style feature variables into the execution parameters of each processing node in the target workflow template based on the previously parsed generation parameters and reference material features, completing the automated replacement of prompts, model specifications, brand specification parameters, etc. Furthermore, due to the ever-changing personalized needs of users, the fixed directed acyclic graph execution flow of the template may have functional redundancy or deficiencies. Therefore, in step S105, the system intelligently infers the necessary procedures for the current task based on the parsed creative goal, automatically determining whether certain nodes need to be skipped or specific processing nodes need to be added to the link, thereby adjusting the node links of the target workflow template. For example, in an e-commerce product image generation task, if it is determined that the user does not need the image retouching step, the system automatically skips the intelligent image retouching node in the link; in the scenario of comics or short video series, the system can also automatically retain or add character and scene consistency control nodes, and automatically embed brand specification and compliance verification nodes in the enterprise brand compliance scenario. After completing parameter filling and link adjustment, the system renders and displays the optimized scenario-based workflow in the front-end visualization canvas. At this point, the logical connections between nodes and internal parameters have been configured, and users can directly view and execute the generation task with one click.

[0024] In this embodiment, creation requirement data is acquired and parsed to construct scene matching feature vectors. Multi-level matching is performed in a scene-based template library to determine the target workflow template. Based on the parsing results, execution parameters are automatically filled in and node links are added or removed and adjusted. Multimodal understanding technology is used to transform unstructured natural language and materials into precise quantitative feature vectors and perform multi-level intelligent retrieval in the feature space. At the same time, automated variable injection and graph topology reconstruction technology are used to achieve dynamic customization of the workflow, which significantly improves the accuracy of matching the target workflow template with the actual business scenario, avoids repeated manual trial and error due to inapplicable template links, and greatly reduces the operation cost of users manually adjusting complex parameters and modifying node connection logic. This greatly lowers the usage threshold of the AI ​​creation system and improves the flexibility and automated production efficiency of content workflow in various complex scenarios.

[0025] In one embodiment, step S102, which involves parsing the creation requirement data to obtain the target business scenario, creation goal, and quantifiable generation parameters, specifically includes the following steps: S201. Call the large language model to parse the requirement text, identify the industry tags and core business scenarios in the requirement text, and use them as the target business scenarios. S202. Analyze the intent of the requirements in the requirements text, identify the target content type and visual style description, and use them as the creative goal; S203. Extract the generation quantity, aspect ratio, resolution, and target publishing platform information from the requirement text as quantifiable generation parameters.

[0026] In practice, when users input requirements in text format, natural language text is unstructured data and is generally not suitable for direct use in controlling workflow topology or configuring API parameters. Therefore, the system needs to parse and structure the requirement text.

[0027] First, the system executes step S201, calling a preset large language model to parse the request text. Utilizing the natural language processing capabilities of the large language model, it identifies industry tags and core business scenarios within the request text, using these as the target business scenario to clarify the business domain to which the user's content creation pertains. In practical applications, step S201 allows the system to identify and distinguish the specific business scenario the user is in, such as "e-commerce product image scenario," "short video production scenario," "AI comic creation scenario," or "enterprise new media marketing scenario," etc.

[0028] Next, the system executes step S202, parsing the requirement intent in the requirement text using a large language model to identify the target content type and visual style description, which are then used as the creation target. This step mainly extracts descriptions of the output content form and visual characteristics from the text, identifying content types such as posters, main images, storyboards, and video frames, as well as visual style descriptions such as realistic style, Instagram style, traditional Chinese style, and cyberpunk style, providing a basis for the dynamic addition, removal, or replacement of processing nodes in the subsequent workflow topology.

[0029] Meanwhile, to obtain specific quantitative control indicators, the system executes step S203 in parallel or sequentially, extracting data such as the number of units generated, aspect ratio, resolution, and target publishing platform information from the requirement text, and using these as quantifiable generation parameters. These parameters form structured control variables, which can be directly injected into the corresponding execution nodes in subsequent processes, replacing the default values ​​in the original workflow template.

[0030] To illustrate the above analysis process more clearly, the following example uses a specific application scenario: Suppose the user's input text is: "Help me generate images of women's clothing e-commerce products, 3:4 aspect ratio, Instagram style, for use on Taobao product detail pages."

[0031] The system breaks down the text segment by executing steps S201 to S203 as described above: In step S201, the large language model identifies the industry tag as "women's clothing" and the core business scenario as "e-commerce product image scenario", thereby determining the target business scenario; In step S202, the large language model parses the user's intent to create product display content, identifies the target content type as "product image" and the visual style as "Instagram style", thereby establishing the creation goal; In step S203, the large language model extracts "3:4" from the text as the screen ratio, extracts "Taobao details page" as the target publishing platform information, and can automatically associate and match the corresponding resolution value based on the platform's corresponding specification configuration, thereby obtaining quantifiable generation parameters.

[0032] Furthermore, in business scenarios with multilingual environments, such as cross-border e-commerce, the parsing process of the aforementioned large language model can also support the parsing of requirement texts in different languages. The system can identify requirement texts in languages ​​other than its native language, extract the specific format requirements and compliance rules of the corresponding cross-border platform, and convert them into structured data to populate quantifiable generation parameters, so that the content generated by subsequent workflows conforms to the business specifications of the corresponding country or region.

[0033] In this embodiment, unstructured requirement text is decoupled and transformed into structured parameters consisting of the target business scenario, creation goal, and quantifiable generation parameters. The benefits are: it reduces the reliance on users manually selecting or filling in complex forms during workflow configuration, improving human-computer interaction efficiency; simultaneously, by decomposing the text into clearly defined business scenarios, creation goals, and multi-dimensional quantifiable parameters, it provides aligned feature data for subsequent template matching and corresponding parameter sources for adaptive filling of node parameters, thus improving the overall executability and operational stability of intelligent workflow adaptation.

[0034] In one embodiment, step S102, which involves extracting features from the reference material to obtain reference material features, specifically includes the following steps: S301. Perform multimodal understanding on the reference material to extract visual element features, composition layout features, main color tone features, and copywriting selling point information from the reference material; S302. Convert the extracted features into feature vectors to construct the reference material features.

[0035] After acquiring unstructured reference materials such as reference images, reference videos, or script documents uploaded by users, the system needs to deeply mine the rich creative information they implicitly contain. Step S301 involves the system calling a preset computer vision model or cross-modal analysis model to perform multimodal understanding on the reference materials. Specifically, multimodal understanding refers to the process of parallel or serial analysis of data streams of different modalities, such as image pixels, video time frames, and text. The system scans the pixel distribution and temporal changes of the materials, identifies and extracts specific content such as the main objects and background decorations contained in the image, and uses these as visual element features. Simultaneously, the system analyzes the spatial relationship between the foreground and background and the camera angle in the image to obtain compositional layout features; and uses algorithms such as color clustering to extract the dominant color distribution that determines the overall visual atmosphere of the image, as the main color tone feature of the image. Furthermore, for materials containing text, such as promotional posters with accompanying text or plain text script documents, the system uses optical character recognition or natural language processing technology to extract text content with marketing orientation or core expressive meaning, using it as copywriting selling point information.

[0036] After completing the multi-dimensional feature extraction described above, the extracted information still belongs to heterogeneous data and is difficult to directly participate in the calculation and matching of the underlying algorithm. Therefore, in step S302, the system sends the extracted features into a preset feature encoding layer and performs feature vector transformation through dimensionality reduction and mapping algorithms such as matrix multiplication. This feature vector transformation process aims to map high-dimensional image pixel features and discrete text symbols into a unified, computer-readable, continuous, dense vector space, thereby constructing reference material features that characterize the comprehensive physical and semantic attributes of the material. As a high-dimensional mathematical expression, this reference material feature can be directly concatenated or fused with the aforementioned required text parsing results for use in subsequent template matching and automated parameter filling stages.

[0037] By performing multimodal understanding on the reference materials, visual element features, composition layout features, main color tone features, and copywriting selling points are extracted. The extracted features are then transformed into feature vectors to construct the reference material features. The multimodal analysis model is used to reduce the dimensionality of heterogeneous multimedia data and project it into a unified implicit feature space. This makes up for the limitations of single natural language text input in expressing visual details. The system can accurately capture the intuitive visual style and screen structure that users cannot accurately describe in words, enriching the information dimensions of requirement analysis and thus improving the accuracy of subsequent workflow template matching and automatic parameter adaptation.

[0038] In one embodiment, step S103, which involves performing multi-level matching based on scene matching feature vectors in a preset scene-based template library to determine the target workflow template, specifically includes the following steps: S401. Extract industry tags, content types and applicable platforms from the scene matching feature vector, and filter out candidate template sets from the scene-based template library; S402. For each template in the candidate template set, calculate the vector similarity between the vector representation of the requirement text and the corresponding functional description and scene tag of the template, and select the first preset number of templates as the selected template set based on the vector similarity. S403. Obtain the node link complexity, historical success rate and user rating weight of each template in the selected template set, sort the selected template set according to the degree of requirement matching, and determine the template with the first position in the sort as the target workflow template.

[0039] To improve matching efficiency and ensure accuracy when searching within the scenario-based template library, the system employs a hierarchical matching mechanism to process feature data. First, in step S401, the system extracts explicit classification indicators from the scenario matching feature vector, namely, industry tags, content type, and applicable platform. The system then uses these explicit classification indicators to perform a preliminary comparison within the scenario-based template library, filtering out templates with mismatched attributes. This results in a smaller set of candidate templates, narrowing the data range for subsequent similarity calculations.

[0040] Next, the system performs semantic-level filtering on the candidate template set. For each template in the candidate template set, the system calculates the vector similarity between the vector representation of the requirement text and the pre-configured functional description and scene tags of that template. This vector similarity quantifies the degree of similarity between the user's requirement description and the template's preset functions in the feature space. The system sorts the calculated vector similarities in descending order and selects the top preset number of templates with the highest similarity values, defining them as the curated template set. As an example, this preset number can be set to the top five (i.e., Top 5) to provide an appropriate number of highly relevant alternatives.

[0041] After semantic filtering, the system performs a final ranking of the selected template set based on multi-dimensional operational metrics. The system obtains the node link complexity, historical success rate, and user rating weight for each template in the selected template set. Node link complexity reflects the topology scale and computational resource consumption of the template during execution; historical success rate characterizes the reliability of the template in historical generation tasks; and user rating weight represents the objective evaluation feedback from historical users regarding the template's content generation effect. The system performs a comprehensive weighted calculation of these operational metrics to rank the selected template set according to their demand matching degree. The system determines the template with the highest comprehensive evaluation score and ranked first as the target workflow template. As an example, while outputting the target workflow template, the system can also output 2 to 3 templates ranked next in the order as alternative templates.

[0042] After determining the target workflow template and alternative templates, the system will display matching template cards on the interactive front end. The displayed information includes the template name, scenario adaptation instructions, core links, and historical usage count. Users can preview the specific workflow structure and default parameters of the template through these cards to further verify and select the appropriate template.

[0043] By extracting basic tags from scene matching feature vectors to filter candidate template sets, using vector similarity to filter and refine the template set, and combining node link complexity, historical success rate and user rating weights to finally rank the requirements matching degree to determine the target workflow template, the hierarchical and progressive retrieval logic and the introduction of historical operation feedback data as evaluation parameters effectively reduces the computational resource consumption required for full-database feature comparison of templates, while taking into account the accuracy of semantic understanding and the reliability of business execution, and improving the applicability of the determined target workflow template in real application scenarios.

[0044] In one embodiment, step S104, which involves filling in the execution parameters of each node in the target workflow template based on the parsed generation parameters and reference material features, specifically includes the following steps: S501. Based on the scale and resolution in the generation parameters, configure the output specification parameters of the generated nodes in the target workflow template; S502. Based on the target business scenario, match the corresponding pre-trained generation model in the model mapping library and load the matched pre-trained generation model into the generation node; S503. Based on the features of the reference material, generate optimized prompt words and replace the default prompt words of the prompt word nodes in the target workflow template with the optimized prompt words.

[0045] After determining the target workflow template, the system needs to configure the parameters of each processing node within the template to adapt it to the specific task requirements of the current user.

[0046] First, the system extracts the generation parameters obtained from the aforementioned analysis, mainly involving the aspect ratio (e.g., 16:9 or 3:4) and resolution data. The system inputs these values ​​into the generation node corresponding to the target workflow template, thereby configuring the output specification parameters of the generation node in the target workflow template. These output specification parameters are used to control the size and clarity of the output content at the algorithm's underlying level to meet the requirements of specific social media or content publishing platforms.

[0047] Because different business scenarios have different preferences for the capabilities of generative models, the system internally establishes and maintains a model mapping library to record the mapping relationships between various business scenarios and adapted models. Based on the currently identified target business scenario, the system retrieves and matches the corresponding pre-trained generative model from the model mapping library. For example, in an e-commerce product image scenario, the system matches a high-resolution raw image model with detailed characterization capabilities; in a short video production scenario, the system matches a generative model with lip-syncing or digital human-driven capabilities. After matching, the system loads the network weights or service call interfaces of the matched pre-trained generative model into the generation node to determine the generation control logic of that node.

[0048] Subsequently, in step S503, the system processes the semantic control conditions of the generated content. Based on the features of the reference materials extracted in the preceding steps (including but not limited to visual element features, main color features of the image, and copywriting selling points), the system uses a natural language generation algorithm to convert them into text descriptions that conform to the input specifications of the pre-trained generation model, thereby generating optimized prompt words. The system then uses the optimized prompt words to replace the default prompt words preset in the prompt word node of the target workflow template. As a supplement, during this parameter filling process, the system can also obtain the brand specifications pre-entered by the tenant (such as preset brand logo, watermark, font, color scheme specifications, etc.) and fill them into the execution parameters of the corresponding node to adapt to the enterprise's brand compliance requirements.

[0049] By configuring output specification parameters based on the scale and resolution in the generation parameters, matching the corresponding pre-trained generation model based on the target business scenario, and generating optimized prompt words based on reference material features to replace the default prompt words, the structured parsing results are automatically distributed to each underlying execution node of the workflow template using parameter mapping and instruction conversion mechanisms. This reduces the tedious steps of manually modifying screen resolution, selecting underlying models, and repeatedly adjusting prompt words, lowers the generation error rate caused by human configuration errors, and improves the efficiency of workflow template parameter configuration and the fit of the final generated content with the actual business scenario.

[0050] In one embodiment, step S105, which involves adding or removing nodes in the target workflow template based on the creation goal, specifically includes the following steps: S601. If an audio generation requirement is identified in the creation target, then insert an audio generation node and a subtitle overlay node into the execution chain of the target workflow template. S602. If a batch generation requirement is identified in the generation parameters, a batch export node is added to the end of the target workflow template link.

[0051] At its core, a workflow template is a directed acyclic graph (DAG) consisting of a series of processing nodes with sequential data dependencies, where each node represents an independent data processing or model inference task. Because user needs are diverse in real-world applications, the fixed topology of pre-set templates is insufficient to adapt to all creation tasks. Therefore, the system introduces a dynamic link reconfiguration mechanism to intelligently customize the workflow.

[0052] In practice, the system analyzes the creative goals extracted from the aforementioned requirements analysis stage in real time. When audio generation requirements such as dubbing and narration are identified in the creative goals in scenarios such as short video production or comic book creation, the system automatically calls the graph topology editing algorithm to adjust the original template links. Specifically, the system logically cuts off the corresponding video compositing connection in the execution link of the target workflow template, which may originally only contain visual image or silent video generation, and inserts audio generation nodes and subtitle overlay nodes as new nodes into the link. The audio generation node is used to convert the associated text or script data into corresponding audio waveform data, and the subtitle overlay node is used to render text characters onto the video visual frames according to timestamps. The system then re-establishes the data flow channel between these two newly inserted nodes and the nodes before and after the original link.

[0053] The system performs conditional checks on the extracted generation parameters. If a batch generation requirement for producing a large amount of similar content at once is identified in the generation parameters, the system will automatically append a batch export node to the end of the node chain in the target workflow template. This batch export node is used to uniformly package and format the processing results generated in parallel or cyclically to support subsequent batch downloads or data delivery.

[0054] As an example, and without specifying a particular approach, this dynamic adjustment mechanism for the workflow, in addition to adding nodes as described above, also supports skipping nodes or supplementing functional modules based on the actual task requirements. For instance, in an e-commerce product image generation task, if the system determines that the user does not need subsequent image retouching, it automatically reconnects the "background removal" and "export" nodes in the underlying directed acyclic graph, thus skipping the original template's built-in "intelligent image retouching" node in the actual execution workflow. When handling the needs of a series of short videos or comic series, the system can automatically retain or add character and scene consistency control nodes in the workflow to ensure a consistent style across the series. Furthermore, in enterprise new media marketing scenarios, the system can automatically embed brand specification nodes and compliance verification nodes before the content output end to ensure that the generated products comply with the enterprise's brand specifications and advertising compliance requirements.

[0055] By inserting audio generation nodes and subtitle overlay nodes when audio generation needs are identified, and adding batch export nodes at the end of the link when batch generation needs are identified, the system utilizes the dynamic graph editing of the directed acyclic graph topology and the automatic reconnection mechanism of node data flow. Based on specific contextual requirements, the system intelligently infers necessary procedures, enabling the originally static workflow template to have the ability to adaptively expand and simplify the topology reconstruction on demand. This effectively reduces the cumbersome process of users manually dragging and dropping nodes and configuring port connections in the operation canvas, reduces the risk of data flow logic errors caused by improper manual connections, and improves the system's business compatibility with complex and long-tail content creation needs.

[0056] In one embodiment, after adding a batch export node to the end of the target workflow template link, the following steps are also included: S701, Identify variable parameters in scenario-based workflows; S702. Generate corresponding batch data import templates based on variable parameters, so that users can fill in task information in batches through the batch data import templates.

[0057] For enterprise-level industrial content production scenarios, single-processing mechanisms often fall short of meeting the demands of high-concurrency business. By adding batch export nodes to the workflow and forming a complete scenario-based workflow, the system enters the batch generation and adaptation phase. Specifically, by traversing all execution nodes in this scenario-based workflow, the system automatically extracts and identifies variable parameters set as external dynamic inputs. These variable parameters refer to placeholders or information fields that need to be frequently replaced in multiple batch generation tasks. As an example, in e-commerce or short video generation scenarios, variable parameters might include product information, marketing copy, and relevant material links that require user input.

[0058] Based on the extracted variable parameters' data structure and field attributes, the system automatically compiles and generates corresponding formatted files, i.e., batch data import templates. As an example, this batch data import template is presented as an automatically generated spreadsheet (e.g., Excel format) import file. The system presents this batch data import template to users, allowing them to batch fill in hundreds or even more corresponding task information entries. After completing and uploading the form, users can submit the batch-generated tasks with a single click through the interactive interface. Preferably, users can also publish the optimized, scenario-based workflow as a shared application within the tenant with a single click, enabling other team members to directly use the workflow solution and corresponding batch data import template without repetitive node or parameter configuration, thus achieving effective reuse of internal production solutions.

[0059] By identifying variable parameters in the scenario-based workflow and generating corresponding batch data import templates based on these parameters for users to fill in task information in batches, the system utilizes parameter variable extraction and formatted form automation construction mechanisms to structurally decompose the data input process and the underlying graph execution logic. This avoids the tedious steps of users repeatedly configuring node parameters for each independent generation task, significantly improving the system's task flow efficiency and solution reusability in an enterprise-level industrialized batch content production environment.

[0060] In one embodiment, after rendering and displaying, the following steps are further included: S801, responding to the run command, executes a scenario-based workflow to generate target content; S802. Conduct quality scoring and compliance testing on the target content; S803: Record the parameter configuration and user feedback data during the execution process, and perform weight iteration on the template parameters and node links in the scenario template library based on the recorded data.

[0061] After the scenario-based workflow completes the front-end rendering and node parameter adaptation, the system enters the task execution and feedback optimization phase. First, the system receives and responds to the execution command triggered by the user interface. Based on the aforementioned dynamically adjusted and configured directed acyclic graph link, the underlying computing engine sequentially schedules computing resources to execute the corresponding workflow nodes, thereby executing the scenario-based workflow and generating corresponding multimodal target content such as images, short videos, or series of videos.

[0062] The system calls the built-in quality assessment model to perform multi-dimensional objective and subjective quality scoring on the target content in order to select high-quality results that meet business expectations. At the same time, the system uses compliance verification rules or nodes to detect whether the target content contains sensitive information, whether it complies with preset corporate brand specifications and advertising rules, in order to achieve compliance detection.

[0063] The system continuously records parameter configuration details (including default parameter modification records) and runtime status throughout the global execution process in the background, and collects user feedback data such as the actual adoption rate and user satisfaction rate of the generated results. Based on this long-term recorded data, the system uses a weighted update algorithm to iterate the weights of the default template parameter values ​​and the importance characteristics of node links in the scenario-based template library, thereby optimizing the template's adaptability. As an example, the system also supports the accumulation of high-scoring, high-quality workflows in the same industry and scenario into the industry best practice template library, and labels their applicable industries and performance data; it also supports tenant administrators to mark efficient workflows within their tenants as enterprise best practice templates and distribute them internally to unify internal content production standards.

[0064] By generating target content in response to execution commands, performing quality scoring and compliance checks on the target content, and recording parameters and feedback data to iterate the weight of the scenario-based template library, a dynamic evolution feedback channel for templates is established using a closed-loop data collection and adaptive weight update mechanism. This not only ensures the output quality and business compliance of the current content production, but also continuously accumulates and promotes best practices in the industry and enterprises, enabling the template library to be continuously optimized as usage data accumulates, thereby improving the accuracy of template matching and the effectiveness of parameter presets in subsequent tasks.

[0065] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0066] In one embodiment, such as Figure 2 As shown, a scenario-based workflow template intelligent adaptation system is also provided, the system including: The scenario-based template library module 101 is used to store a preset template system for core business scenarios; The requirement analysis module 102 is used to analyze the requirement text and / or reference materials to obtain the target business scenario, creation goal, generation parameters and reference material characteristics. The template matching module 103 is used to perform multi-layer matching logic in the scenario-based template library module based on the scenario matching feature vector to determine the target workflow template; The adaptation and optimization module 104 is used to fill the execution parameters of the target workflow template based on the parsed parameters, and adjust the node links of the target workflow template according to the creation goal, so as to output the scenario-based workflow to be executed.

[0067] For details on the implementation of this scenario-based workflow template intelligent adaptation system, please refer to the steps of a scenario-based workflow template intelligent adaptation method, which will not be repeated here.

[0068] In one embodiment, such as Figure 3 As shown, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a scenario-based workflow template intelligent adaptation method as described in any of the preceding claims.

[0069] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of intelligent adaptation of a scenario-based workflow template as described above.

[0070] The details regarding the equipment and media can be found in the description of the aforementioned method embodiments, and will not be repeated here.

[0071] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the methods described above. Any references to memory, storage, user databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory.

[0072] Those skilled in the art will understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0073] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for intelligent adaptation of scenario-based workflow templates, characterized in that, The method includes: Obtain user-inputted creation request data, wherein the creation request data includes at least one of request text and reference materials; The creation requirement data is analyzed to obtain the target business scenario, creation goal and quantifiable generation parameters, and the reference material features are extracted to construct a scene matching feature vector. Based on the scene matching feature vector, multi-level matching is performed in a preset scene-based template library to determine the target workflow template; Based on the generated parameters obtained from the analysis and the features of the reference materials, the execution parameters of each node in the target workflow template are filled in; Based on the creation goal, the node links of the target workflow template are added, removed, or adjusted to generate a scenario-based workflow and render it.

2. The method according to claim 1, characterized in that, The process of parsing the creation requirement data yields the target business scenario, creation objective, and quantifiable generation parameters, including: The large language model is invoked to parse the requirement text, identify the industry tags and core business scenarios in the requirement text, and use them as the target business scenario. The intention of the requirement text is analyzed, and the target content type and visual style description are identified as the creative target. The quantity, aspect ratio, resolution, and target publishing platform information are extracted from the requirement text and used as the quantifiable generation parameters.

3. The method according to claim 1, characterized in that, The step of extracting features from the reference material to obtain reference material features includes: Perform multimodal understanding on the reference materials to extract visual element features, composition layout features, main color tone features, and copywriting selling points from the reference materials; The extracted features are transformed into feature vectors to construct the reference material features.

4. The method according to claim 1, characterized in that, The step of determining the target workflow template by performing multi-level matching in a preset scenario-based template library based on the scenario matching feature vector includes: Industry tags, content types, and applicable platforms are extracted from the scene matching feature vector, and a candidate template set is selected from the scene-based template library. For each template in the candidate template set, the vector similarity between the vector representation of the requirement text and the corresponding functional description and scene tag is calculated, and a preset number of templates are selected as the selected template set based on the vector similarity. Obtain the node link complexity, historical success rate, and user rating weight for each template in the selected template set, sort the selected template set by demand matching degree, and determine the template with the highest ranking as the target workflow template.

5. The method according to claim 1, characterized in that, The process of filling in the execution parameters of each node in the target workflow template based on the generated parameters obtained from the parsing and the features of the reference materials includes: Based on the scale and resolution in the generation parameters, configure the output specification parameters of the generated nodes in the target workflow template; Based on the target business scenario, a corresponding pre-trained generative model is matched in the model mapping library, and the matched pre-trained generative model is loaded into the generative node; Based on the features of the reference material, optimized prompt words are generated, and the optimized prompt words are used to replace the default prompt words of the prompt word nodes in the target workflow template.

6. The method according to claim 1, characterized in that, The process of adding, removing, or adjusting the node links of the target workflow template based on the creation goal includes: If an audio generation requirement is identified in the creation objective, then an audio generation node and a subtitle overlay node are inserted into the execution chain of the objective workflow template; If a batch generation requirement is identified in the generation parameters, a batch export node is added to the end of the link of the target workflow template.

7. The method according to claim 6, characterized in that, After adding a batch export node to the end of the link in the target workflow template, the method further includes: Identify the variable parameters in the scenario-based workflow; Based on the variable parameters, a corresponding batch data import template is generated so that users can fill in task information in batches using the batch data import template.

8. The method according to claim 1, characterized in that, After the rendering and display, it also includes: In response to the execution command, the scenario-based workflow is executed to generate the target content; The target content will be subject to quality scoring and compliance testing. Record the parameter configuration and user feedback data during the execution process, and perform weight iteration on the template parameters and node links in the scenario-based template library based on the recorded data.

9. A scenario-based workflow template intelligent adaptation system, characterized in that, The system includes: The scenario-based template library module is used to store a preset template system for core business scenarios; The requirement parsing module is used to parse requirement text and / or reference materials to obtain the target business scenario, creation goal, generation parameters and reference material characteristics; The template matching module is used to perform multi-layer matching logic in the scenario-based template library module based on the scenario matching feature vector to determine the target workflow template; The adaptation and optimization module is used to fill the execution parameters of the target workflow template based on the parsed parameters, and adjust the node links of the target workflow template according to the creation goal, so as to output the scenario-based workflow to be executed.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images