An advertisement material production method based on artificial intelligence

Abstract

The application relates to the fields of artificial intelligence and advertisement putting, and discloses an advertisement material production method based on artificial intelligence, which comprises the following steps: obtaining advertisement task original data, performing field integrity checking on the advertisement task original data to obtain structured advertisement task data; extracting marketing semantic nodes in the structured advertisement task data, performing relationship edge construction on the marketing semantic nodes to obtain a marketing intention graph; generating candidate materials according to the marketing intention graph; performing quality and compliance evaluation on the candidate materials to obtain available materials. By extracting the marketing semantic nodes and constructing the marketing intention graph, the platform constraint coding formed by platform specification parameters, layout constraint regions, region identifiers and boundary parameters is combined, constraint conditions are constructed and a multi-modal generation model is inputted, so that the material content is generated in a semantic order and is limited to fall into position, and key information is prevented from being cut or shielded.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and advertising technology, specifically to a method for creating advertising materials based on artificial intelligence, an electronic device, and a storage medium. Background Technology

[0002] In multi-platform digital advertising scenarios, this invention addresses the novel technical challenge of automatically generating and delivering advertising materials across various display formats and terminals, such as social media short videos, news feeds, and banners: how to simultaneously satisfy the constraints of each platform regarding aspect ratio, safe zone, text content area, and native visual style during the generation phase, ensuring that key selling points, etc., are effectively utilized. , Information such as these are fully visible and natively presented on various platforms to improve campaign performance and user experience.

[0003] Currently, existing technologies typically treat "material generation" and "platform adaptation" as separate steps—materials are first produced in batches using templates, manual design, or models, and then post-processed according to each platform's rules through cropping, scaling, manual fine-tuning, or independent adaptation modules, or gradually optimized based on post-launch data loops. However, material generation and platform adaptation remain disconnected, leading to mismatches in aspect ratios, safe zones, text coverage areas, and native style requirements across different platforms. This manifests as key information being cropped, etc. or Issues such as obstruction and lack of platform native feel in the creative materials can reduce the effectiveness of ad campaigns and user experience. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the present invention provides an artificial intelligence-based advertising material production method, electronic device and storage medium to at least partially solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides a method for creating advertising materials based on artificial intelligence, comprising:

[0007] Obtain the original data of the advertising task, perform field integrity verification on the original data of the advertising task, and obtain structured advertising task data;

[0008] The marketing semantic nodes are extracted from the structured advertising task data, and relational edges are constructed on the marketing semantic nodes to obtain a marketing intent graph.

[0009] Based on the marketing intent map, candidate materials are generated;

[0010] The candidate materials are evaluated for quality and compliance to obtain usable materials;

[0011] The available materials are style-optimized to obtain refined materials;

[0012] The available and refined materials are formatted to obtain the final advertising materials.

[0013] Preferably, the original data of the advertising task is subjected to field integrity verification, including:

[0014] Based on a preset field template, perform field alignment on the original data of the advertising task to generate field alignment results;

[0015] Each field after field alignment is compared with a preset set of required fields to identify missing fields and complete the missing field validation, generating a missing field validation result.

[0016] The structured advertising task data is generated based on the field alignment results and the missing data validation results.

[0017] Preferably, field alignment is performed on the original data of the advertising task, including:

[0018] Establish the correspondence between the original fields and standard fields in the original data of the advertising task based on the field template;

[0019] Map different original fields with the same semantics to a unified field name;

[0020] According to preset format rules, the mapped fields are standardized to give each field a unified data structure.

[0021] Output the field alignment result.

[0022] Preferably, the marketing semantic nodes extracted from the structured advertising task data include:

[0023] Semantic parsing is performed on the structured advertising task data to obtain marketing information;

[0024] The marketing information is categorized by pre-defined marketing semantic categories.

[0025] The categorized marketing information is then node-based to obtain the marketing semantic nodes.

[0026] Preferably, constructing relationship edges on the marketing semantic nodes to obtain a marketing intent graph includes:

[0027] Based on the marketing semantic category corresponding to the marketing semantic node, node pairing is performed on each marketing semantic node;

[0028] Perform semantic association analysis on the paired marketing semantic nodes to determine whether there is an association relationship between the nodes and the direction of the association;

[0029] Establish relationship edges between marketing semantic nodes that are related;

[0030] The marketing intent graph is generated based on the established relationship edges.

[0031] Preferably, the process of generating the preset platform constraint code includes:

[0032] Extract the platform constraint rules corresponding to the target platform set and generate platform specification parameters;

[0033] The layout constraint area is determined based on the platform specifications.

[0034] Each page's constraint area is coded with area identifiers and area boundary parameters.

[0035] The platform constraint code is obtained by combining the region identifier code and the region boundary parameter code.

[0036] Preferably, the candidate materials generated based on the marketing intent map include:

[0037] The marketing intent graph is analyzed, and the generation order of the marketing semantic nodes is determined based on the relationship edges between the marketing semantic nodes.

[0038] Decode the platform constraint code to obtain the region identifier and region boundary parameters corresponding to each page constraint region;

[0039] Generate constraints are constructed based on the generation order, region identifier, and region boundary parameters.

[0040] The marketing intent map and the generation constraints are input into the multimodal generation model, so that the multimodal generation model generates corresponding material content in the generation order, and restricts the generation location of the material content according to the region identifier and region boundary parameters;

[0041] Output the candidate materials.

[0042] Preferably, the candidate materials are subjected to quality and compliance assessments to obtain usable materials, including:

[0043] Based on the element safe zone occupancy rate of the candidate materials Selling point coverage Brand consistency and initial screen recognition Calculate the pre-deployment availability score The pre-deployment availability score is mentioned above. satisfy: ,

[0044] in, This refers to the occupancy rate of the element's safe zone. To increase the coverage of selling points, For brand consistency, For initial screen recognizability, , , , These are the weighting coefficients;

[0045] The pre-deployment availability score Compared with the preset scoring threshold In comparison, > When the candidate material is determined to be usable material, it is then identified.

[0046] Preferably, the available materials are style-optimized to obtain refined materials, including:

[0047] Obtain the native style requirements and brand guidelines of the target platform;

[0048] Based on the platform's native style requirements and brand guidelines, determine the visual style adjustment items, layout structure adjustment items, and content arrangement adjustment items in the available materials;

[0049] Adjust the color representation, font style, and image presentation of the available materials according to the aforementioned visual style adjustment options;

[0050] The layout relationship of the main visual area, text area, and logo area in the available materials is adjusted according to the layout structure adjustment items;

[0051] The presentation order of the content in the available materials is adjusted according to the content arrangement adjustment items.

[0052] The adjusted and usable materials are identified as the retouched materials.

[0053] Preferably, the available and refined materials are formatted to obtain the final advertising materials, including:

[0054] Based on the output format requirements of the target platform, the target output parameters are determined, including at least the material size parameters and the material format parameters;

[0055] Based on the material size parameters and material format parameters, the available materials and the retouched materials are converted in format and adapted in size to generate materials in the target format;

[0056] Based on the targeting platform's corresponding placement requirements, the target format material is labeled with platform identification information, placement identification information, and material type identification information to obtain the final advertising material after labeling metadata. The material size parameter characterizes the output size of the target material, the material format parameter characterizes the file format of the target material, the platform identification information characterizes the placement platform corresponding to the final advertising material, the placement identification information characterizes the placement position corresponding to the final advertising material, and the material type identification information characterizes the material category corresponding to the final advertising material.

[0057] In a second aspect, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, characterized in that: when the processor executes the computer program, it implements the steps of the method in the first aspect.

[0058] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that: when the computer program is executed by a processor, it implements the steps of the method in the first aspect.

[0059] This invention provides a method for creating advertising materials based on artificial intelligence. It has the following beneficial effects:

[0060] 1. This invention extracts marketing semantic nodes and constructs a marketing intent graph. Combined with platform constraint codes formed by platform specification parameters, layout constraint areas, area identifiers and boundary parameters, it constructs generation constraint conditions and inputs them into a multimodal generation model. This enables the material content to be generated in semantic order and its placement to be limited, avoiding the cutting or obscuring of key information.

[0061] 2. This invention calculates the pre-launch availability score by evaluating the element safety zone occupancy rate, selling point coverage rate, brand consistency, and initial screen recognizability of candidate materials. It then optimizes the visual style, layout structure, and content arrangement in accordance with the platform's native style requirements and brand specifications, thereby improving the availability of materials and reducing manual maintenance costs before launch. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating the method for creating advertising materials based on artificial intelligence, as provided in an embodiment of the present invention. Detailed Implementation

[0063] To enable those skilled in the art to understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0064] The present invention will now be described in detail with reference to the accompanying drawings:

[0065] Combined with appendix Figure 1 This invention provides an AI-based method for creating advertising creatives, applied to the scenario of automatically creating cross-platform advertising creatives for the cold start of new e-commerce products. Instead of generating advertising creatives first and then adapting them to the platform, this method first constructs a marketing intent graph and platform constraint codes, then generates target platform advertising creatives under the joint constraints of the marketing intent graph and platform constraint codes, and makes corrections based on pre-launch availability scores. This achieves automatic generation of multiple versions, native platform adaptation, brand consistency, and improved initial launch availability with minimal rework.

[0066] In this embodiment, the original data of the advertising task is first obtained. This original data may include product title, product attributes, main product image, selling point copy, landing page content, brand guidelines, campaign objectives, target platform set, and historical creative performance data. This data can be sourced from the product database, product detail page, brand manual, campaign configuration, and historical campaign data system. Preferably, product attributes, prices, and discount information are directly sourced from the product database and landing page; brand guidelines are preferably sourced from the brand manual; and historical creative performance data preferably includes impressions, clicks, conversions, initial frame dwell time, and completion rate.

[0067] Specifically, field integrity verification is used to organize raw advertising task data from different sources, in different formats, and with different expressions into a unified data structure that can be directly called upon for marketing intent modeling; field integrity verification includes at least field alignment processing and missing data verification processes. Field alignment processing may include:

[0068] Map similar fields from different sources to a unified field name; group product information, platform information, brand information, and campaign information into a pre-defined field structure; organize text fields, image fields, and historical performance fields into parsable data units; and integrate several candidate values ​​for the same field from different sources to form advertising task data with a unified field structure. For example, discount descriptions in product detail pages, price fields in product libraries, and promotional information in campaign copy can be unified into benefit-related fields; brand tone, prohibited expressions, and visual guidelines in brand manuals can be unified into brand guidelines fields, avoiding parsing errors caused by inconsistent field naming and scattered field sources during the node extraction stage.

[0069] The missing data verification includes checking for missing product titles, main product images, selling point copy, brand guidelines, campaign objectives, and target platform sets. For historical creative performance data, it further checks for at least some of the following statistics: impressions, clicks, conversions, initial frame dwell time, and completion rate. When a field is found to be missing, it is marked as a field to be supplemented, and the reliance on the corresponding metrics for the missing field is reduced during the modeling process. For example, when historical creative performance data is insufficient, platform constraint weights and brand constraint weights can be retained first, reducing reliance on historical conversion data to ensure that usable structured advertising task data can still be formed during the new product cold start phase. Through the above field integrity verification, structured advertising task data can be obtained and sent to the subsequent marketing intent graph construction step to provide a unified input for marketing intent modeling.

[0070] In this embodiment, structured advertising task data is extracted to obtain marketing semantic nodes. This extraction process is used to transform product information, placement information, and brand information from their original descriptive state into semantic units suitable for organizing advertising expression.

[0071] Specifically, product selling points, audience pain points, benefits, evidence points, activity information, and calls to action can be extracted from structured advertising task data to obtain corresponding marketing semantic nodes. Among these, product selling points characterize the core advantages of the product itself, such as functional advantages, material advantages, price advantages, or user experience advantages; audience pain points characterize the main problems encountered by target users in consumption or usage scenarios; benefits characterize the direct benefits the product can bring to users; evidence points characterize the factual basis supporting selling points and benefits, such as parameter descriptions, test results, sales information, user review summaries, or scenario demonstration results; activity information characterizes promotional activities, discount formats, or time-sensitive information; and calls to action guide users to further click, purchase, inquire, or redirect.

[0072] In one optional implementation, the text fields in the structured advertising task data can first be segmented and semantically categorized. Then, based on preset marketing semantic categories, the segmentation results are assigned to the node sets corresponding to product selling points, audience pain points, benefits, evidence points, activity information, and calls to action. For image fields, supplementary semantics related to the product subject, usage scenario, and visual style can be extracted based on the image content description results to assist in subsequent node weight calculations and visual placement constraints. Through the above processing, information originally scattered in product titles, landing page content, selling point copy, and brand guidelines can be transformed into a structured set of marketing semantic nodes.

[0073] After obtaining the marketing semantic nodes, relationship edges are constructed on them to obtain the marketing intent graph. The purpose of constructing relationship edges is to no longer treat the marketing semantic nodes as isolated pieces of information, but to organize them into a graph structure with an expressive logical order and dependencies.

[0074] Specifically, edge connections can be established based on the relationships between "selling point - evidence - benefit - audience - call to action". That is, connections can be established between the selling point node and the evidence node to explain "what evidence supports this selling point", between the evidence node and the benefit node to explain "how this evidence supports user benefits", between the benefit node and the audience node to explain "which audience this benefit targets or what pain point it addresses", and between the audience node and the call to action node to explain "what guiding actions should be taken for this audience". The marketing intent map generated in this way enables the advertising generation model to know "what to say", "what to say first, what to say next, and why the various expressions are related".

[0075] Furthermore, during the construction of the marketing intent graph, weights are calculated for the nodes in the graph to determine their priority order. This priority order characterizes the degree to which each marketing semantic node in the advertising creative should be presented first. The priority score can be expressed as:

[0076] ,

[0077] in, For marketing semantic node sequence number, , , , These are the weighting coefficients. Marketing semantic nodes The degree of match between the selling points and the target audience. Marketing semantic nodes Completeness of evidence, Marketing semantic nodes Consistency with brand proposition Marketing semantic nodes Adaptability to the target platform scenario. The aforementioned weighting coefficients can be obtained through regression analysis of historical high-performing creatives; if historical data is insufficient, initial values ​​can be manually provided and then updated based on feedback. This weighting calculation process determines the core message of candidate ad creatives, ensuring that high-priority marketing elements are prioritized and highlighted during the generation process.

[0078] In this embodiment, platform constraint codes are generated in advance before generating advertising materials. These platform constraint codes are used to incorporate the specifications of different advertising platforms into the generation stage, rather than passively trimming and adapting them after generation.

[0079] Specifically, the platform can retrieve the placement type, aspect ratio rules, safe zone rules, text length rules, audio rules, and material quantity rules from the target platform collection to obtain platform specification parameters. Placement types can include news feeds, short video feeds, and asset-based ad placements; aspect ratio rules can include display ratios such as horizontal, square, and vertical; and safe zone rules can include edge areas where text should not be displayed. The rules specify the areas that must not be obscured and the central area where the main product should be presented first. Text length rules may include maximum title character length, maximum description field length, or ad copy length limits. Audio rules may include whether voice is allowed, the duration of background music, and subtitle synchronization requirements. Material quantity rules may include the number of material versions required for a single campaign. Platform specifications are preferably derived from the official specification libraries of each platform. When platform specifications are updated, the platform constraint rule library needs to be updated, but the generation model does not need to be retrained. For short video platforms, the 9:16 vertical format can be set as a high-priority constraint; for information flow graphic platforms, the title area and the initial product main body area can be prioritized for constraint.

[0080] Furthermore, platform specifications can be mapped to screen composition areas, text placement areas, product main body security areas, and... The safe zone is used to obtain platform constraint codes. That is, after converting abstract platform rules into layout constraint information that the generative model can directly perceive and execute, the platform constraint codes can be bound to the marketing intent map to form a platform-based creative blueprint, so that the generative model is constrained in both content expression and layout.

[0081] In this embodiment, the marketing intent graph and platform constraint coding are used to generate candidate generated materials.

[0082] Specifically, marketing intent maps and platform constraint codes can be input into a multimodal generation model. The multimodal generation model can receive text semantic structure, layout constraint information, and product-related image materials, and output image ad creatives, short video ad creatives, or mixed text and image ad creatives. During the generation process, on the one hand, the expression order, keyword coverage, and visual placement of high-priority marketing semantic nodes are locked to ensure that high-weight nodes are given priority in the title, first sentence, main visual center area, or video opening shot; on the other hand, restrictions are placed on text, Product content and calls to action are placed in the non-safe zone to ensure platform rules are not violated and to reduce rework during later editing. Candidate materials can be generated through this method.

[0083] In one optional implementation, when the target platform is a short video platform, it is preferable to first generate a sequence of shots and then generate the corresponding video footage; when the target platform is an asset-based graphic and text ad, it is preferable to first generate the layout skeleton and then generate the corresponding images and text. For new product launch tasks, it is preferable to simultaneously generate three types of candidate ad materials: strong promotional, strong pain point, and strong experience, in order to improve the first-round launchability under different platforms and different audience segmentation scenarios.

[0084] In this embodiment, a quality and compliance assessment is performed on the candidate generated materials to obtain usable materials.

[0085] Specifically, candidate generated materials can be processed. Recognition, visual saliency analysis, key element localization analysis, and speech-text alignment analysis. Among them, Text content is extracted from the materials to check for completeness, whether it exceeds the designated area, or if there are any misidentifications. Visual salience analysis identifies the areas users are most likely to focus on during the initial viewing phase, determining whether key selling points, product elements, and brand logos are located in high-attention areas. Key element positioning analysis is used to identify text, The positional relationship of the product, promotional information, and call to action within the frame is analyzed to determine whether they have entered unsafe zones and whether they obscure each other. Voice-text alignment analysis is primarily used in short video advertising scenarios to determine whether the voiceover, subtitles, and camera content are consistent and synchronized. Through these analyses, the material detection results can be obtained.

[0086] Furthermore, platform compatibility, brand consistency, selling point completeness, and initial visual recognizability can be determined based on the analysis results. Platform compatibility can be determined based on the occupancy rate of key elements' safe zones, i.e., based on text, The following criteria are used to determine whether the product's main content and call to action information are located within the permitted area and whether they meet the target platform's proportion, safety zone, and copywriting requirements; brand consistency is used to characterize the degree of consistency between the candidate generated materials and the brand's guidelines, brand proposition, brand tone, and brand visual requirements; the completeness of selling points can be determined based on the selling point coverage rate, that is, the degree to which high-priority selling points, benefits, and evidence points in the marketing intent map are expressed and displayed in the candidate generated materials; the initial visual recognizability is used to characterize the comprehensibility and attractiveness of the materials in the initial presentation stage, and can be determined by combining the results of visual salience analysis, the position of the main body, the position of the title, and the visibility of key information.

[0087] The pre-launch usability score is calculated based on platform compatibility, brand consistency, completeness of selling points, and initial visual recognizability. The pre-launch usability score can be expressed as:

[0088] ,in, This refers to the occupancy rate of the element's safe zone. To increase the coverage of selling points, For brand consistency, For initial screen recognizability, , , , This is a weighting coefficient. The preset threshold is preferably determined based on historical samples of "directly deployable materials". If the pre-launch usability score is lower than the preset threshold, the process returns to the constraint generation step to regenerate candidate materials in a targeted manner. Targeted regeneration can correct low-scoring items; for example, when platform compatibility is low, priority is given to adjusting the placement of key elements and the occupancy of safe zones; when the completeness of selling points is low, priority is given to supplementing high-priority selling points or evidence; when brand consistency is low, priority is given to correcting brand tone or visual style to obtain usable materials. Through this process, unusable materials can be eliminated before actual launch.

[0089] In this implementation, after obtaining usable materials, the materials undergo style optimization to obtain refined materials. Style optimization primarily addresses the differences in native expression preferences across different platforms and the issue of maintaining brand consistency.

[0090] Specifically, based on the target platform type and brand guidelines, available creative materials can be optimized for platform native style adaptation and brand tone consistency. Platform native style adaptation mainly involves ensuring that the ad creatives align with the common content formats of the target platform in terms of shot rhythm, composition, copywriting tone, subtitle style, and visual style. Brand tone consistency optimization mainly involves ensuring that the ad creatives are consistent with brand guidelines in terms of word choice, color style, visual quality, product presentation, and call to action.

[0091] In one alternative implementation, when the target platform is a short video platform, the camera and narration rhythm are optimized to attract user attention from the initial stage and quickly convey the selling points. When the target platform is an asset-based graphic ad, the layout structure is optimized to clearly distinguish the product content, title, and guiding information. Through the above style optimization, refined materials that balance platform nativeness and brand consistency can be obtained.

[0092] In this embodiment, available and refined creative materials are formatted and annotated with metadata to obtain the final advertising creative. This step transforms the aforementioned generation and optimization results into standard output results that can be directly used for system calls, management, and reuse.

[0093] Specifically, formatted output processing may include: adjusting image size, video resolution, file format, encoding format, duration, bitrate, cover image format, and subtitle file format according to the target platform's material upload requirements; for text and image ad creatives, further outputting title, description, and supplementary information fields that match the platform's requirements; for short video ad creatives, further outputting video files, initial frame cover images, interstitial text, and subtitle content. Metadata annotation may include: associating output creatives with product identifiers, campaign identifiers, target platform identifiers, creative type identifiers, marketing theme identifiers, selling point tags, audience tags, brand version identifiers, generation time identifiers, and rating result identifiers, etc. Through the above processing, the final ad creative can be obtained.

[0094] In a further implementation, after selecting usable creatives through pre-launch usability scoring, a small-scale trial launch can be conducted to obtain feedback data such as click-through rate, conversion rate, initial frame dwell time, completion rate, and bounce rate. This feedback data is then written back into the marketing intent graph and the pre-launch usability scoring model to obtain updated weight parameters, which are then used to generate the next batch of creatives. The small-scale trial launch is preferably set with a controlled budget and controlled placement, avoiding a full rollout. When the objective is conversion, the weights of selling point completeness and call-to-action related items can be prioritized. When the objective is clicks or video views, the weights of initial screen recognizability and platform native feel related items can be prioritized. Online updates preferably only correct marketing node weights and platform constraint priorities, rather than frequently modifying the underlying large model parameters. This approach can further shorten the cold start exploration cycle for new products.

[0095] Example 1: Taking "Dynamic Adaptation to Short Video Platforms" as a Scenario

[0096] The scenario in this example is as follows: A 3C digital brand needs to create a vertical short video (9:16) advertisement for a "new noise-canceling headphone" and place it on short video platform A, which has complex UI obstruction areas (such as the right-side like and comment area and the bottom text area). This example focuses on how to solve the problem of key information being cropped or obscured through front-end platform constraint coding.

[0097] Step 1: Obtain and verify structured advertising task data

[0098] The system obtains raw data from the product information database and brand specification documents, including: product main image, selling point copy (50dB deep noise reduction), and target advertising platform (Short video platform A). The system performs field integrity checks to confirm that no required fields are missing and that the data format is consistent, and outputs structured advertising task data.

[0099] Step 2: Extract marketing semantic nodes and construct a marketing intent graph

[0100] The system invokes a semantic extraction model based on the BERT architecture to extract entity class nodes from structured data: selling point nodes (50dB deep noise reduction), pain point nodes (noisy environment), and CTA nodes (click-to-buy). Directed edges are established according to the FABE rule. Finally, a linear weighted model is used to calculate the priority expression score. :

[0101] ,

[0102] In this scenario, the "noise reduction" attribute has a historically high click-through rate on platform A, resulting in high performance data. The score was extremely high, and the system assigned the highest priority score to "50dB deep noise reduction".

[0103] Step 3: Parse the preset platform constraint code for page layout rules

[0104] The system calls the official API of short video platform A to dynamically obtain the placement rules.

[0105] Hard constraint analysis: Extract the 9:16 screen ratio requirement, and parse the specific coordinate range of the "Like / Comment Button Interaction Area" on the right side of Platform A and the "Account Name / System Subtitle Area" at the bottom into the coordinates of the unobstructed area.

[0106] Soft constraint parsing: Extract the native style preference of platform A ("high saturation, fast pace") and the text density limit of no more than 15 characters per frame for subtitles. The system encapsulates these rules into tensor and vector formats recognizable by the multimodal generation model, generating platform constraint codes.

[0107] Step 4: Constraint-based generation based on anchor box mechanism

[0108] The platform constraint codes and graphs mentioned above are input into the multimodal generative model for generation:

[0109] Hard constraint enforcement: Before denoising in the model's latent space, the right and bottom coordinates are mapped to an uncoverable region mask using an anchor box mechanism. The system forcibly locks key information with the highest priority expression score (such as the headphone image and the "50dB noise reduction" label) in a safe area outside the uncoverable region, absolutely preventing key information from being obscured by the UI of platform A at the physical level.

[0110] Soft Constraints and Attention Guidance: During generation, the priority expression score is injected as an attention guidance weight into the cross-attention layer, ensuring the core focus of the image is on the headphones. Simultaneously, by dynamically adjusting the loss function during generation and adding a text density penalty term, it is ensured that the generated auxiliary text in the image does not exceed 15 characters, complying with soft constraints. The output yields candidate materials.

[0111] Step 5: Formatting and Metadata Output The system performs timeline cropping on the candidate materials, outputs multi-scale variations with a standard 15-second duration, and extracts information such as "triggering the safety box constraint of the right-side like area of ​​platform A" as metadata and writes it into the file to obtain the final advertising material, achieving instant adaptation and completely solving the adaptation disconnect problem.

[0112] Example 2: Taking the "cold launch of high-end beauty products on an image and text-based product recommendation platform" as an example

[0113] The scenario in this example is as follows: A high-end international beauty brand launches a new "anti-aging face cream" and plans to conduct a cold start campaign on a social media platform (B) that emphasizes text and image content. This example focuses on how to ensure brand consistency and extremely high usability of creative materials in the absence of historical campaign data by using a strict semantic weight scoring and pre-campaign evaluation closed loop.

[0114] Step 1: Obtain and Verify Structured Advertising Task Data. The system acquires anti-aging face cream data, where the "Brand Tone Standard" field explicitly specifies "minimalist, high-end, professional". After system verification, it outputs structured data.

[0115] Step 2: A high-weight, brand-oriented marketing intent mapping system extracts selling point nodes (anti-aging ingredients), evidence point nodes (clinical trial data), etc., and calculates the priority expression score. hour:

[0116] ,

[0117] For this high-end luxury brand scenario, the system dynamically adjusts the alignment with the brand's proposition. Corresponding weight coefficients and the completeness of evidence Corresponding weight coefficients Because this is a cold start for the first team, historical performance data... If missing, the system will assign weights. Reduced to 0. The ranking is calculated to ensure that brand tone and professional endorsement become the highest priority expression nodes, generating a marketing intent map.

[0118] Step 3: Perform constraint generation to obtain candidate materials. The system pulls the constraint codes of platform B (such as 4:3 aspect ratio, soft native filter), and combines them with the above map to call the multimodal generation model to generate the first batch of "candidate materials".

[0119] Step 4: Perform a quality and compliance assessment

[0120] To avoid distributing substandard materials that could damage the brand image, the system intercepts these materials at the front end and uses a multi-dimensional AI model for pre-distribution and review.

[0121] Visual salience analysis and OCR: Detect whether the focus of the image is cluttered and whether the copy contains promotional words that are inconsistent with the high-end style.

[0122] Multimodal semantic alignment detection: Image feature vectors extracted using the CLIP model are compared with text feature vectors to confirm whether the generated image and text are aligned. A pre-deployment usability score is then calculated; the system detected that a certain candidate material had a lower brand consistency score due to overly vibrant background decorations. Low. Total score. The material is below the ideal threshold for direct deployment, but is still within the correctable range, therefore it is considered usable material.

[0123] Step 5: Automatic Correction and Style Optimization

[0124] For the aforementioned areas where points were lost, the system automatically generates correction instructions without the need for human art intervention:

[0125] The system generates a "scale main element proportion" instruction, requiring the face cream bottle to be enlarged by 1.2 times to highlight the main element; and a "simplify copy text" instruction, removing unnecessary modifiers. The system inputs the "available materials" and "correction instructions" into an image editing model based on mask redrawing, fine-tuning local coordinates and re-performing color mapping to align with a high-end minimalist style. The resulting "refined materials" fully meet the requirements. Finally, the refined materials are annotated with metadata and output as the final advertising material. This solution, with zero test data during a cold start, ensures high-quality materials and absolute brand consistency through a pre-evaluation and self-correction closed loop.

[0126] Embodiments of the present invention have been presented and described. It will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0127] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device to execute the methods of each embodiment or some parts of the embodiments.

[0128] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0129] It is understood that the relevant features in the above methods and apparatus can be referenced interchangeably. Furthermore, the terms "first," "second," etc., in the above embodiments are used to distinguish between embodiments and do not represent the superiority or inferiority of any particular embodiment.

[0130] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0131] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0132] In addition, the memory may include non-permanent memory in computer-readable media, random access memory and / or non-volatile memory, such as read-only memory or flash memory, and the memory includes at least one memory chip.

[0133] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0134] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0135] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0136] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0137] In a typical configuration, an electronic device includes one or more processors, input / output interfaces, network interfaces, and memory.

[0138] Memory may include non-persistent memory in computer-readable media, random access memory, and / or non-volatile memory, such as read-only memory or flash memory. Memory is an example of computer-readable media.

[0139] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory, static random access memory, dynamic random access memory, other types of random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other memory technologies, optical disc read-only memory, digital versatile optical disc or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by electronic devices. As defined herein, computer-readable media do not include temporary computer-readable media, such as modulated data signals and carrier waves.

[0140] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0141] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0142] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0143] It should be understood that when the terms "first," "second," "third," and "fourth," etc., are used in the claims, specification, and drawings of this application, they are used only to distinguish different objects and not to describe a specific order. The terms "comprising" and "including" as used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0144] It should be understood that the terminology used herein is for the purpose of describing particular embodiments and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should be further understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0145] The embodiments described above are for the purpose of facilitating understanding of this application and are not intended to limit the scope or application scenarios of this application. Any person skilled in the art may make any modifications and changes in form and detail of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.

Claims

1. A method for creating advertising materials based on artificial intelligence, characterized in that, include: Obtain the original data of the advertising task, perform field integrity verification on the original data of the advertising task, and obtain structured advertising task data; The marketing semantic nodes are extracted from the structured advertising task data, and relational edges are constructed on the marketing semantic nodes to obtain a marketing intent graph. Based on the marketing intent map, candidate materials are generated; The candidate materials are evaluated for quality and compliance to obtain usable materials; The available materials are style-optimized to obtain refined materials; The available and refined materials are formatted to obtain the final advertising materials.

2. The method for producing advertising materials according to claim 1, characterized in that, Perform field integrity checks on the original data of the advertising task, including: Based on a preset field template, perform field alignment on the original data of the advertising task to generate field alignment results; Each field after field alignment is compared with a preset set of required fields to identify missing fields and complete the missing field validation, generating a missing field validation result. The structured advertising task data is generated based on the field alignment results and the missing data validation results.

3. The method for producing advertising materials according to claim 2, characterized in that, Field alignment is performed on the original data of the advertising task, including: Establish the correspondence between the original fields and standard fields in the original data of the advertising task based on the field template; Map different original fields with the same semantics to a unified field name; According to preset format rules, the mapped fields are standardized to give each field a unified data structure. Output the field alignment result.

4. The method for producing advertising materials according to claim 1, characterized in that, Extracting marketing semantic nodes from the structured advertising task data includes: Semantic parsing is performed on the structured advertising task data to obtain marketing information; The marketing information is categorized by pre-defined marketing semantic categories. The categorized marketing information is then node-based to obtain the marketing semantic nodes.

5. The method for producing advertising materials according to claim 1, characterized in that, By constructing relational edges on the aforementioned marketing semantic nodes, a marketing intent graph is obtained, including: Based on the marketing semantic category corresponding to the marketing semantic node, node pairing is performed on each marketing semantic node; Perform semantic association analysis on the paired marketing semantic nodes to determine whether there is an association relationship between the nodes and the direction of the association; Establish relationship edges between marketing semantic nodes that are related; The marketing intent graph is generated based on the established relationship edges.

6. The method for producing advertising materials according to claim 1, characterized in that, The process of generating the pre-defined platform constraint code includes: Extract the platform constraint rules corresponding to the target platform set and generate platform specification parameters; The layout constraint area is determined based on the platform specifications. Each page's constraint area is coded with area identifiers and area boundary parameters. The platform constraint code is obtained by combining the region identifier code and the region boundary parameter code.

7. The method for producing advertising materials according to claim 1, characterized in that, Based on the marketing intent map, candidate materials are generated, including: The marketing intent graph is analyzed, and the generation order of the marketing semantic nodes is determined based on the relationship edges between the marketing semantic nodes. Decode the platform constraint code to obtain the region identifier and region boundary parameters corresponding to each page constraint region; Generate constraints are constructed based on the generation order, region identifier, and region boundary parameters. The marketing intent map and the generation constraints are input into the multimodal generation model, so that the multimodal generation model generates corresponding material content in the generation order, and restricts the generation location of the material content according to the region identifier and region boundary parameters; Output the candidate materials.

8. The method for producing advertising materials according to claim 1, characterized in that, The candidate materials are subjected to quality and compliance assessments to obtain usable materials, including: Based on the element safe zone occupancy rate of the candidate materials Selling point coverage Brand consistency and initial screen recognition Calculate the pre-deployment availability score The pre-deployment availability score is mentioned above. satisfy: , in, For element safe zone occupancy rate, To increase the coverage of selling points, For brand consistency, For initial screen recognizability, , , , These are the weighting coefficients; The pre-deployment availability score Compared with the preset scoring threshold In comparison, > When the candidate material is determined to be usable material, it is then identified.

9. The method for producing advertising materials according to claim 1, characterized in that, The available materials are style-optimized to obtain refined materials, including: Obtain the native style requirements and brand guidelines of the target platform; Based on the platform's native style requirements and brand guidelines, determine the visual style adjustment items, layout structure adjustment items, and content arrangement adjustment items in the available materials; Adjust the color representation, font style, and image presentation of the available materials according to the aforementioned visual style adjustment options; The layout relationship of the main visual area, text area, and logo area in the available materials is adjusted according to the layout structure adjustment items; The presentation order of the content in the available materials is adjusted according to the content arrangement adjustment items. The adjusted and usable materials are identified as the retouched materials.

10. The method for producing advertising materials according to claim 1, characterized in that, The available and refined materials are formatted to obtain the final advertising materials, including: Based on the output format requirements of the target platform, the target output parameters are determined, including at least the material size parameters and the material format parameters; Based on the material size parameters and material format parameters, the available materials and the retouched materials are converted in format and adapted in size to generate materials in the target format; Based on the targeting platform's corresponding placement requirements, the target format material is labeled with platform identification information, placement identification information, and material type identification information to obtain the final advertising material after labeling metadata. The material size parameter characterizes the output size of the target material, the material format parameter characterizes the file format of the target material, the platform identification information characterizes the placement platform corresponding to the final advertising material, the placement identification information characterizes the placement position corresponding to the final advertising material, and the material type identification information characterizes the material category corresponding to the final advertising material.

Images