Advertisement pushing picture generation method and system based on style transfer

Abstract

The application provides an advertisement pushing picture generation method and system based on style migration. First, the advertisement style demand is analyzed to obtain advertisement style guiding parameters for defining the style migration direction of a generative artificial intelligence model. Then, the parameters are used to drive a pre-trained generative artificial intelligence model to perform style migration processing to obtain a style migration base picture. The base picture is used to adapt advertisement core elements to obtain an element integrated advertisement picture, so that the elements and the picture style are consistent. The element integrated advertisement picture is visually adapted and adjusted in combination with advertisement pushing scene information to obtain a pushing adapted advertisement picture that meets the display specifications of a pushing terminal. Finally, the pushing adapted advertisement picture is output to an advertisement pushing system and pushed to a target audience terminal device. The application quickly and accurately generates an advertisement pushing picture that adapts to different scenes and audience demands, thereby improving the advertisement propagation effect.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a method and system for generating advertising push screens based on style transfer. Background Technology

[0002] In the advertising industry, with increasingly fierce market competition and ever-diversifying consumer demands, generating more attractive and targeted advertising visuals has become a key issue for improving advertising effectiveness. Traditional methods of generating advertising visuals primarily rely on manual design and production. Designers need to manually conceive and draw the visuals based on factors such as the target audience and brand positioning. This approach is not only inefficient, but the design results also largely depend on the designer's personal experience and creative level, making it difficult to guarantee that the advertising visuals accurately meet the aesthetic preferences and brand communication needs of different target audiences.

[0003] In recent years, while some computer-aided design (CAD) technologies for generating ad visuals have emerged, these technologies typically only provide simple design templates and tools. Designers still need to spend a significant amount of time and effort manually adjusting and optimizing them. Furthermore, these technologies have significant shortcomings in handling ad style migration and adaptation to different push notification scenarios. For example, when applying an ad visual with a specific style to different push notification terminals or scenarios, it often requires redesign and adjustment, failing to achieve automated style migration and scenario adaptation. This leads to increased ad production costs and makes it difficult to quickly respond to market changes and user needs. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for generating ad push screens based on style transfer, the method comprising:

[0005] The advertising style requirements are analyzed to obtain advertising style guidance parameters. The advertising style requirements include the visual style type preferred by the advertising target audience and the visual identity features of the advertising brand. The advertising style guidance parameters are used to define the style transfer direction of the generative artificial intelligence model.

[0006] The pre-trained generative artificial intelligence model is driven by the advertising style guidance parameters to perform style transfer processing to obtain the style transfer base image. The generative artificial intelligence model includes a style feature conversion module and a content feature fusion module. The style transfer base image carries the visual style corresponding to the advertising style guidance parameters.

[0007] Based on the style transfer base image, the core elements of the advertisement are adapted to obtain the element-integrated advertisement image. The core elements of the advertisement include the visual image of the advertised product and the advertising text information. The visual style of the core elements of the advertisement in the element-integrated advertisement image is consistent with that of the style transfer base image.

[0008] By combining the advertising push scenario information, the elements are integrated into the advertising screen for visual adaptation and adjustment to obtain the push-adapted advertising screen. The advertising push scenario information includes the display characteristics of the push terminal and the ambient light adaptation requirements of the push scenario. The push-adapted advertising screen conforms to the display specifications of the push terminal.

[0009] The push-adapted advertising screen is output to the advertising push system, which then pushes the push-adapted advertising screen to the terminal devices of the target audience.

[0010] In another aspect, embodiments of the present invention also provide an advertising push screen generation system based on style transfer, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the machine-readable storage medium to implement the above-described method.

[0011] Based on the above, this invention obtains advertising style guidance parameters by analyzing advertising style requirements, enabling the generated advertising images to accurately match the visual style type preferred by the target audience and the visual identity characteristics of the advertising brand, breaking the ambiguity and limitations of traditional advertising image generation in style positioning. The advertising style guidance parameters drive a pre-trained generative artificial intelligence model to perform style transfer processing. Utilizing the style feature conversion module and content feature fusion module in the model, the target style can be efficiently transferred to the base image, quickly obtaining a base image carrying the corresponding visual style, greatly improving the efficiency of advertising image generation. Based on the style transfer base image, the core elements of the advertisement are adapted, ensuring that the visual images of the advertising product and the advertising text information are consistent with the visual style of the image, enhancing the overall coordination and attractiveness of the advertising image, and improving the communication effect of the advertising information. Combining the advertising push scenario information, the elements of the integrated advertising image are visually adapted and adjusted, ensuring that the generated push-adapted advertising image meets the display specifications of the push terminal and the ambient light adaptation requirements of the push scenario, achieving optimized display of the advertising image in different scenarios, further improving the advertising delivery effect. The final output is a customized ad display to the ad push system and pushed to the target audience's terminal devices. It can quickly and accurately generate ad push displays that adapt to different scenarios and audience needs, effectively reducing ad production costs and improving the efficiency and accuracy of ad dissemination. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the style transfer-based ad push screen generation method provided in the embodiments of the present invention.

[0013] Figure 2 This is a schematic diagram of exemplary hardware and software components of the style transfer-based advertising push screen generation system provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an embodiment of the ad push screen generation method based on style transfer provided by the present invention. The following is a detailed description of the ad push screen generation method based on style transfer.

[0015] Step S110: Analyze the advertising style requirements to obtain advertising style guidance parameters. The advertising style requirements include the visual style type preferred by the target audience of the advertisement and the visual identity features of the advertising brand. The advertising style guidance parameters are used to define the style transfer direction of the generative artificial intelligence model.

[0016] This embodiment uses the summer new product promotion of a new tea beverage brand targeting young consumers as an example for illustration. The advertising style requirements clearly point to a fresh and natural visual style preferred by the target audience, while also incorporating the brand's own visual identity characteristics. In practice, the first step is to comprehensively structure the collected advertising style requirements document.

[0017] This means accurately extracting various descriptive information related to the visual style from the document, such as specific statements like "the image should convey the fresh feeling of a summer beach, with blue and green as the main colors, paired with simple geometric compositions, and an overall texture that presents a delicate frosted feel." At the same time, it's also necessary to extract relevant descriptions of the brand's visual identity characteristics, such as "the brand's main color is a specific mint green, the iconic graphic is a leaf shape, and the brand font is a rounded sans-serif font." Through systematically sorting out this information, the approximate range of advertising style guidance parameters can be initially determined.

[0018] Step S111: Deconstruct the visual style type in the advertising style requirements, and divide it into color tone dimension, composition structure dimension and texture dimension. Each dimension corresponds to the specific visual expression form preferred by the advertising target audience.

[0019] Based on step S110, the extracted visual style type description needs to be broken down in detail. For the description "fresh and natural visual style, presenting the fresh feeling of a summer beach, with blue and green as the main colors, matched with simple geometric composition, and the overall texture presents a delicate frosted texture", it is clearly divided into three dimensions.

[0020] The color tone dimension corresponds to a refreshing color scheme dominated by blue and green; the composition dimension corresponds to a simple geometric arrangement; and the texture dimension corresponds to a delicate matte finish. Each dimension is strictly based on the extraction of key elements from the visual style description, ensuring that each dimension accurately reflects the preferences of the target audience.

[0021] Step S112: Extract the visual identity features of the advertising brand from the advertising style requirements, and separate the brand-specific color system, the brand's iconic graphic elements, and the brand's font style. The brand-specific color system includes the matching relationship between the brand's primary color and secondary colors.

[0022] Next, specific extraction of the brand visual identity features in the advertising style requirements was carried out. From the description that "the brand's main color is a specific mint green, the iconic graphic is a leaf shape, and the brand font is a rounded sans-serif font," the brand's exclusive color system, the brand's iconic graphic elements, and the brand font style were gradually separated.

[0023] The brand's exclusive color system is defined as using a specific mint green as the main color, supplemented by light white as the auxiliary color based on the brand's past visual data. The combination of the two has been verified through long-term market testing, with mint green accounting for a higher proportion and light white accounting for a lower proportion. The brand's iconic graphic element is defined as a leaf shape, with its unique outline curve and vein distribution characteristics. The brand's font style is a rounded sans-serif font with specific letter weight and character spacing characteristics. These characteristics together constitute the brand's unique visual identity.

[0024] Step S113: For the color tone dimension, map the color expression form preferred by the target audience of the advertisement to color space parameters that can be recognized by the generative artificial intelligence model. The color space parameters are used to define the direction of color adjustment in the style transfer process.

[0025] After defining the color tone dimensions, it is necessary to accurately map the corresponding color expressions to color space parameters. For the expression of "a refreshing color scheme dominated by blue and green", professional color analysis tools are needed to define the specific range of blue and green.

[0026] Specifically, this involves determining the numerical range of each component in commonly used color spaces, such as the range of red, green, and blue components for blue and the range of red, green, and blue components for green in the RGB color space. These numerical ranges are then systematically integrated into color space parameters that can be recognized by a generative artificial intelligence model. These parameters will guide the model to make targeted adjustments to the colors of the image during the subsequent style transfer process, ensuring that they conform to the color tone preferred by the target audience.

[0027] Step S1131: Extract the main color type from the color expression form preferred by the target audience of the advertisement. The main color type includes warm color series or cool color series, and each color series corresponds to multiple specific color types.

[0028] To map the color tone dimension, the first step is to extract the dominant color type. From the phrase "a refreshing color scheme dominated by blue and green," we can clearly identify the dominant color type as a cool color family. The cool color family contains a wide variety of specific colors, including blue, green, and cyan.

[0029] In this embodiment, two specific color types are mainly involved: blue and green. The blue is closer to sky blue, which can convey a sense of openness and freshness; the green is closer to grass green, which can show the characteristics of nature and vitality. The choice of both colors is consistent with the theme of summer new product promotion.

[0030] Step S1132: Determine the color range parameter corresponding to the main color type. The color range parameter includes the hue range and saturation range of the main color. The hue range is used to limit the color bias of the main color, and the saturation range is used to limit the color intensity of the main color.

[0031] After determining that the main color scheme is a cool color family, it is necessary to further determine its color gamut parameters. For the blue in the main color scheme, its hue range is determined using professional color analysis tools, corresponding to a specific angular range of blue on the color wheel.

[0032] At the same time, the saturation range also needs to be carefully determined. The saturation in this range should present a refreshing visual effect without being too intense and glaring. For green, the hue range is also determined as a specific angular range corresponding to green on the color wheel. The principle for determining the saturation range is the same as for blue, to ensure the harmony and unity of blue and green in terms of saturation and avoid color conflicts.

[0033] Step S1133: Extract the auxiliary color type from the color expression forms preferred by the target audience of the advertisement. The auxiliary color type is used to match the main color type to enhance the visual hierarchy of the image.

[0034] After determining the primary color scheme, it is necessary to extract the secondary color scheme. From the perspective of achieving a fresh and natural visual style, the selection of the secondary color scheme is crucial; it needs to complement the primary color scheme well while enhancing the visual depth of the image.

[0035] Based on the analysis of the target audience's aesthetic preferences, white and light gray were selected as the auxiliary color types. White and light gray can be well paired with cool-toned blue and green. Through a reasonable contrast in brightness, the image becomes more layered, avoiding monotony due to an overly concentrated main color, while also enhancing the image's transparency.

[0036] Step S1134: Determine the color matching ratio parameter corresponding to the auxiliary color type. The color matching ratio parameter is used to define the proportion of the auxiliary color in the overall picture color. This proportion needs to form a visual balance with the main color.

[0037] To achieve visual balance between the secondary and primary colors, it is necessary to determine the color proportion parameters of the secondary colors. In this embodiment, the total proportion of the primary colors blue and green in the overall image color scheme was determined to be a suitable value after multiple adjustments, and the proportions of blue and green were also optimized based on visual effects.

[0038] The proportions of the secondary colors, white and light gray, were also carefully designed, with each color's proportion carefully defined. This arrangement highlights the freshness of the main color scheme while enhancing the visual clarity of the image through the secondary colors, achieving overall visual balance.

[0039] Step S1135: Integrate the color range parameters of the primary color tone and the color matching ratio parameters of the secondary color tone, and convert them into a numerical parameter form that can be recognized by the color processing module of the generative artificial intelligence model. The numerical parameter form is matched with the color feature input format of the generative artificial intelligence model to obtain the color space parameters.

[0040] The color gamut parameters of the primary color tone (including the hue and saturation ranges of blue and green) and the color matching ratio parameters of the secondary color tone (the ratio of white to light gray) are systematically integrated. During the integration process, the input format required by the color processing module of the generative artificial intelligence model must be strictly followed.

[0041] The above parameters are converted into numerical arrays. For example, the angular range of the hue interval, the percentage range of the saturation interval, and the percentage of the color matching ratio are all converted into corresponding values, forming a multi-dimensional numerical array. This numerical array is the color space parameter, which can be directly read and processed by the model's color processing module.

[0042] Step S114: For the composition structure dimension, the composition representation preferred by the target audience of the advertisement is transformed into composition layout parameters that can be executed by the generative artificial intelligence model. The composition layout parameters are used to standardize the arrangement of elements in the style transfer process.

[0043] Regarding the compositional structure dimension, the corresponding compositional expression is "simple geometric composition." This expression needs to be translated into specific compositional layout parameters. First, it's essential to deeply analyze the characteristics of simple geometric composition, including key elements such as the type of shape (e.g., circle, square, triangle), the size and proportion of the shapes, and their distribution within the image.

[0044] Then, these characteristics are translated into executable parameters for the model, such as determining the positional coordinate range of the main geometric shapes in the image, the size ratio of the shapes, and the spacing between the shapes. These parameters together constitute the composition layout parameters. During the style transfer process, the model will arrange the image elements according to these parameters to ensure that the composition conforms to the preferences of the target audience.

[0045] Step S115: For the texture dimension, the texture representation preferred by the target audience of the advertisement is converted into texture feature parameters that can be processed by the generative artificial intelligence model. The texture feature parameters are used to control the surface texture presentation during the style transfer process.

[0046] To address the "delicate matte texture" aspect of texture quality, it's necessary to convert it into texture feature parameters. First, a comprehensive analysis of the matte texture is conducted, extracting key features such as particle size, distribution density, and contrast.

[0047] The above features are converted into corresponding numerical parameters, such as the pixel size range corresponding to particle size, the range of particle count per unit area corresponding to distribution density, and the range of brightness difference corresponding to contrast. These parameters constitute the texture feature parameters. During processing, the model will use these parameters to precisely control the surface texture of the image, giving the image a delicate frosted texture.

[0048] Step S116: Integrate the color space parameters, composition layout parameters, texture feature parameters, and brand-specific color system, brand-identical graphic elements, and brand font styles corresponding to the brand visual identity features to form the advertising style guidance parameters. Each dimension of the advertising style guidance parameters corresponds one-to-one with the style transfer module of the generative artificial intelligence model.

[0049] Finally, the color space parameters, composition layout parameters, texture feature parameters, and the brand-specific color system (mint green as the main color, light white as the auxiliary color and matching ratio) corresponding to the brand visual identity features, the brand's iconic graphic elements (leaf shape and features), and the brand font style (rounded sans-serif font and features) are fully integrated.

[0050] During the integration process, the parameters were strictly organized into a structured set according to the parameter input requirements of the style transfer module of the generative artificial intelligence model. This ensured that the parameters in each dimension matched the corresponding style transfer module in the model; for example, the color space parameters corresponded to the model's color transfer module, and the composition layout parameters corresponded to the model's composition arrangement module. Through this integration, a complete set of advertising style guidance parameters was formed.

[0051] Step S120: Drive the pre-trained generative artificial intelligence model with the advertising style guidance parameters to perform style transfer processing to obtain the style transfer base image. The generative artificial intelligence model includes a style feature conversion module and a content feature fusion module. The style transfer base image carries the visual style corresponding to the advertising style guidance parameters.

[0052] After obtaining the advertising style guidance parameters, they are input into a pre-trained generative AI model to drive the model to perform style transfer processing. This generative AI model has been pre-trained on a large number of advertising image style transfer samples and has the ability to generate corresponding style images based on the input style parameters.

[0053] The style feature conversion module in the model is responsible for in-depth processing of the advertising style guidance parameters and extracting style features; the content feature fusion module is responsible for introducing basic content materials and extracting content features. Through the collaborative work of the two modules, the model organically integrates style features and content features, ultimately generating a style transfer base image that carries the visual style corresponding to the advertising style guidance parameters. In this embodiment, the generated style transfer base image should present a visual effect with blue and green as the main colors, simple geometric composition, and a delicate frosted texture, while subtly reflecting the brand's mint green main color and leaf graphic elements.

[0054] Step S121: Input the advertising style guidance parameters into the style feature conversion module of the generative artificial intelligence model. The style feature conversion module performs feature encoding on the color space parameters, composition layout parameters and texture feature parameters in the advertising style guidance parameters to obtain a style feature vector. The dimension of the style feature vector matches the feature processing channel of the generative artificial intelligence model.

[0055] The advertising style guidance parameters generated in step S116 are input into the style feature conversion module of the generative artificial intelligence model. This style feature conversion module first performs a detailed analysis of the color space parameters, composition layout parameters, and texture feature parameters in the advertising style guidance parameters.

[0056] For color space parameters, key information such as the dominant hue range, saturation range, and auxiliary hue proportions are extracted. For composition and layout parameters, important information such as the position, size, and spacing of geometric shapes are extracted. For texture feature parameters, core information such as grain size, distribution density, and contrast is extracted. Then, a feature encoding algorithm transforms this information into a multi-dimensional numerical vector, namely the style feature vector. The dimension of this style feature vector is set to match the number of feature processing channels in the subsequent generative AI model to ensure that features can be smoothly transferred and processed within the model.

[0057] Step S122: Call the content feature fusion module of the generative artificial intelligence model, import the basic advertising content material, and use the content feature fusion module to extract content features from the basic advertising content material to obtain a content feature vector. The basic advertising content material contains general visual elements related to the advertising theme.

[0058] The content feature fusion module of the generative artificial intelligence model is invoked to import basic advertising content materials related to the summer new product promotion theme of the new-style tea drinks. These materials include images of common summer elements, such as waves, beaches, fruits (lemons, watermelons, etc.), ice cubes, and other general visual elements.

[0059] The content feature fusion module preprocesses these materials, including size standardization and color normalization, to ensure that the material format meets the model's processing requirements. Then, feature extraction algorithms such as convolutional neural networks are used to perform in-depth analysis of the preprocessed materials, extracting content features such as the shape of ocean waves, the texture of sand, and the shape of fruit. These features are then integrated into a content feature vector, which also has multi-dimensional attributes. The dimensions of this vector are determined based on the complexity of the content materials and the model's design requirements for subsequent feature fusion.

[0060] Step S123: In the feature interaction layer of the generative artificial intelligence model, the style feature vector and the content feature vector are cross-dimensionally associated and fused to generate a fused feature matrix, which simultaneously carries the association information of style features and content features.

[0061] In the feature interaction layer of the generative artificial intelligence model, the style feature vector obtained in step S121 and the content feature vector obtained in step S122 are fused across dimensions. First, the feature interaction layer performs a comprehensive analysis of each dimension of the two vectors to determine the relationship between the style feature dimension and the content feature dimension.

[0062] For example, the color style dimension is related to the color element dimension in the content material, and the composition style dimension is related to the element layout dimension in the content material. Then, based on these relationships, a feature fusion algorithm is used to fuse the corresponding dimensions of the two vectors, organically integrating the information of style features and content features. After fusion processing, a two-dimensional fusion feature matrix is ​​generated. The rows and columns of this two-dimensional fusion feature matrix correspond to different feature dimensions, and each element in the matrix carries the correlation information between style features and content features in their corresponding dimensions.

[0063] Step S1231: Determine the number of feature dimensions of the style feature vector and the feature meaning of each dimension. The feature meaning includes color feature dimension, composition feature dimension and texture feature dimension, and each feature dimension corresponds to multiple feature sub-items.

[0064] Before performing cross-dimensional correlation and fusion, it is first necessary to determine the number of feature dimensions of the style feature vector and the meaning of each dimension. Assuming that the style feature vector has multiple dimensions, one part of which is the color feature dimension, corresponding to the feature sub-items such as primary hue, saturation, and secondary hue proportion in the color space parameters, with each sub-item corresponding to one or more dimensions.

[0065] The other part is the composition feature dimension, which corresponds to the graphic position, size, spacing and other feature sub-items in the composition layout parameters; the remaining part is the texture feature dimension, which corresponds to the grain size, distribution density, brightness and contrast and other feature sub-items in the texture feature parameters.

[0066] Step S1232: Determine the number of feature dimensions of the content feature vector and the feature meaning of each dimension. The feature meaning includes the content subject feature dimension, the background element feature dimension, and the spatial relationship feature dimension. Each feature dimension corresponds to multiple feature sub-items.

[0067] Similarly, it is necessary to determine the number of feature dimensions of the content feature vector and the meaning of each dimension. The dimensions of the content feature vector are consistent with those of the style feature vector, with one part being the main content feature dimension, corresponding to the shape, outline, and other feature sub-items of the main elements (such as waves and fruits) in the basic advertising content material.

[0068] Another part is the background element feature dimension, corresponding to the texture, color, and other features of secondary elements in the material (such as the beach and the sky); the remaining part is the spatial relationship feature dimension, corresponding to the position, hierarchy, and other features between the main element and the background element. Each feature sub-item also corresponds to one or more specific dimensions, ensuring that the content features can be comprehensively and accurately described.

[0069] Step S1233: Construct a dimension association mapping table in the feature interaction layer. The dimension association mapping table is used to establish the corresponding association relationship between each dimension of the style feature vector and each dimension of the content feature vector.

[0070] After clarifying the meaning of each dimension of the style feature vector and content feature vector, a dimension association mapping table is constructed in the feature interaction layer. During the construction process, based on the correlation of feature meanings, the color feature dimension of the style feature vector is associated with the color-related sub-dimensions in the background element feature dimension of the content feature vector.

[0071] The compositional feature dimension of the style feature vector is associated with the spatial relationship feature dimension of the content feature vector; the texture feature dimension of the style feature vector is associated with the content subject feature dimension and the texture-related sub-dimensions of the background element feature dimension of the content feature vector. The mapping table records in detail the corresponding dimension numbers of each dimension of the style feature vector and the content feature vector, forming a one-to-one association.

[0072] Step S1234: Based on the dimensional association mapping table, perform dimensional feature value association calculations on the style feature vector and content feature vector to generate a dimensional association feature matrix. Each element in the dimensional association feature matrix represents a combination of feature values ​​of a set of associated dimensions.

[0073] Step S12341: Extract the first set of associated dimension pairs from the dimension association mapping table. The associated dimension pair includes a feature dimension of the style feature vector and a feature dimension of the content feature vector, such as the composition feature dimension of the style feature vector and the spatial relationship feature dimension of the content feature vector.

[0074] From the constructed dimension association mapping table, the first set of associated dimension pairs is extracted in a preset order. In this embodiment, the graphic position dimension under the composition feature dimension in the style feature vector and the main element position dimension under the spatial relationship feature dimension in the content feature vector are selected to form the first set of associated dimension pairs. The above two dimensions reflect the composition position requirements at the style level and the element spatial distribution characteristics at the content level, respectively. The correlation between the two can ensure that the position layout of the main elements in the picture conforms to the preset style composition specifications.

[0075] Step S12342: Extract the feature value set of the style feature dimension and the feature value set of the content feature dimension in the first set of associated dimensions. The feature value set contains the specific values ​​of all feature sub-items under the content feature dimension.

[0076] For the first pair of related dimensions, feature value sets are extracted for both the style feature dimension and the content feature dimension. The feature value set for the graphic position dimension in the style feature vector encompasses all specific values ​​related to the graphic position within that dimension, including the values ​​of the graphic's horizontal and vertical position ranges within the image. The feature value set for the subject element position dimension in the content feature vector contains the values ​​of the subject element's (e.g., waves, fruit) horizontal and vertical coordinate ranges within the material. These values ​​together constitute two sets of feature values ​​that can be used for association calculations.

[0077] Step S12343: Use a feature association algorithm to perform item-by-item association calculation on the two sets of feature values. The feature association algorithm transforms the two sets of feature values ​​into a set of associated feature values ​​by calculating the association coefficient between corresponding feature items. The association coefficient is used to represent the degree of fit between the two sets of feature items.

[0078] A feature association algorithm is used to perform item-by-item association calculations on the two extracted feature value sets. For the value of the graphic horizontal position range sub-item in the style feature dimension and the value of the main element horizontal coordinate range sub-item in the content feature dimension, a correlation coefficient is calculated between them. This correlation coefficient reflects the degree of fit between the horizontal position of the main element and the horizontal position of the style graphic. Similarly, a correlation coefficient is calculated between the graphic vertical position range sub-item and the main element vertical coordinate range sub-item. These correlation coefficients and the corresponding feature item values ​​are integrated to transform them into a new set of associated feature values. Each value in this set comprehensively reflects the degree of association between the style requirements and content features of the corresponding feature item.

[0079] Step S1235: Perform channel dimension expansion processing on the dimension-related feature matrix, supplement the preset association weight parameters of the feature interaction layer, so that the number of channels of the dimension-related feature matrix matches the number of input channels of the subsequent image generation layer of the generative artificial intelligence model, and obtain the fused feature matrix.

[0080] After generating the dimensional correlation feature matrix, it needs to undergo channel dimension expansion processing. The feature interaction layer pre-sets correlation weight parameters corresponding to each correlation dimension pair. These parameters are determined based on a large number of training samples and are used to adjust the importance of different correlation dimension pairs in the overall feature fusion. The pre-set correlation weight parameters are then added to the dimensional correlation feature matrix according to the corresponding correlation dimension pairs, increasing the number of channels in the matrix. After expansion processing, the number of channels in the dimensional correlation feature matrix is ​​adjusted to be completely consistent with the number of input channels in the subsequent image generation layer of the generative artificial intelligence model, ultimately forming a fused feature matrix. This fused feature matrix fully carries the correlation information and weight adjustment information between style features and content features.

[0081] Step S124: The fused feature matrix is ​​visualized by the image generation layer of the generative artificial intelligence model to generate an initial style transfer image. The image generation layer converts the fused feature matrix into a visual image by using a layer-by-layer feature decoding method.

[0082] After receiving the fused feature matrix, the image generation layer of the generative AI model initiates the visualization transformation process. The image generation layer consists of multiple decoding sub-layers, processing the fused feature matrix using a layer-by-layer feature decoding approach. First, the first decoding sub-layer extracts and performs preliminary transformations on the basic features of the fused feature matrix, generating a low-dimensional feature map containing the basic color and contour information of the image. Subsequent decoding sub-layers refine the feature map output from the previous layer, gradually increasing the spatial dimension and detail information of the features, supplementing visual elements such as texture and layering. After layer-by-layer processing by all decoding sub-layers, the fused feature matrix is ​​finally transformed into an initial style-transferred image with complete visual information. This initial style-transferred image initially presents the visual effect of the fusion of style and content features.

[0083] Step S125: Based on the parameters corresponding to the brand visual identity features in the advertising style guidance parameters, perform brand element implantation processing on the initial style transfer image, adjust the visual details related to the brand visual identity features in the initial style transfer image, and obtain the style transfer base image.

[0084] Based on the brand visual identity characteristics parameters in the advertising style guidance parameters, brand element integration was implemented on the initial style migration screen. For the brand's exclusive color system, the areas in the image related to the brand's primary color, mint green, and secondary color, light white, were fine-tuned to ensure that the distribution and presentation of these colors conformed to the brand's color specifications. For the brand's iconic graphic element, the leaf shape, its outline curves and vein distribution characteristics were used to integrate the graphic into predetermined locations (such as corners or edges) on the image, and the size and transparency of the graphic were adjusted to blend naturally with the overall style. Simultaneously, any text reservation areas in the image were checked, and a preliminary font style definition was performed according to the characteristics of the brand's font style, completing the integration of brand elements. After these visual detail adjustments, the initial style migration screen was optimized into a basic style migration screen, which reflects the visual style preferred by the target audience while incorporating the brand's core visual identity.

[0085] Step S130: Adapt the core elements of the advertisement based on the style transfer base image to obtain an element-integrated advertisement image. The core elements of the advertisement include visual images of the advertised product and advertising text information. The core elements of the advertisement in the element-integrated advertisement image maintain the same visual style as the style transfer base image.

[0086] After obtaining the base visuals for style transfer, the adaptation and integration of the core advertising elements began. These core elements include product visuals of the new summer tea drinks (such as beverage cups and fruit ingredients) and advertising text (such as product names and promotional slogans). During the adaptation process, it was crucial to ensure that the visual style of the product visuals and promotional text remained highly consistent with the fresh and natural style, color scheme, composition, and texture of the base visuals for style transfer, avoiding any visual clashes. Through style adjustments and optimized placement of the core elements, they were organically integrated into the base visuals for style transfer, ultimately forming an integrated advertising image.

[0087] Step S131: Extract the visual style attributes of the style transfer base image. The visual style attributes include color distribution rules, composition ratio relationship and texture presentation method. The visual style attributes are used to guide the adaptation direction of the core elements of the advertisement.

[0088] First, a comprehensive extraction of the visual style attributes from the base image for style transfer was conducted. Image analysis tools were used to statistically analyze the distribution of each color in the image, determining the distribution range and proportion of the primary colors blue and green, secondary colors white and light gray, and the brand color mint green in different areas of the image, thus forming color distribution pattern data. The compositional structure of the image was analyzed, measuring parameters such as the size ratio, positional spacing of major geometric shapes, and the golden ratio of the overall image, summarizing the compositional proportions. Simultaneously, the texture features of the image were sampled and analyzed to determine the texture presentation methods, such as the particle distribution density and size variation patterns of the frosted texture. These extracted visual style attributes will serve as specific guidance for the adaptation direction of the core elements of the advertisement.

[0089] Step S132: Adjust the style of the advertising product visual image in the core elements of the advertisement so that the color distribution pattern of the advertising product visual image is consistent with the color distribution pattern of the style transfer base image, the composition ratio is coordinated with the composition ratio of the style transfer base image, and the texture presentation method is unified with the texture presentation method of the style transfer base image, so as to obtain the style-adapted product image.

[0090] A style adaptation adjustment process was initiated for the visual images of the advertising products. Regarding color, based on the color distribution patterns of the base image for style transfer, the colors of the beverage cups and fruit ingredients in the product images were adjusted to harmonize the product's primary color with the main blue and green tones of the image, and to match the product's secondary colors with the secondary white and light gray tones of the image. Simultaneously, the brand's mint green was appropriately incorporated to ensure that the distribution of product colors within the image conformed to the overall pattern. Regarding composition, based on the proportions of the image, the size and placement of the product images were adjusted to ensure that the spacing and proportions between the product images and surrounding geometric shapes conformed to the overall compositional guidelines. Regarding texture, image texture transfer technology was used to apply a matte texture to the surface of the product images, ensuring that the texture presentation of the products remained consistent with the image, ultimately resulting in style-adapted product images.

[0091] Step S1321: Extract the color distribution pattern of the style transfer base image, and use the color analysis module of the generative artificial intelligence model to count the pixel proportion of each color type in the style transfer base image, determine the pixel proportion of the main color, the pixel proportion of the auxiliary color, and the pixel proportion of the accent color, and form color proportion distribution data.

[0092] The color analysis module of the generative AI model performs pixel-level scanning analysis on the basic image for style transfer. It counts the number of pixels belonging to the primary colors blue and green, calculates their proportion in the total number of pixels, and obtains the primary color pixel percentage. Similarly, it counts the number and proportion of pixels belonging to the secondary colors white and light gray, determining the secondary color pixel percentage. For colors that serve as accents, such as the brand color mint green, their pixel count and proportion are counted, representing the accent color pixel percentage. This percentage data is then categorized and organized according to color type to form complete color percentage distribution data, clearly reflecting the color composition of the image.

[0093] Step S1322: Extract the spatial distribution pattern of colors in the style transfer base image, determine the main distribution area of ​​the primary color in the image, the transitional distribution area of ​​the auxiliary color in the image, and the key distribution area of ​​the accent color in the image, and form color spatial distribution data.

[0094] Based on the color proportion distribution data, the spatial distribution patterns of colors are further extracted. Image segmentation technology is used to divide the style transfer base image into multiple regions, and the distribution of the primary color, secondary color, and accent color in each region is analyzed. The main areas occupied by the primary colors blue and green are identified, as these areas are usually the visual center of the image; the areas where the secondary colors white and light gray are distributed as transition colors are identified, as these areas connect and balance the primary color areas; and the key distribution areas of the accent color mint green are located, as these areas are often where the brand logo needs to be highlighted in the image. The boundary coordinates and range information of the above regions are compiled into color space distribution data, providing a positional reference for color space adjustments of product images.

[0095] Step S1323: Input the visual image of the advertising product into the product color adjustment module of the generative artificial intelligence model, and extract the original color distribution data of the visual image of the advertising product. The original color distribution data includes the original main color, the original auxiliary color and the original color ratio of the product.

[0096] The visual image of the advertised product is input into the product color adjustment module of the generative artificial intelligence model. This module performs color analysis on the product image, identifying the main colors in the image and determining the original primary color (e.g., the original color of a beverage cup) and original secondary colors (e.g., the colors of the pattern on the cup). Pixel statistics are used to calculate the pixel percentage of the original primary and secondary colors in the product image, forming original color percentage data. The original primary color, primary secondary color, and original color percentages are then integrated into original color distribution data, fully presenting the initial color state of the product image.

[0097] Step S1324: Based on the color proportion distribution data of the style transfer base image, adjust the original color proportion of the advertising product visual image so that the main color proportion and auxiliary color proportion of the advertising product visual image are consistent with the main color proportion and auxiliary color proportion of the style transfer base image, and obtain a color proportion adapted product image.

[0098] The product color adjustment module adjusts the original color proportions of the product image based on the color proportion distribution data of the style transfer base image. If the proportion of the product's original primary color is higher than that of the image's primary color, its pixel proportion is reduced by decreasing the saturation of the original primary color or shrinking its area in the image. If the proportion of the original primary color is lower than that of the image's primary color, its proportion is increased by expanding the primary color area or increasing its saturation. The same adjustment method is used for the secondary color proportions to match the secondary color proportions of the product image with those of the image. After adjustment, the color proportions of the product image are consistent with the color composition ratios of the style transfer base image, generating a color proportion-adapted product image.

[0099] Step S1325: Based on the color space distribution data of the style transfer base image, adjust the spatial distribution position of each color in the color proportion-adapted product image so that the distribution area of ​​the main color in the advertising product visual image is coordinated with the main distribution area of ​​the main color in the style transfer base image, and the distribution area of ​​the auxiliary color in the advertising product visual image corresponds to the transitional distribution area of ​​the auxiliary color in the style transfer base image, thus obtaining a color distribution-adapted product image.

[0100] Referring to the color space distribution data of the base image for style transfer, the color space distribution of the product image is adjusted to match the color proportions. The distribution area of ​​the main color in the product image is adjusted to correspond to the main distribution area of ​​the main color in the image. For example, if the main color is mainly distributed in the upper half of the image, the main color area of ​​the product is concentrated in the upper half of the product image to achieve harmony with the main color area of ​​the image. For the product's secondary color area, it is adjusted to correspond to the transitional distribution area of ​​secondary colors in the image, ensuring that the secondary colors play a connecting and transitional role in space. Through the above spatial distribution adjustments, the color distribution of the product image and the spatial layout of the image's colors form an organic whole, resulting in a product image with color distribution adaptation.

[0101] Step S133: Visually adjust the advertising text information in the core elements of the advertisement. Determine the font type of the advertising text information according to the font style tendency of the style transfer base image. Determine the font color of the advertising text information according to the color distribution law of the style transfer base image. Determine the placement position of the advertising text information according to the composition ratio of the style transfer base image to obtain visually adapted text information.

[0102] For the advertising text, visual style adjustments were made. The implicit font style tendencies in the base image were analyzed. Given the overall minimalist and clean style of the image, the font type for the advertising text was determined to be a rounded sans-serif font consistent with the brand font style, ensuring consistency between the font style and the overall image style. Based on the color distribution patterns of the image, a color that contrasts sharply with the background color and falls within the primary or secondary color range was selected as the font color. For example, white font was used in blue background areas, and green font in white background areas, ensuring readability. Considering the compositional proportions of the image, the text information was placed in a position conforming to the golden ratio, such as the upper third or right third of the image, to harmonize the text position with the overall composition, ultimately creating visually appropriate text information.

[0103] Step S134: Import the style-adapted product image into the preset element placement area of ​​the style transfer base image, adjust the size of the style-adapted product image to blend the overall visual effect of the style-adapted product image with the style transfer base image, and obtain the product embedding image.

[0104] In the style transfer base image, a pre-defined element placement area is established. This area is determined based on composition and layout parameters, suitable for placing product images without disrupting the overall balance of the image. Style-adapted product images are imported into this pre-defined area, and their size is adjusted using a scaling algorithm to ensure the height and width of the product image match the proportions of the placement area. This also ensures the product image occupies a suitable proportion of the image, clearly showcasing product details without appearing obtrusive. During the adjustment process, the visual effects of the product image and surrounding elements are compared in real-time to ensure a natural transition in color and texture with the surrounding environment, achieving overall visual integration with the style transfer base image and generating the product placement image.

[0105] Step S135: Overlay the visually adapted text information onto the text reserved area of ​​the product embedded screen, adjust the transparency and font size of the visually adapted text information, and obtain the element integrated advertising screen.

[0106] The product placement image features a dedicated text area, pre-defined in the design for placing advertising text. Visually appropriate text is overlaid on this area, and by adjusting the text's transparency, a clear distinction is maintained between the text and the background without creating a strong visual disconnect. Furthermore, the font size is fine-tuned based on the size of the text area and the overall proportions of the image to ensure the text is fully displayed and occupies an appropriate visual proportion. After adjusting transparency and font size, the text seamlessly integrates with the product placement image, creating a unified advertising visual.

[0107] Step S140: Combine the advertising push scenario information to visually adapt and adjust the integrated advertising screen of the elements to obtain the push-adapted advertising screen. The advertising push scenario information includes the display characteristics of the push terminal and the ambient light adaptation requirements of the push scenario. The push-adapted advertising screen conforms to the display specifications of the push terminal.

[0108] After integrating the elements into the advertisement visuals, visual adaptation adjustments are needed based on the advertisement push scenario information. This scenario information is obtained through a scenario analysis module. The display characteristics of the push terminal include parameters such as screen size, resolution, and color display mode. Ambient light adaptation requirements for the push scenario are determined based on the scene's light intensity, such as the brightness requirements for different scenarios like low-light indoor environments and high-light outdoor environments. Based on this information, the integrated element advertisement visuals are adjusted to ensure optimal display on the target push terminal and adapt to viewing needs under different ambient light conditions, ultimately resulting in a push-adapted advertisement visual that conforms to the push terminal's display specifications.

[0109] Step S141: Analyze the display characteristics of the push terminal in the advertising push scenario information. The display characteristics of the push terminal include the display resolution specification and the display color mode. The display resolution specification is used to determine the pixel dimension of the image, and the display color mode is used to determine the color representation of the image.

[0110] The display characteristics of the push terminal in the advertising push scenario information are analyzed and processed. The model and technical parameters of the push terminal are obtained through the terminal information collection interface, from which the display resolution specification is extracted. This display resolution specification consists of the number of horizontal and vertical pixels, used to determine the pixel dimensions that the image needs to be adjusted to. Simultaneously, the terminal's display color mode is determined, such as RGB mode, CMYK mode, or other specific color modes. Different color modes determine the color representation and color gamut of the image. The display resolution specification and display color mode are then organized into structured terminal display characteristic data.

[0111] Step S142: Adjust the pixel dimension of the element-integrated advertisement screen according to the display resolution specification. Expand or compress the number of pixels of the element-integrated advertisement screen through the resolution adaptation module of the generative artificial intelligence model so that the pixel dimension of the element-integrated advertisement screen matches the display resolution specification, and obtain a resolution-adapted screen.

[0112] After receiving the display resolution specifications, the resolution adaptation module of the generative AI model adjusts the pixel dimensions of the integrated advertising image. If the original resolution of the integrated advertising image is lower than the display resolution specifications, the resolution adaptation module uses a pixel interpolation algorithm to expand the image pixels. By inserting calculated transition pixels between existing pixels, the horizontal and vertical pixel count of the image is increased, raising the resolution to the display specification requirements. If the original resolution is higher than the display resolution specifications, a pixel compression algorithm is used to selectively merge adjacent pixels or reduce pixel sampling points, reducing the pixel count to match the display specifications. After adjustment, the pixel dimensions of the image perfectly match the display resolution specifications of the push terminal, generating a resolution-adapted image.

[0113] Step S143: Perform color mode conversion on the resolution-adapted screen according to the display color mode, adjust the color parameters of each pixel in the resolution-adapted screen, so that the color representation of the resolution-adapted screen is consistent with the display color mode, and obtain the color-adapted screen.

[0114] After the resolution-adapted image is generated, a color mode conversion is required based on the display color mode. First, the original color mode of the current resolution-adapted image is analyzed. If the original mode is RGB mode while the terminal display color mode is CMYK mode, the color mode conversion engine needs to be activated. During the conversion process, the RGB color parameters of each pixel in the image are calculated according to preset color space conversion rules, converting the parameter values ​​of the red, green, and blue channels into the parameter values ​​of the cyan, magenta, yellow, and black channels. For any color gamut deviations that may occur during the conversion process, a color compensation algorithm is used to fine-tune color areas with large deviations, ensuring that the converted colors are visually consistent with the original image. If the terminal display color mode is another specific mode, the corresponding conversion rules are used to process it, ultimately ensuring that the color representation of the image perfectly matches the display color mode, generating a color-adapted image.

[0115] Step S144: Analyze the ambient light adaptation requirements in the ad push scene information. The ambient light adaptation requirements include the screen brightness adjustment direction corresponding to the ambient light intensity. The higher the ambient light intensity, the more the screen brightness adjustment direction is biased towards increasing brightness. The lower the ambient light intensity, the more the screen brightness adjustment direction is biased towards decreasing brightness.

[0116] This paper analyzes the ambient light adaptation requirements in the advertising push scenario. Ambient light intensity data of the push scenario is collected using an ambient light sensor, reflecting the brightness of the light within the scene. The ambient light intensity data is divided into multiple intervals, each corresponding to a different brightness adjustment direction. When the ambient light intensity is in a high interval, it indicates sufficient scene lighting, and the brightness adjustment direction is set to increase brightness to avoid the image appearing dark under strong light. When the ambient light intensity is in a low interval, it indicates dim scene lighting, and the brightness adjustment direction is set to decrease brightness to prevent the image from being too bright and causing visual stimulation to the viewer. The ambient light intensity data and the corresponding brightness adjustment direction are compiled into ambient light adaptation requirement parameters.

[0117] Step S145: Adjust the brightness parameter of the color-adapted image according to the ambient light adaptation requirements, and simultaneously adjust the contrast parameter of the color-adapted image to obtain the push-adapted advertisement image.

[0118] After obtaining the ambient light adaptation requirements, the brightness and contrast parameters of the color-adapted image are adjusted. First, the target brightness range of the image is determined based on the brightness adjustment direction corresponding to the ambient light intensity. If the adjustment direction is to increase brightness, the brightness values ​​of each pixel in the image are gradually increased to bring the overall brightness of the image closer to the target range; if the adjustment direction is to decrease brightness, the brightness values ​​of each pixel are decreased. During the brightness adjustment process, the contrast of the image is monitored simultaneously, and the current contrast is determined by calculating the brightness difference between the brightest and darkest areas of the image. When brightness is increased, the contrast is appropriately increased to make the light and dark levels in the image clearer; when brightness is decreased, the contrast is appropriately decreased to avoid losing details in the dark areas of the image. After multiple iterations of adjustments, the brightness and contrast of the image are made to meet the ambient light adaptation requirements, and the final push-adapted advertisement image is generated.

[0119] For example, step S1451: convert the ambient light adaptation requirements in the advertising push scene information into brightness adjustment parameters, wherein the brightness adjustment parameters include a target brightness range, and the target brightness range is determined according to the ambient light intensity.

[0120] The ambient light adaptation requirements are translated into specific brightness adjustment parameters. A preset mapping table is consulted based on the ambient light intensity data. This table records the target brightness range corresponding to different ambient light intensities. For example, when the ambient light intensity is in a relatively high range, the corresponding target brightness range is a relatively high brightness value range; when the ambient light intensity is in a relatively low range, the target brightness range is a relatively low brightness value range. The retrieved target brightness range is used as the brightness adjustment parameter to determine the required range of screen brightness.

[0121] Step S1452: Call the brightness adjustment module of the generative artificial intelligence model, extract the current brightness value of the color-adapted image, calculate the average brightness of the color-adapted image by pixel-by-pixel statistics, and determine whether the current brightness value is within the target brightness range.

[0122] The brightness adjustment module of the generative artificial intelligence model is invoked to process the color-adapted image. The brightness adjustment module extracts the brightness value of each pixel in the image by scanning pixel by pixel, then adds up the brightness values ​​of all pixels and divides by the total number of pixels to calculate the average brightness of the image, i.e., the current brightness value. The current brightness value is compared with the target brightness range to determine if it falls within the target range. If the current brightness value is within the target range, no brightness adjustment is needed; if it is below the lower limit or above the upper limit of the target range, appropriate brightness adjustment is required.

[0123] Step S1453: If the current brightness value is lower than the lower limit of the target brightness range, the brightness value of each pixel in the color-adapted image is increased by the brightness adjustment module. The increase is determined by the difference between the lower limit of the target brightness range and the current brightness value, while keeping the relative brightness relationship between each pixel unchanged.

[0124] When the current brightness value is lower than the lower limit of the target brightness range, the brightness adjustment module initiates a brightness enhancement procedure. First, it calculates the difference between the lower limit of the target brightness range and the current brightness value, determining the overall brightness enhancement amount based on this difference. Then, it increases the brightness value of each pixel in the image according to this enhancement amount. During this process, it ensures that the brightness ratio between pixels remains unchanged; that is, previously brighter pixels remain relatively bright, and previously darker pixels remain relatively dark, only the overall brightness level is improved. Through this method, the average brightness of the image reaches or exceeds the lower limit of the target brightness range.

[0125] Step S1454: If the current brightness value is higher than the upper limit of the target brightness range, the brightness value of each pixel in the color adaptation image is reduced by the brightness adjustment module. The reduction range is determined according to the difference between the current brightness value and the upper limit of the target brightness range, while keeping the relative brightness relationship between each pixel unchanged.

[0126] When the current brightness value is higher than the upper limit of the target brightness range, the brightness adjustment module initiates a brightness reduction procedure. It calculates the difference between the current brightness value and the upper limit of the target brightness range, and determines the brightness reduction amount based on this difference. The brightness value of each pixel is then reduced according to this reduction amount, while maintaining the relative brightness relationship between pixels. Through this adjustment, the average brightness of the image is reduced to below the upper limit of the target brightness range, preventing the image from becoming too bright.

[0127] Step S1455: During the process of adjusting the brightness parameters, the current contrast value of the color-adapted image is extracted simultaneously. The current contrast value is calculated by the brightness difference between the brightest and darkest pixels in the image. The contrast parameters are adjusted synchronously according to the brightness adjustment range. When the brightness is increased, the contrast value is increased accordingly, and when the brightness is decreased, the contrast value is decreased accordingly. Finally, a push-adapted advertisement image with both brightness and contrast adapted to the ambient light requirements is obtained.

[0128] While adjusting the brightness parameters, the brightness adjustment module simultaneously extracts the current contrast value of the image. It scans the image to find the brightness values ​​of the brightest and darkest pixels, and uses the difference between them as the current contrast value. A contrast adjustment strategy is then formulated based on the brightness adjustment magnitude: when the brightness increase is significant, the contrast value is increased accordingly, enhancing the image's depth by increasing the brightness difference between the brightest and darkest pixels; conversely, when the brightness decreases significantly, the contrast value is appropriately decreased to reduce the brightness difference and prevent blurring of details in dark areas. Through this coordinated adjustment of brightness and contrast, the final push notification ad image presents a good visual effect under various ambient lighting conditions.

[0129] Step S150: Output the push-adapted advertising screen to the advertising push system, and the advertising push system pushes the push-adapted advertising screen to the terminal devices of the target audience.

[0130] After generating the push-adaptive ad screen, the screen is encapsulated in a preset format, such as JPEG or PNG, to ensure the integrity of the image data and compliance with the ad push system's receiving requirements. The encapsulated push-adaptive ad screen is then transmitted to the ad push system via a data transmission interface. Upon receiving the image data, the ad push system selects an appropriate push channel based on the target audience's terminal device information and sends the image data to the target audience's terminal devices. The terminal devices receive the image data, decode and render it, and finally display the push-adaptive ad screen on the screen, completing the entire process of ad push screen generation and delivery.

[0131] Figure 2 The illustration shows exemplary hardware and software components of a style transfer-based ad push screen generation system 100 that can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the style transfer-based ad push screen generation system 100 and to perform the functions in this application.

[0132] For example, the style transfer-based ad push screen generation system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the style transfer-based ad push screen generation system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The style transfer-based ad push screen generation system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0133] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-described method for generating advertising push screens based on style transfer is implemented.

[0134] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for generating ad push screens based on style transfer, characterized in that, The method includes: The advertising style requirements are analyzed to obtain advertising style guidance parameters. The advertising style requirements include the visual style type preferred by the advertising target audience and the visual identity features of the advertising brand. The advertising style guidance parameters are used to define the style transfer direction of the generative artificial intelligence model. The pre-trained generative artificial intelligence model is driven by the advertising style guidance parameters to perform style transfer processing to obtain the style transfer base image. The generative artificial intelligence model includes a style feature conversion module and a content feature fusion module. The style transfer base image carries the visual style corresponding to the advertising style guidance parameters. Based on the style transfer base image, the core elements of the advertisement are adapted to obtain the element-integrated advertisement image. The core elements of the advertisement include the visual image of the advertised product and the advertising text information. The visual style of the core elements of the advertisement in the element-integrated advertisement image is consistent with that of the style transfer base image. By combining the advertising push scenario information, the elements are integrated into the advertising screen for visual adaptation and adjustment to obtain the push-adapted advertising screen. The advertising push scenario information includes the display characteristics of the push terminal and the ambient light adaptation requirements of the push scenario. The push-adapted advertising screen conforms to the display specifications of the push terminal. The push-adapted advertising screen is output to the advertising push system, which then pushes the push-adapted advertising screen to the terminal devices of the target audience. The step of visually adapting and adjusting the integrated advertising screen based on the advertising push scenario information to obtain a push-adapted advertising screen includes: The display characteristics of the push terminal in the advertising push scenario information are analyzed. The display characteristics of the push terminal include the display resolution specification and the display color mode. The display resolution specification is used to determine the pixel dimension of the image, and the display color mode is used to determine the color representation of the image. The pixel dimension of the integrated element advertisement is adjusted according to the display resolution specification. The resolution adaptation module of the generative artificial intelligence model expands or compresses the number of pixels of the integrated element advertisement so that the pixel dimension of the integrated element advertisement matches the display resolution specification, thus obtaining a resolution-adapted image. According to the display color mode, the color mode of the resolution-adapted image is converted, and the color parameters of each pixel in the resolution-adapted image are adjusted so that the color representation of the resolution-adapted image is consistent with the display color mode, thus obtaining the color-adapted image. The ambient light adaptation requirements in the advertising push scenario information are analyzed. The ambient light adaptation requirements include the screen brightness adjustment direction corresponding to the ambient light intensity. The higher the ambient light intensity, the more the screen brightness adjustment direction is biased towards increasing the brightness. The lower the ambient light intensity, the more the screen brightness adjustment direction is biased towards decreasing the brightness. The brightness parameters of the color-adapted image are adjusted according to the ambient light adaptation requirements, and the contrast parameters of the color-adapted image are adjusted simultaneously to obtain the push-adapted advertisement image.

2. The method for generating ad push screens based on style transfer according to claim 1, characterized in that, The process of analyzing advertising style requirements yields advertising style guidance parameters, including: The visual style types in the advertising style requirements are broken down into color tone dimension, composition structure dimension and texture dimension, and each dimension corresponds to the specific visual expression form preferred by the advertising target audience. Extract the visual identity features of the advertising brand from the advertising style requirements, and separate the brand-specific color system, brand-identical graphic elements and brand font styles. The brand-specific color system includes the matching relationship between the brand's primary color and secondary colors. For the aforementioned color tone dimension, the color expression preferred by the target audience of the advertisement is mapped to color space parameters that can be recognized by the generative artificial intelligence model. The color space parameters are used to define the direction of color adjustment during the style transfer process. For the aforementioned compositional structure dimension, the compositional representation preferred by the advertising target audience is transformed into compositional layout parameters that can be executed by the generative artificial intelligence model. These compositional layout parameters are used to standardize the arrangement of elements during the style transfer process. For the texture dimension, the texture representation preferred by the target audience of the advertisement is converted into texture feature parameters that can be processed by the generative artificial intelligence model. The texture feature parameters are used to control the surface texture presentation during the style transfer process. The color space parameters, composition layout parameters, texture feature parameters, and brand visual identity features are integrated to form the brand-specific color system, brand-identical graphic elements, and brand font styles corresponding to the brand, forming the advertising style guidance parameters. Each dimension of the advertising style guidance parameters corresponds one-to-one with the style transfer module of the generative artificial intelligence model.

3. The method for generating ad push screens based on style transfer according to claim 2, characterized in that, The process of mapping the color representation preferred by the target audience of the advertisement to color space parameters that can be recognized by the generative artificial intelligence model, for the aforementioned color tone dimension, includes: Extract the dominant color type from the color expression forms preferred by the target audience of the advertisement. The dominant color type includes warm color series or cool color series, and each color series corresponds to multiple specific color types. Determine the color range parameters corresponding to the main color type. The color range parameters include the hue range and saturation range of the main color. The hue range is used to limit the color bias of the main color, and the saturation range is used to limit the color intensity of the main color. Extract the auxiliary color type from the color expression forms preferred by the target audience of the advertisement. The auxiliary color type is used to match the main color type to enhance the visual hierarchy of the image. Determine the color matching ratio parameter corresponding to the auxiliary color type. The color matching ratio parameter is used to define the proportion of the auxiliary color in the overall picture color. This proportion needs to form a visual balance with the main color. The color range parameters of the primary color tone and the color matching ratio parameters of the secondary color tone are integrated and transformed into a numerical parameter form that can be recognized by the color processing module of the generative artificial intelligence model. The numerical parameter form is matched with the color feature input format of the generative artificial intelligence model to obtain the color space parameters.

4. The method for generating ad push screens based on style transfer according to claim 1, characterized in that, The style transfer processing is performed by a pre-trained generative artificial intelligence model driven by the advertising style guidance parameters to obtain a style transfer base image, including: The advertising style guidance parameters are input into the style feature conversion module of the generative artificial intelligence model. The style feature conversion module performs feature encoding on the color space parameters, composition layout parameters and texture feature parameters in the advertising style guidance parameters to obtain a style feature vector. The dimension of the style feature vector matches the feature processing channel of the generative artificial intelligence model. The content feature fusion module of the generative artificial intelligence model is invoked to import basic advertising content materials. The content feature fusion module is used to extract content features from the basic advertising content materials to obtain a content feature vector. The basic advertising content materials contain general visual elements related to the advertising theme. In the feature interaction layer of the generative artificial intelligence model, the style feature vector and the content feature vector are cross-dimensionally associated and fused to generate a fused feature matrix, which simultaneously carries the association information of style features and content features. The fused feature matrix is ​​visualized by the image generation layer of the generative artificial intelligence model to generate an initial style transfer image. The image generation layer uses a layer-by-layer feature decoding method to convert the fused feature matrix into a visual image. Based on the parameters corresponding to the brand visual identity features in the advertising style guidance parameters, the initial style transfer image is processed by embedding brand elements, and the visual details related to the brand visual identity features in the initial style transfer image are adjusted to obtain the basic style transfer image.

5. The method for generating ad push screens based on style transfer according to claim 4, characterized in that, In the feature interaction layer of the generative artificial intelligence model, the style feature vector and the content feature vector are cross-dimensionally correlated and fused to generate a fused feature matrix, including: Determine the number of feature dimensions of the style feature vector and the feature meaning of each dimension. The feature meaning includes color feature dimension, composition feature dimension and texture feature dimension, and each feature dimension corresponds to multiple feature sub-items. Determine the number of feature dimensions of the content feature vector and the feature meaning corresponding to each dimension. The feature meaning includes the content subject feature dimension, the background element feature dimension, and the spatial relationship feature dimension. Each feature dimension corresponds to multiple feature sub-items. In the feature interaction layer, a dimension association mapping table is constructed. The dimension association mapping table is used to establish the corresponding association relationship between each dimension of the style feature vector and each dimension of the content feature vector. Based on the dimensional association mapping table, the style feature vector and the content feature vector are subjected to dimensional feature value association calculation to generate a dimensional association feature matrix. Each element in the dimensional association feature matrix represents a combination of feature values ​​of a set of associated dimensions. The dimensional correlation feature matrix is ​​subjected to channel dimension expansion processing, and the preset correlation weight parameters of the feature interaction layer are added to make the number of channels of the dimensional correlation feature matrix match the number of input channels of the subsequent image generation layer of the generative artificial intelligence model, so as to obtain the fused feature matrix.

6. The method for generating ad push screens based on style transfer according to claim 5, characterized in that, The step of performing a dimension-wise feature value association calculation on the style feature vector and content feature vector based on the dimensional association mapping table to generate a dimensional association feature matrix includes: Extract the first set of associated dimension pairs from the dimension association mapping table. The associated dimension pair contains a feature dimension of the style feature vector and a feature dimension of the content feature vector. Extract the feature value set of the style feature dimension and the feature value set of the content feature dimension in the first set of associated dimensions. The feature value set contains the specific values ​​of all feature sub-items under the content feature dimension. A feature association algorithm is used to perform item-by-item association calculations on two sets of feature values. The feature association algorithm transforms two sets of feature values ​​into a set of associated feature values ​​by calculating the association coefficient between corresponding feature items. The association coefficient is used to represent the degree of fit between the two sets of feature items. The set of associated feature values ​​is used as a row of data in the dimension-associated feature matrix, and the number of columns in this row is consistent with the number of feature items. Repeat the above steps to extract other related dimension pairs from the dimension association mapping table, perform feature value association calculations on each pair, generate the corresponding set of related feature values, and use them as other row data in the dimension association feature matrix until all related dimension pairs have been processed, and finally form a complete dimension association feature matrix.

7. The method for generating ad push screens based on style transfer according to claim 1, characterized in that, The process of adapting the core elements of the advertisement based on the style transfer to obtain an integrated advertisement image includes: The visual style attributes of the base image for style transfer are extracted. These visual style attributes include color distribution patterns, compositional proportions, and texture presentation methods. These visual style attributes are used to guide the adaptation direction of the core elements of the advertisement. Style adaptation adjustments are made to the visual images of the advertising products in the core elements of the advertisement, so that the color distribution pattern of the visual images of the advertising products is consistent with the color distribution pattern of the base image of the style transfer, the composition ratio is coordinated with the composition ratio of the base image of the style transfer, and the texture presentation method is unified with the texture presentation method of the base image of the style transfer, thus obtaining a style-adapted product image. The advertising text information in the core elements of the advertisement is visually styled. The font type of the advertising text information is determined according to the font style tendency of the style transfer base image. The font color of the advertising text information is determined according to the color distribution law of the style transfer base image. The placement of the advertising text information is determined according to the composition ratio of the style transfer base image, so as to obtain visually adapted text information. The style-adapted product image is imported into the preset element placement area of ​​the style transfer base image. The size of the style-adapted product image is adjusted to blend the overall visual effect of the style-adapted product image with the style transfer base image, resulting in a product placement image. The visually adapted text information is overlaid onto the text reserved area of ​​the product embedded image, and the transparency and font size of the visually adapted text information are adjusted to obtain the element integrated advertising image.

8. The method for generating ad push screens based on style transfer according to claim 7, characterized in that, The style adaptation and adjustment of the visual images of the advertised products in the core elements of the advertisement includes: The color distribution pattern of the style transfer base image is extracted, and the pixel proportion of each color type in the style transfer base image is statistically analyzed through the color analysis module of the generative artificial intelligence model to determine the pixel proportion of the main color, the pixel proportion of the auxiliary color, and the pixel proportion of the accent color, thus forming color proportion distribution data. Extract the spatial distribution pattern of colors in the style transfer base image, determine the main distribution area of ​​the main color in the image, the transitional distribution area of ​​the auxiliary color in the image, and the key distribution area of ​​the accent color in the image, and form color spatial distribution data; The product color adjustment module of the generative artificial intelligence model inputs the visual image of the advertising product to extract the original color distribution data of the visual image of the advertising product. The original color distribution data includes the original main color, the original auxiliary color and the original color ratio of the product. Based on the color proportion distribution data of the style transfer base image, adjust the original color proportion of the advertising product visual image so that the main color proportion and auxiliary color proportion of the advertising product visual image are consistent with the main color proportion and auxiliary color proportion of the style transfer base image, and obtain a color proportion adapted product image. Based on the color space distribution data of the style transfer base image, the spatial distribution position of each color in the color proportion-adapted product image is adjusted so that the distribution area of ​​the main color in the advertising product visual image is coordinated with the main distribution area of ​​the main color in the style transfer base image, and the distribution area of ​​the auxiliary color in the advertising product visual image corresponds to the transitional distribution area of ​​the auxiliary color in the style transfer base image, thus obtaining a color distribution-adapted product image.

9. A style transfer-based advertising push screen generation system, characterized in that, The style transfer-based ad push screen generation system includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the style transfer-based ad push screen generation method according to any one of claims 1-8.

Images