Advertisement image generation method and device, computer equipment and storage medium

Abstract

The invention discloses an advertisement image generation method and device, computer equipment and a storage medium. The method comprises the following steps: receiving text prompt information input by a user and a specified material type; performing semantic extraction on the text prompt information to obtain semantic features; encoding the semantic feature to obtain an encoding vector; and inputting the coding vector and the specified material type into the image generation model to generate an advertisement image. According to the method, a material sensing mechanism is introduced, the specified material type and the coding vector are combined and input into the image generation model, the expressive force of the material can be accurately controlled in the image generation process, fine and smooth material effects such as a metal feeling, a frosted texture and a velvet background can be visually presented, and the image generation efficiency is improved. The reality sense and the high-end texture of the image are enhanced, it can be ensured that the generated image can closely meet the personalized requirements and brand tonality of the user, and the technical scheme can be applied to the fields of finance and medical health.

Description

Technical Field

[0001] The present invention relates to the technical field of advertising image generation, and more specifically to an advertising image generation method, apparatus, computer equipment, and storage medium. Background Art

[0002] As the advertising industry continues to evolve, advertising content is shifting towards visuals and emotional appeal. Personalized ad design has become a hot topic and trend. Especially in the financial and healthcare sectors, advertising images must not only convey product information but also evoke emotional resonance through visual elements to enhance brand image and user trust.

[0003] However, current image generation models on the market still have significant shortcomings in personalized advertising design. These models often focus on scene composition or character expression, while neglecting the expressive power of "texture"—a key visual element in images. For example, in financial advertisements, material elements such as metallic cards, high-end business cards with a matte texture, and products with velvet backgrounds often have a crucial impact on users' perception of brand attributes such as "premium," "safety," and "trust." However, existing image generation technologies often struggle to accurately capture and reproduce the details and texture of these materials, resulting in generated advertising images that lack a high-end quality in visual effects and struggle to effectively convey brand value and product characteristics.

[0004] In particular, existing Diffusion models, while demonstrating strong image generation capabilities, lack clear modeling and control mechanisms for common material elements found in advertisements. As a result, when using Diffusion models to generate financial advertising images, the output images often have a disorganized style that struggles to align with brand tone and product strategy. This deficiency not only impacts the visual quality of advertising images but also limits their potential for application in personalized advertising design.

[0005] Therefore, developing a generative advertising image solution that can accurately capture and reproduce the texture of images is of great significance for improving the style consistency and perceived quality of advertisements, thereby generating personalized advertising images that are more in line with brand tone and product strategy. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the prior art and provide a method, device, computer equipment and storage medium for generating an advertising image.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] An advertising image generation method, comprising:

[0009] Receive the text prompt information and specified material type entered by the user;

[0010] Perform semantic extraction on text prompt information to obtain semantic features;

[0011] Encode the semantic features to obtain an encoding vector;

[0012] Input the encoding vector and the specified material type into the image generation model to generate the advertising image.

[0013] The present invention also provides an advertising image generating device, comprising:

[0014] A receiving unit, configured to receive text prompt information and a specified material type input by a user;

[0015] An extraction unit, configured to perform semantic extraction on the text prompt information to obtain semantic features;

[0016] The encoding unit is used to encode the semantic features to obtain an encoding vector;

[0017] The input generation unit is used to input the encoding vector and the specified material type into the image generation model to generate the advertising image.

[0018] The present invention further provides a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0019] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0020] The beneficial effects of the present invention compared with the prior art are: by receiving the text prompt information input by the user and the specified material type, it is ensured that the generated image can closely fit the user's personalized needs and brand tone, and then by performing semantic extraction on the text prompt information, it can accurately understand the user's intention, including the target user group, product positioning, and the emotion or value to be conveyed. At the same time, on the basis of encoding the semantic features, a material perception mechanism is introduced. By combining the specified material type with the encoding vector and inputting it into the raw image model, the expressiveness of the material can be accurately controlled during the image generation process. This means that the generated advertising image can not only accurately convey the text information, but also visually present delicate material effects such as metallic feel, matte texture, velvet background, etc., thereby enhancing the realism and high-end texture of the image.

[0021] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 A schematic diagram of an application scenario of the method for generating an advertisement image provided by an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of a flow chart of a method for generating an advertisement image according to an embodiment of the present invention;

[0025] Figure 3 A schematic block diagram of an advertising image generating device provided by an embodiment of the present invention;

[0026] Figure 4 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0028] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0029] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0030] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0031] See also Figure 1 and Figure 2 , Figure 1A schematic diagram of an application scenario of the advertising image generation method provided by an embodiment of the present invention. Figure 2 A schematic flow chart of an advertising image generation method provided in an embodiment of the present invention. The advertising image generation method is applied to a server that interacts with a terminal for data exchange. By receiving text prompt information and a specified material type input by a user, the server ensures that the generated image closely matches the user's personalized needs and brand tone. Then, by performing semantic extraction on the text prompt information, the user's intention can be accurately understood. At the same time, based on the encoding of semantic features, a material perception mechanism is introduced. By combining the specified material type with the encoding vector and inputting it into the raw image model, the expressiveness of the material can be precisely controlled during the image generation process. The generated advertising image can not only accurately convey the text information, but also visually present delicate material effects such as metallic, matte texture, velvet background, etc., enhancing the realism and high-end texture of the image.

[0032] Figure 2 FIG. 1 is a flow chart of the method for generating an advertisement image according to an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S140.

[0033] S110, receiving text prompt information and a specified material type input by a user;

[0034] Specifically, the system provides a user-friendly interface that allows users to enter text prompt information related to the advertisement, such as the theme of the advertisement, the target audience, product features, etc. At the same time, the interface also provides a material type selection function, allowing users to select the desired material type from a predefined list, such as metal, glass, velvet, etc.

[0035] This step ensures that the ad image generation process fully considers the user's personalized needs and the specific requirements of the ad. By receiving the user's input text prompt information and material type, the system can more accurately understand the user's intent, providing a foundation for subsequent semantic extraction and material encoding.

[0036] In one embodiment, receiving the text prompt information and the specified material type input by the user includes:

[0037] A user input interface is established and a material type selector is provided in the user input interface for receiving text prompt information input by the user and the material type specified by the material type selector.

[0038] Specifically, a graphical user interface (GUI) should be developed that can be presented to users through a web page, mobile application, or desktop software. The interface design should be simple, intuitive, and easy for users to operate. For example, it can include elements such as text input boxes, buttons, and drop-down menus to guide users through the various inputs required to generate advertising images. Input requirements should be clearly marked on the interface, such as prompting users to enter text information such as the advertisement's theme, target audience, and product features, so that users can accurately understand and fill in the relevant content. A material type selector should be integrated into the user input interface. This selector can be in the form of a drop-down menu, a set of buttons, or a slider, and its specific design should be considered based on user habits and interface aesthetics. The material type selector should include common material types used in financial advertisements, such as metal, glass, velvet, and kraft paper. When a user clicks or selects a material type, the system should be able to record and recognize this selection for subsequent use in the advertising image generation process.

[0039] In other words, by establishing a user input interface, the system can provide a unified, easy-to-use entry point, allowing users to conveniently enter the various information required to generate advertising images. Interface design helps improve the user experience, reduces the user's learning curve, and makes it easy for non-professional users to get started. The unified input interface facilitates the system's subsequent processing and analysis of input information, improving the accuracy and efficiency of advertising image generation. In addition, by providing a material type selector, the system allows users to clearly specify the type of material they wish to include in the advertising image, thereby enhancing the user's control over the advertising image generation process. Explicitly specifying the material type helps the system more accurately understand the user's intent and generate advertising images that meet the user's expectations. The integration of the material type selector simplifies the user's operation process. Users no longer need to manually enter material information; they can complete the material type specification with a simple click or selection, improving the user experience and operational efficiency.

[0040] S120, performing semantic extraction on the text prompt information to obtain semantic features;

[0041] Specifically, the system uses natural language processing technology (such as NLP models) to perform semantic analysis on the text prompt information entered by the user, extracting semantic features related to the advertising theme, target audience, product features, etc. These semantic features are represented in the form of vectors to facilitate subsequent processing and encoding.

[0042] In other words, the semantic extraction step enables the system to deeply understand the text information entered by the user and convert the abstract text description into a computable semantic vector, which helps to more accurately generate advertising images that meet the user's intentions in subsequent steps and improve the semantic relevance and accuracy of the image.

[0043] In one embodiment, the semantic extraction of the text prompt information to obtain semantic features includes:

[0044] Clean the text prompt information to obtain clean text;

[0045] Specifically, construct a text cleaning rule library. The rules include removing special characters (such as redundant symbols in punctuation marks, non-text characters, etc.), removing stop words (such as words with no practical meaning in semantic understanding like "de", "le", "he", etc.), and removing repetitive and redundant expressions. Write a text cleaning program, and use tools such as regular expressions to scan and match the text prompt information input by the user character by character, and perform deletion or replacement operations on characters, words, etc. that do not meet the requirements according to the rules in the rule library. For example, for the text "High-end, financial products!!", after cleaning, "High-end financial products" is obtained.

[0046] That is to say, remove the noise information in the text, such as special characters, stop words, and redundant expressions, to make the text more concise and standardized, and improve the accuracy and efficiency of subsequent processing. Avoid interference from irrelevant information to semantic extraction and advertisement image generation.

[0047] Perform word segmentation on the clean text to obtain segmented text;

[0048] Specifically, select a suitable word segmentation tool, such as a dictionary-based word segmentation algorithm (forward maximum matching method, backward maximum matching method, etc.) or a statistics-based word segmentation model (such as hidden Markov model, conditional random field, etc.). Input the cleaned text into the word segmentation tool, and the tool segments the text according to the preset dictionary or trained model to obtain the word segmentation result. For example, for the text "High-end financial products", after word segmentation, three words "High-end", "financial", and "products" are obtained.

[0049] That is to say, by splitting the continuous text into meaningful lexical units, it is convenient for subsequent semantic analysis and processing of each word. The segmented text is more easily understood and processed by the computer, which helps to accurately extract the semantic information of the text.

[0050] Parse the segmented text to obtain parsed text;

[0051] Specifically, first, the part-of-speech tagging function in natural language processing technology is used to mark the part of speech (such as noun, verb, adjective, etc.) for each word segmentation to help understand the grammatical role of the word in the sentence. Secondly, named entity recognition is performed to identify specific entities such as names of people, places, institutions, and products in the text. For example, for the word segmentation text "high-end, finance, product", after parsing, "high-end (adjective), finance (noun), product (noun)" is obtained, and it is recognized that "financial product" may be a specific concept. Syntactic analysis can also be performed to analyze the grammatical relationship between words in a sentence, construct structures such as syntax trees, and further understand the semantic structure of the text.

[0052] In other words, through technologies like part-of-speech tagging, named entity recognition, and syntactic analysis, we gain a deeper understanding of the text's grammatical structure and semantic relationships, providing richer information for subsequent semantic extraction. The parsed text can more accurately reflect the user's intent and the text's core content, helping to generate advertising images that better meet user needs.

[0053] Expanding the parsed text to obtain an expanded text;

[0054] Specifically, a knowledge base is first constructed, containing information such as vocabulary, phrases, sentence templates, and common semantic associations related to the financial field. Secondly, based on the semantic information of the parsed text, relevant extended content is searched from the knowledge base. For example, for the term "high-end financial products" in the parsed text, the knowledge base may contain extended information such as "high-end financial products generally have high returns and low risks, and are suitable for investors with a certain amount of asset accumulation." Finally, a text generation algorithm (such as template-based generation or neural network-based generation) is used to fuse the extended content with the parsed text to generate the expanded text.

[0055] In other words, expanding the semantic content of the text makes it richer and more detailed, containing more information relevant to ad image generation. This expanded text provides more context for semantic extraction, helping to generate more accurate and comprehensive semantic features, thereby improving the quality and relevance of ad image generation.

[0056] Semantic extraction is performed on the expanded text to obtain semantic features.

[0057] Specifically, a suitable semantic extraction model is selected, such as a word vector model (Word2Vec, GloVe, etc.) or a deep learning-based semantic encoding model (such as BERT, GPT, etc.). The expanded text is input into the semantic extraction model, which converts the text into a semantic feature vector by learning the semantic representations of the words in the text and the relationships between the words. For example, using the BERT model to process the expanded text produces a fixed-dimensional vector that contains the semantic information of the expanded text.

[0058] In other words, by converting the expanded text into a semantic feature vector, the text's semantic information is represented numerically, making it easier for computers to process and calculate. Semantic feature vectors accurately capture the text's core semantics, providing critical semantic guidance for subsequent ad image generation, ensuring that the resulting ad image better meets the semantic requirements of the text's prompt information.

[0059] S130, encoding the semantic features to obtain an encoding vector;

[0060] Specifically, the system uses an encoder to encode the extracted semantic features, converting them into encoding vectors suitable for subsequent image generation model processing. This step may involve operations such as feature scaling and normalization to ensure the stability and consistency of the encoding vectors.

[0061] In other words, the encoding step converts semantic features into machine-processable encoded vectors, providing the foundation for subsequent splicing with material vectors and input into the image generation model. Through encoding, the system can more effectively utilize semantic information to guide the generation of advertising images, improving the semantic accuracy and expressiveness of the generated images.

[0062] In one embodiment, encoding the semantic features to obtain an encoding vector includes:

[0063] The semantic features are input into the encoder, which converts and calculates the semantic features to obtain the encoding vector.

[0064] Specifically, select an appropriate encoder architecture based on the characteristics and requirements of the advertising image generation task. Common encoder architectures include encoders based on recurrent neural networks (RNNs) and their variants (such as LSTM and GRU), encoders based on convolutional neural networks (CNNs), and encoders based on Transformers. For example, Transformer encoders have advantages in handling long-range dependencies and parallel computing, and can better capture complex relationships in semantic features. They are suitable for processing input such as financial advertising text, which may contain complex semantics and logical relationships. Use deep learning frameworks (such as TensorFlow and PyTorch) to build a model of the selected encoder architecture. Define hyperparameters such as the number of encoder layers, the number of neurons in each layer, and the activation function. For example, for a Transformer-based encoder, you can set up multiple encoder layers, each layer containing a multi-head self-attention mechanism and a feedforward neural network, and choose an appropriate activation function (such as ReLU) to enhance the model's nonlinear expression capabilities.

[0065] Before inputting the semantic features into the encoder, the semantic features are preprocessed as necessary. If the semantic features are represented in the form of word vectors, the word vectors may need to be normalized so that their numerical range is within an appropriate interval to avoid affecting the training and performance of the encoder due to excessive numerical differences. To improve the computational efficiency and stability of the encoder, multiple semantic feature samples are usually grouped into a batch for input. The batch size is set appropriately based on hardware resources (such as GPU memory) and model complexity. For example, during training, 32 or 64 semantic feature samples can be input into the encoder as a batch.

[0066] After receiving the input semantic features, the encoder begins forward propagation. For a Transformer-based encoder, the semantic features are first processed using a multi-head self-attention mechanism. This allows the model to focus on different parts of the semantic features in different representation subspaces, thereby capturing richer semantic relationships. The output of the self-attention mechanism is then further transformed and computed using a feedforward neural network, which typically consists of two fully connected layers with an activation function performing a nonlinear transformation. If the encoder contains multiple layers, the output of each layer serves as the input to the next layer, and information is continuously passed and processed between layers. By stacking multiple layers of encoders, the model can gradually extract higher-level and more abstract semantic feature representations. After computation across all encoder layers, the final encoding vector is obtained. The dimensionality of the encoding vector can be set based on the specific task and model design. It contains the key semantic information of the input semantic features after transformation and computation by the encoder.

[0067] In other words, the encoder compresses the original semantic features, removing redundant information and retaining the most critical semantic components. For example, financial advertisements may contain a large amount of descriptive vocabulary and detailed information. The encoder extracts the semantic information most relevant to the core theme of the advertisement (such as product features and target audience) and represents it in a more compact form. Furthermore, by converting semantic features of varying forms and dimensions into encoding vectors of fixed dimensions, a unified representation of semantic information is achieved. This enables subsequent models (such as raw graph models) to easily process and understand semantic information. Regardless of the source of the input semantic features, they can interact and fuse them using the same encoding vector. Furthermore, through complex computations within the encoder (such as the multi-head self-attention mechanism), complex relationships between semantic features can be captured. In financial advertisements, different words may have semantic connections and dependencies. The encoder identifies these relationships and reflects them in the encoding vector. For example, the semantic connection between "high-end" and "financial products" is captured by the encoder, allowing the encoding vector to better represent the overall semantic meaning of the advertisement. Furthermore, the encoder's processing of semantic features helps improve semantic consistency. In advertising image generation, ensuring that the generated image is semantically consistent with the input semantic cues is crucial. The encoder encodes semantic features so that the encoded vector accurately conveys semantic information, providing reliable semantic guidance for subsequent image generation and reducing the deviation between the generated image and the semantic cues.

[0068] S140: Input the encoding vector and the specified material type into the image generation model to generate an advertising image.

[0069] Specifically, the system concatenates the encoding vector and the specified material type (in the form of a material vector) and inputs it into a raw image model (such as a Diffusion model). During each denoising step in the raw image model, the encoding vector and the material vector participate in the cross-attention calculation, guiding the model to generate advertising images that meet semantic and material requirements.

[0070] In other words, this step is the core of advertising image generation. By inputting the encoding vector and material vector into the raw image model, the system can simultaneously consider semantic information and material expressiveness during the generation process, thereby generating advertising images that both meet user intent and have realistic textures. This greatly improves the visual effect and appeal of advertising images, helping to enhance brand image and user trust. At the same time, because the system can automatically generate advertising images, it also significantly reduces design costs and time costs.

[0071] In one embodiment, the raw image model is obtained by training with big data, and the specific training steps are as follows:

[0072] Collect relevant pictures and process them into the same size;

[0073] Specifically, collect pictures related to financial products from channels such as financial advertising material libraries, commercial image libraries, and online resources, such as promotional pictures of credit cards, financial products, and bank service scenes. Ensure that the collected pictures cover a variety of material expressions, such as metal cards, glass display cabinets, velvet backgrounds, etc. In addition, use image processing tools (such as OpenCV, Pillow, etc.) to uniformly process the collected pictures into the same size, such as 256×256 pixels. This step helps standardize the input data during subsequent model training and improve training efficiency and stability.

[0074] Perform text tagging on the processed images, and mark the image content and material category to obtain pre-processed images;

[0075] Specifically, provide a text description of the processed image, describing the main content of the image, such as "high-end credit card display, metal texture, blue velvet background." The text description should be accurate and detailed, covering the core information and material characteristics of the image. For materials appearing in the image, label their specific categories, such as metal, glass, velvet, plastic, etc. A tag file (such as JSON or XML) can be used to record the text description and material category annotation information for each image.

[0076] Predefine material types and encode them as low-dimensional composable material vectors;

[0077] Specifically, a set of material types, such as metal, glass, velvet, kraft paper, and plastic, is predefined based on common materials found in financial advertisements. Furthermore, appropriate encoding methods (such as one-hot encoding and embedded coding) are used to encode these predefined material types into low-dimensional, composable material vectors. For example, using embedded coding, a fixed-dimensional vector representation, such as a 32-dimensional vector, is learned for each material type. These material vectors can then be used in subsequent model training to help the model learn material characteristics.

[0078] Randomly combine the preprocessed image and the texture vector to form a preprocessed image;

[0079] Specifically, during training, an image is randomly selected from a preprocessed image set, and one or more material vectors are randomly selected from a predefined set of material vectors (based on actual needs). The image and material vectors are then combined. For example, a credit card image is combined with a metal material vector and a blue velvet background material vector to form a preprocessed image containing both image and material information. To increase the diversity of the training data and improve the model's generalization ability, data augmentation operations can be performed during the random combination process, such as random cropping, rotation, and flipping of the image.

[0080] Performing denoising on the preprocessed image to obtain a denoised image;

[0081] Specifically, random noise is added to the preprocessed image to simulate the uncertainty in the image generation process. The type and intensity of the noise can be adjusted according to actual needs, such as Gaussian noise, salt and pepper noise, etc. In addition, a denoising algorithm (such as a deep learning-based denoising model such as DnCNN or FFDNet) is used to denoise the noisy image to obtain a denoised image. The denoised image will serve as the training target for the model, which needs to learn to recover a clear and accurate image from the noisy preprocessed image.

[0082] The above operations are iteratively repeated until the denoised image reaches the preset performance index to obtain the raw image model.

[0083] Specifically, a noisy preprocessed image is used as input and a denoised image is used as output to construct a training dataset. A raw image model (such as a variant based on the Diffusion model) is built using a deep learning framework (such as TensorFlow, PyTorch, etc.), and the model is trained using the training dataset. During the training process, an optimization algorithm (such as Adam, SGD, etc.) is used to update the model parameters to minimize the loss function (such as mean square error loss, perceptual loss, etc.) between the model output and the denoised image. In addition, the performance of the training model is regularly evaluated using a validation dataset. The evaluation indicators may include image quality indicators (such as PSNR, SSIM), style consistency indicators, user satisfaction, etc. The model structure and training parameters, such as learning rate and batch size, are adjusted based on the evaluation results. When the performance indicator of the denoised image on the validation dataset reaches a preset threshold, the model training is stopped to obtain the final raw image model.

[0084] In other words, uniformly processing images to the same size avoids model training instability caused by inconsistent image sizes, helping the model learn more accurate image features, thereby improving the quality of generated images. By denoising the preprocessed images, the model learns to recover clear images from noise, resulting in clearer, sharper, and less noisy and blurry advertising images. Furthermore, by labeling images with material categories and encoding material types as material vectors, the model explicitly learns the characteristics and representations of different materials. When generating images, the model can accurately represent the corresponding material textures, such as the luster of metal or the softness of velvet, based on the input material vectors, enhancing the realism and appeal of advertising images. Furthermore, by randomly combining preprocessed images and material vectors, the model increases the diversity of training data, exposing the model to more images with different material combinations. This enhances the model's understanding of materials and generalization capabilities, enabling it to generate more diverse advertising images that meet real-world needs. Furthermore, performing data augmentation during the random combination process further increases the diversity of training data, allowing the model to learn more diverse image features and variations, improving its generalization capabilities and preventing overfitting. By iteratively repeating training operations and adjusting model parameters based on performance evaluation results, the model's performance can be continuously optimized, enabling it to consistently generate high-quality advertising images for a variety of inputs. Furthermore, text tagging enables the model to understand the content and subject matter of images, while the introduction of material vectors allows users to specify the material type for generated images. This combination enables the model to generate personalized advertising images based on user-entered text prompts and material requirements, meeting the needs of different financial products and brands and improving the relevance and effectiveness of advertising.

[0085] In one embodiment, inputting the encoding vector and the specified material type into a raw image model to generate an advertisement image includes:

[0086] Convert the specified material type into a material vector;

[0087] Specifically, a set of material types, such as metal, glass, velvet, leather, and plastic, is predefined based on common materials in the financial advertising field. Then, an embedding encoding method is used to construct a coding library for each material type. Specifically, using the embedding layer in a deep learning framework (such as TensorFlow or PyTorch), a fixed-dimensional vector representation is assigned to each material type. For example, if the dimension of the material vector is set to 32, each material type will correspond to a 32-dimensional vector. These vectors are continuously learned and optimized during model training to better represent the characteristics of the materials. In addition, when the user specifies a material type, the system searches for the corresponding material type from the predefined set of material types. The material vector corresponding to this material type is then retrieved from the coding library. For example, if the user specifies the material type as "metal," the system retrieves the 32-dimensional vector corresponding to "metal" from the coding library.

[0088] Concatenate the encoding vector and the material vector to form a combined vector;

[0089] Specifically, after obtaining the encoding vector and texture vector, they are combined into a single combined vector using vector concatenation. For example, assuming the encoding vector has 512 dimensions and the texture vector has 32 dimensions, the combined vector will have a dimension of 512 + 32 = 544. In deep learning frameworks, specialized concatenation functions (such as tf.concat in TensorFlow or torch.cat in PyTorch) can be used to implement vector concatenation.

[0090] The combined vector is input into the image generation model to generate the advertising image.

[0091] Specifically, a suitable image generation model architecture is selected, such as a variant based on the diffusion model. The diffusion model has excellent performance in image generation, capable of generating high-quality and diverse images. A deep learning framework is used to build the image generation model, defining hyperparameters such as the number of layers, the number of neurons in each layer, and the activation function. For example, a diffusion model may include multiple denoising layers, each implemented using a convolutional neural network (CNN) or Transformer architecture. Furthermore, the concatenated combined vector is input into the image generation model. The image generation model gradually generates the advertising image based on the semantic and material information contained in the combined vector. During the generation process, the model iterates through a series of random noise calculations, gradually removing the noise and generating a clear image. Each iteration adjusts the image generation direction based on the combined vector, ensuring that the generated image both conforms to the semantic cues and possesses the specified material texture. Ultimately, the image generation model outputs a generated advertising image that visually matches the input textual cues and material type.

[0092] Specifically, by converting the specified material type into a material vector and concatenating it with the encoding vector before inputting it into the raw image model, the model can clearly receive information about the material. This allows the generated advertising images to more accurately represent the material, accurately displaying user-specified textures such as the luster of metal, the transparency of glass, and the softness of velvet, thereby enhancing the realism and appeal of the advertising images. Furthermore, the encoding vector contains the semantic information of the user-entered text prompt, while the material vector provides material feature information. Combining these two and inputting them into the raw image model allows the model to consider both semantic and material requirements when generating images. The resulting advertising images are semantically consistent with the text prompt while accurately representing the specified material, avoiding mismatches between semantics and material, and improving the quality and usability of the advertising images. Furthermore, users can freely specify the material type according to their needs, and the system generates advertising images with the corresponding material texture based on their specifications. This personalized generation method meets the promotional needs of different financial products and brands, allowing advertisements to better highlight the product's features and advantages and attract the attention of the target audience. Furthermore, by introducing the material vector, the raw image model can learn more about material characteristics and representations, enriching its knowledge base. This enables the model to generate more diverse, high-quality advertising images for different inputs, improving its generative and generalization capabilities. Furthermore, the concatenated vectors provide the model with more information dimensions, helping it better understand the input requirements and generate images that better meet expectations.

[0093] In one embodiment, after inputting the encoding vector and the specified material type into the raw image model to generate the advertisement image, the method further includes:

[0094] Visualize the advertising image.

[0095] Specifically, choose a visualization display platform:

[0096] Web platform: Develop a web-based advertising image display page, using front-end technologies such as HTML, CSS, and JavaScript to build page layout and interactive effects. For example, using HTML5 <canvas>Elements or CSS layout features to display ad images and add interactive elements such as buttons and links to facilitate user interaction with the images.

[0097] Mobile app platforms: If you're displaying ad images within a mobile app, you can use native development technologies (such as Java / Kotlin for Android or Swift / Objective-C for iOS) or cross-platform development frameworks (such as Flutter or React Native) to create the app interface. Set up a dedicated ad display area within the app to display the generated ad images.

[0098] Desktop software platform: For desktop applications, use appropriate development tools (such as Python's PyQt, Tkinter, or C#'s WPF) to design the software interface. Reserve a window or panel within the interface for displaying advertising images so that the generated images can be presented to the user.

[0099] In addition, the advertising images generated by the raw image model may be in a specific format (such as the intermediate format generated by some deep learning frameworks) and need to be converted to common image formats (such as JPEG and PNG) for normal display on the display platform. Image processing libraries (such as Python's Pillow and OpenCV) can be used to perform format conversion. In order to improve the image loading speed and display effect, the converted images are optimized. For example, the image resolution can be adjusted to fit the screen size of the display platform, and the image file size can be compressed to reduce network transmission time and storage space. Image compression algorithms (such as JPEG's lossy compression and PNG's lossless compression) can be used to achieve image optimization.

[0100] If the ad image is stored on the local device, load the image into memory through the corresponding file read operation. For example, in Web development, use JavaScript's fetchAPI or Use the src attribute of the tag to load local image files. In mobile app development, use file manipulation APIs to read locally stored images. If the ad image is stored on a server, retrieve the image data from the server through a network request. In web development, use AJAX or the modern Fetch API to send HTTP requests to retrieve image resources from the server. In mobile app development, use network request libraries (such as Retrofit for Android and Alamofire for iOS) to load images from the network. Once loaded, decode the image data and display it in the corresponding interface element.

[0101] Additionally, add a click event listener to the ad image. When a user clicks the image, it triggers a corresponding action, such as redirecting to the product details page or displaying more relevant information. In web development, use JavaScript's addEventListener method to listen for click events; in mobile app development, use the corresponding gesture recognition API to implement click interactions. For ad images that require detailed viewing, you can implement zooming and dragging. In web development, use the CSS transform property and JavaScript to achieve zooming and dragging effects; in mobile app development, leverage the platform's touch event handling mechanisms to implement these interactive features.

[0102] In other words, by visually displaying the generated ad images, users can intuitively see the final ad effect, including the image's color, composition, and material performance. This helps users evaluate whether the ad meets expectations and effectively conveys product information and brand image. Furthermore, displaying ad images on appropriate platforms and providing good interactive features can enhance the user experience. For example, in mobile apps, users can easily view ad images and interact with them through actions such as clicking and zooming, increasing user attention and engagement with the ad. Furthermore, visually displayed ad images provide an intuitive basis for evaluating ad effectiveness. Based on the display results, users can collect feedback and adjust and optimize ad text prompts, material type selection, and image generation model parameters, thereby improving ad quality and effectiveness.

[0103] For example, consider the generation of advertising images for high-end credit cards in the financial sector. A bank plans to launch a high-end credit card designed to attract high-net-worth customers. To effectively promote this card, the bank needs to generate a series of high-quality, personalized advertising images that showcase the card's appearance, feel, and associated financial services and benefits.

[0104] A bank employee logs into the user input interface of the advertising image generation system. The interface contains a text input box for entering textual information about the credit card, such as "High-end credit card, exclusive private banking services, high credit limit, global travel insurance." The interface also provides a material selector. The employee selects "Metal" from the drop-down menu as the material type for the card, as this high-end card is intended to be made of metal to reflect its premium quality.

[0105] The system cleans the input text prompt, removing extra spaces and punctuation, to produce the following: "High-end credit card, exclusive private banking services, high credit limit, global travel insurance." A word segmentation tool is used to segment the cleaned text, resulting in segmented text such as "high-end," "credit card," "exclusive," "private banking services," "high amount," "credit limit," "global," and "travel insurance." Furthermore, parsing operations such as part-of-speech tagging and named entity recognition are performed. For example, "credit card," "private banking services," and "travel insurance" are identified as nouns, and "high-end," "exclusive," and "high amount" as adjectives, further understanding the semantic structure of the text. Based on the parsing results, relevant information is searched in the knowledge base and expanded, resulting in an expanded text: "This high-end credit card offers cardholders exclusive private banking services, a high credit limit, and complimentary global travel insurance, giving you greater peace of mind when traveling." The expanded text is then fed into a semantic extraction model to produce a semantic feature vector containing core semantic information about the credit card, such as "high-end," "private banking services," "high credit limit," and "travel insurance."

[0106] The semantic feature vector is input into the encoder, which converts and calculates the semantic features to produce an encoding vector. The encoding vector further compresses and refines the semantic information, facilitating subsequent combination with the material vector. The system converts the specified "metal" material type into a material vector. According to a predefined material type encoding library, "metal" corresponds to a vector of a specific dimension, such as a 32-dimensional vector, which contains characteristic information about the metal material, such as gloss and texture. The encoding vector and the material vector are concatenated to form a combined vector. This combined vector contains both semantic and material information, providing comprehensive input for the raw image model. The combined vector is then input into the raw image model, which gradually generates the advertising image based on the input information. The model takes into account the high-level semantics of the credit card and the metallic material characteristics, generating an image depicting a metallic credit card. The image may also contain elements related to private banking services or travel insurance, such as bank logos and travel maps.

[0107] The generated advertising image can be visualized in the bank's internal advertising planning system. Staff can view the generated image on a web page and evaluate its color, composition, and other aspects. If adjustments are needed, they can return to modify the text prompt or material type and regenerate the image. The finalized advertising image can be used across the bank's online and offline promotional channels, such as the bank's website, mobile banking app, and posters in offline branches, to attract target customers.

[0108] Specifically, through semantic extraction and encoding, the generated advertising images accurately convey the core selling points of high-end credit cards, such as private banking services, high credit limits, and travel insurance, allowing customers to quickly understand the product's advantages. By converting the metal material type into a material vector and incorporating it into image generation, the advertising images realistically showcase the credit card's metallic texture, enhancing the product's appeal and high-end image. Compared to traditional manually designed advertising images, this method can rapidly generate multiple advertising images of varying styles, significantly improving advertising production efficiency and shortening product promotion cycles. Banks can flexibly adjust text prompts and material types based on target customer groups and promotional needs, generating personalized advertising images and improving advertising relevance and effectiveness.

[0109] Another example is the generation of advertising images for smart health monitoring wristbands in the healthcare sector. A medical device company plans to launch a smart health monitoring wristband with multiple functions, such as heart rate monitoring, sleep monitoring, and activity tracking. The wristband's outer shell is made of soft, skin-friendly silicone. To promote this new product, the company needs to generate a series of attractive advertising images that showcase the wristband's functions and material characteristics to attract consumers.

[0110] The company's marketing personnel logged into the user input interface of the advertising image generation system. The interface contained a text input box for entering a description of the smart health monitoring wristband, such as "New smart health monitoring wristband, accurate heart rate monitoring, deep sleep analysis, comprehensive exercise tracking." The interface also included a material selector. The marketing personnel selected "Silicone" from the drop-down menu as the wristband's outer shell is made of silicone, which is soft, skin-friendly, and comfortable to wear.

[0111] The system cleans the input text prompt, removing extra spaces and special characters, and obtains the following: "New smart health monitoring bracelet with accurate heart rate monitoring, deep sleep analysis, and comprehensive exercise tracking." A word segmentation tool is used to segment the cleaned text, resulting in segmented text such as "new," "smart," "health monitoring," "bracelet," "accurate," "heart rate monitoring," "deep," "sleep analysis," "comprehensive," and "exercise tracking." Furthermore, through parsing operations such as part-of-speech tagging and semantic understanding, the system identifies "bracelet" as a noun, "smart," "accurate," "deep," and "comprehensive" as adjectives, and "health monitoring," "heart rate monitoring," "sleep analysis," and "exercise tracking" as noun phrases, further understanding the bracelet's functions and features described in the text. Based on the parsing results, the system searches the knowledge base for relevant information and expands the text, obtaining the expanded text: "This new smart health monitoring bracelet features accurate heart rate monitoring, which can accurately record your heart rate changes in real time. It also has a deep sleep analysis function to help you understand your sleep quality. Furthermore, it can comprehensively track your exercise status, providing strong support for your healthy lifestyle." In addition, the expanded text is input into the semantic extraction model to obtain a semantic feature vector, which contains the core semantic information about the bracelet, such as intelligence, health monitoring, heart rate monitoring, sleep analysis, exercise tracking, etc.

[0112] The semantic feature vector is input into the encoder, which converts and calculates the semantic features to produce an encoding vector. The encoding vector further refines and compresses the semantic information, facilitating subsequent combination with the material vector. The system converts the specified "silicone" material type into a material vector. According to a predefined material type encoding library, "silicone" corresponds to a vector of a specific dimension, such as a 32-dimensional vector, which contains characteristic information about the silicone material, such as softness, skin-friendliness, and color. The encoding vector and the material vector are concatenated to form a combined vector. This combined vector contains both semantic and material information, providing comprehensive input for the raw image model. The combined vector is then input into the raw image model, which gradually generates an advertising image based on the input information. The model takes into account the smart health monitoring capabilities of the wristband and the characteristics of the silicone material to generate an image depicting the wristband's appearance. The image shows the wristband worn on the wrist, with the screen displaying heart rate, sleep, and activity data. The texture of the silicone material is clearly visible, creating a soft and skin-friendly effect.

[0113] The generated advertising image can be visualized within the company's marketing system. Marketing personnel can view the generated image on a web page and evaluate its color, composition, and information presentation. If adjustments are needed, they can return to modify the text prompts or material type and regenerate the image. The finalized advertising image can be used across the company's online and offline promotional channels, such as the company website, social media platforms, and offline product posters, attracting consumer attention and increasing product awareness and market competitiveness.

[0114] In other words, through semantic extraction and encoding, the generated advertising image accurately conveys the core functions of the smart health monitoring wristband, such as heart rate monitoring, sleep monitoring, and activity tracking, allowing consumers to quickly understand the product's advantages and uses. By converting the silicone material type into a material vector and incorporating it into image generation, the advertising image truly showcases the wristband's silicone texture, highlighting its softness and skin-friendliness, and increasing consumer trust and favorability. Compared to traditional manually designed advertising images, this method can quickly generate multiple advertising images in different styles, significantly improving advertising production efficiency and shortening product promotion cycles. Furthermore, companies can flexibly adjust text prompts and material types based on different promotional objectives and target audiences, generating personalized advertising images for greater relevance and effectiveness. For example, for sports enthusiasts, the fitness tracking function of the wristband can be emphasized, while for those concerned about health, the health monitoring function can be highlighted.

[0115] The above-mentioned advertising image generation method ensures that the generated image can closely fit the user's personalized needs and brand tone by receiving the text prompt information and the specified material type input by the user. Then, by performing semantic extraction on the text prompt information, it can accurately understand the user's intention, including the target user group, product positioning, and the emotion or value to be conveyed. At the same time, on the basis of encoding the semantic features, a material perception mechanism is introduced. By combining the specified material type with the encoding vector and inputting it into the raw image model, the expressiveness of the material can be accurately controlled during the image generation process. This means that the generated advertising image can not only accurately convey the text information, but also visually present delicate material effects such as metallic feel, matte texture, velvet background, etc., thereby greatly enhancing the realism and high-end texture of the image.

[0116] Figure 3 FIG is a schematic block diagram of an advertising image generating device 300 provided by an embodiment of the present invention. Figure 3 As shown, corresponding to the above advertising image generation method, the present invention also provides an advertising image generation device 300. The advertising image generation device 300 includes a unit for executing the above advertising image generation method, and the device can be configured in a server. Figure 3 , the advertising image generating device 300 includes:

[0117] The receiving unit 301 is configured to receive text prompt information and a specified material type input by a user;

[0118] Extraction unit 302, used for performing semantic extraction on the text prompt information to obtain semantic features;

[0119] The encoding unit 303 is used to encode the semantic features to obtain an encoding vector;

[0120] The input generation unit 304 is configured to input the encoding vector and the specified material type into the raw image model to generate an advertisement image.

[0121] In one embodiment, the receiving unit 301 includes:

[0122] A receiving module is established to establish a user input interface and provide a material type selector in the user input interface, so as to receive text prompt information input by the user and the material type specified by the material type selector.

[0123] In one embodiment, the extraction unit 302 includes:

[0124] A cleaning module is used to clean the text prompt information to obtain a cleaned text;

[0125] A processing module is used to perform word segmentation processing on the cleaned text to obtain a word segmentation text;

[0126] The parsing module is used to parse the segmented text to obtain the parsed text;

[0127] An expansion module, used to expand the parsed text to obtain an expanded text;

[0128] The extraction module is used to perform semantic extraction on the expanded text to obtain semantic features.

[0129] In one embodiment, the encoding unit 303 includes:

[0130] The input conversion module is used to input the semantic features into the encoder, and the encoder converts and calculates the semantic features to obtain the encoding vector.

[0131] In one embodiment, the input generation unit 304 includes:

[0132] The conversion module is used to convert the specified material type into a material vector;

[0133] A concatenation module, used to concatenate the encoding vector and the material vector to form a combined vector;

[0134] The input generation module is used to input the combined vector into the image generation model to generate an advertising image.

[0135] In one embodiment, the raw image model is obtained by big data training, and the specific training is as follows: relevant pictures are collected and uniformly processed into the same size; text is marked on the processed pictures, and the picture content and material category are annotated to obtain preprocessed pictures; material types are predefined, and the material types are encoded into low-dimensional combinable material vectors; the preprocessed pictures and material vectors are randomly combined to form preprocessed images; the preprocessed images are denoised to obtain denoised images; the above operations are iteratively repeated until the denoised images meet preset performance indicators to obtain the raw image model.

[0136] In one embodiment, the apparatus further comprises:

[0137] The display unit is used to visually display the advertising image.

[0138] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned advertising image generation device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment, and for the convenience and brevity of description, it will not be repeated here.

[0139] The above-mentioned advertising image generating device 300 can be implemented in the form of a computer program. The computer program can be used in Figure 4 Runs on the computer equipment shown.

[0140] See also Figure 4 , Figure 4 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0141] See Figure 4 The computer device 500 includes a processor 502 , a memory, and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0142] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, which, when executed, can cause the processor 502 to execute a method for generating an advertisement image.

[0143] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0144] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503 . When the computer program 5032 is executed by the processor 502 , the processor 502 can execute a method for generating an advertisement image.

[0145] The network interface 505 is used to communicate with other devices through the network. Figure 4 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0146] The processor 502 is configured to execute a computer program 5032 stored in the memory to implement the following steps:

[0147] Receive text prompt information and specified material type input by the user; perform semantic extraction on the text prompt information to obtain semantic features; encode the semantic features to obtain an encoding vector; input the encoding vector and the specified material type into the raw image model to generate an advertising image.

[0148] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0149] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a storage medium that is computer-readable. The program instructions are executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0150] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor performs the following steps:

[0151] Receive text prompt information and specified material type input by the user; perform semantic extraction on the text prompt information to obtain semantic features; encode the semantic features to obtain an encoding vector; input the encoding vector and the specified material type into the raw image model to generate an advertising image.

[0152] The storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0153] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0154] In the several embodiments provided herein, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the various units is merely a logical functional division, and actual implementation may employ other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0155] The steps in the methods of the embodiments of the present invention may be adjusted in order, combined, or deleted as needed. The units in the devices of the embodiments of the present invention may be combined, divided, or deleted as needed. Furthermore, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0156] If this integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, terminal, or network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present invention.

[0157] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.< / canvas>

Claims

1. A method for generating an advertising image, characterized in that: include: Receive the text prompt information and specified material type entered by the user; Perform semantic extraction on text prompt information to obtain semantic features; Encode the semantic features to obtain an encoding vector; Input the encoding vector and the specified material type into the image generation model to generate the advertising image.

2. The advertising image generation method according to claim 1, wherein: The receiving of the text prompt information input by the user and the specified material type includes: A user input interface is established and a material type selector is provided in the user input interface for receiving text prompt information input by the user and the material type specified by the material type selector.

3. The advertising image generation method according to claim 1, wherein: The semantic extraction of the text prompt information to obtain semantic features includes: Cleaning the text prompt information to obtain a cleaned text; Perform word segmentation processing on the cleaned text to obtain word segmentation text; Parse the segmented text to obtain the parsed text; Expanding the parsed text to obtain an expanded text; Semantic extraction is performed on the expanded text to obtain semantic features.

4. The advertising image generation method according to claim 1, wherein: The encoding of the semantic features to obtain an encoding vector includes: The semantic features are input into the encoder, which converts and calculates the semantic features to obtain the encoding vector.

5. The advertising image generation method according to claim 1, wherein: The step of inputting the encoding vector and the specified material type into the image generation model to generate the advertisement image includes: Convert the specified material type into a material vector; Concatenate the encoding vector and the material vector to form a combined vector; The combined vector is input into the image generation model to generate the advertising image.

6. The advertising image generation method according to claim 1, wherein: The raw image model is obtained by big data training. The specific training steps are as follows: Collect relevant pictures and process them into the same size; Perform text tagging on the processed images, and mark the image content and material category to obtain pre-processed images; Predefine material types and encode them as low-dimensional composable material vectors; Randomly combine the preprocessed image and the texture vector to form a preprocessed image; Performing denoising on the preprocessed image to obtain a denoised image; The above operations are iteratively repeated until the denoised image reaches the preset performance index to obtain the raw image model.

7. The advertising image generation method according to claim 1, wherein: After inputting the encoding vector and the specified material type into the raw image model to generate the advertisement image, the method further includes: Visualize the advertising image.

8. An advertising image generating device, characterized in that include: A receiving unit, configured to receive text prompt information and a specified material type input by a user; An extraction unit, configured to perform semantic extraction on the text prompt information to obtain semantic features; The encoding unit is used to encode the semantic features to obtain an encoding vector; The input generation unit is used to input the encoding vector and the specified material type into the image generation model to generate the advertising image.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images