Advertisement generation method, system, medium, and device

Abstract

The application provides an advertisement generation method, system, medium and system, wherein the method comprises: obtaining product information and scene information of at least one product; generating a product image corresponding to the product information based on artificial intelligence content production AIGC, measuring each dimension of the product image to obtain label information of each dimension of the product image, and performing semantic combination on the label information of each dimension and the scene information to generate an advertisement script; and matching an advertisement template from an advertisement template library according to the label information of each dimension, and synthesizing the product image and the advertisement script into the advertisement template to generate a product advertisement. The scheme has the technical advantages of simplicity and efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet, and in particular to an advertisement generation method, system, medium and device. BACKGROUND

[0002] Advertisement has been an important way of product promotion. With the rapid development of the Internet, it is necessary to promote the advertisement of online sales products on other platforms, so as to attract users. Therefore, how to simply and efficiently generate an advertisement is a technical problem that technicians in the field need to face. SUMMARY

[0003] Therefore, the present application provides an advertisement generation method, system, medium and electronic device, which mainly aims to provide a simple and efficient advertisement generation scheme.

[0004] According to one aspect of the present application, an advertisement generation method is provided, comprising:

[0005] obtaining product information and scene information of at least one product;

[0006] generating a product image corresponding to the product information based on artificial intelligence content production AIGC, measuring each dimension of the product image to obtain label information of each dimension of the product image, and combining the label information of each dimension and the scene information to generate an advertisement script;

[0007] According to the label information of each dimension, an advertisement template is matched from an advertisement template library, and the product image and the advertisement script are synthesized into the advertisement template to generate a product advertisement.

[0008] In one implementation,

[0009] The product information includes product text description information or product category information;

[0010] The AIGC generates the product image corresponding to the product information, comprising: inputting the product text description information or product category information into an AIGC model, and outputting the product image corresponding to the product text description information according to an interactive prompt algorithm or a text-to-image algorithm of the AIGC model.

[0011] In one implementation,

[0012] The product information includes product text description information or product category information and product seed image, or the product information includes product text description information or product category information and product outline image;

[0013] The AIGC-based production content generates a product image corresponding to the product information, including: inputting the product text description information or the product category information and the product seed image, or the product text description information or the product category information and the product outline image into an AIGC model, and outputting a product image corresponding to the product text description information or the product category information and the product seed image according to a graph generation algorithm of the AIGC model, or outputting a product image corresponding to the product text description information or the product category information and the product outline image.

[0014] In an implementation manner, the AIGC model is further trained, and the training further includes:

[0015] On the basis of an open-source stable diffusion model, the historical platform product data of a platform to which the product belongs is used for secondary training to obtain a base model of a local text-to-image model and a local image-to-image model;

[0016] A low-order adaptation LORA of a large language model is performed on the base model to generate a corresponding LORA model for each product category;

[0017] The base model and the plurality of LORA models constitute the AIGC model.

[0018] In an implementation manner, the measuring of each label dimension of the product image to obtain label information of each dimension of the product image includes:

[0019] The visual information of the product image is converted into semantic information based on an image single classification algorithm and / or an image multi-label classification algorithm of the image label model;

[0020] Based on the semantic information, label information of each dimension is measured, and the label information includes any one or more of image color distribution information, image subject information, image category information, image name information, product name information, and product associated scene information.

[0021] In an implementation manner, the matching of an advertisement template from an advertisement template library according to the label information of each dimension includes:

[0022] According to any one or more of the image color distribution information, the image subject information, the image category information, the image name information, the product name information, and the product associated scene information, each template in the advertisement template library is matched to obtain an advertisement template according to a matching effect.

[0023] In an implementation manner, the semantic combination of the label information of each dimension and the scene information to generate an advertisement script includes:

[0024] inputting the label information of each dimension and the scene information into a large language model to obtain syntax and semantics corresponding to the information, and obtaining label semantic information and scene semantic information;

[0025] combining the label semantic information and the scene semantic information to obtain the advertising copy.

[0026] In an implementation manner, the synthesizing the product image and the advertising copy into the advertising template comprises:

[0027] arranging the product image in a main body position of the advertising template, arranging the advertising copy in a non-main body position of the advertising template, or arranging the product image in the main body position of the advertising template and highlighting a short number or text extracted from the advertising copy in the advertising template.

[0028] In an implementation manner, after the product image corresponding to the product information is generated, the method further comprises:

[0029] performing quality detection on the product image in at least one dimension to determine a quality detection result of each dimension; the at least one dimension comprises an image color dimension, a picture naturalness dimension, an image color value dimension, an image optical dimension, an image definition dimension, a product main body integrity dimension, a font dimension, a special character dimension, and a watermark dimension.

[0030] In the advertising synthesis process, the quality of each dimension is compared according to a detection threshold of each dimension, and it is determined whether the image quality detection result passes; only the image that passes the quality detection is subjected to advertising synthesis.

[0031] In an implementation manner, the method further comprises:

[0032] For the product image that does not pass the image quality detection result, image correction is performed to obtain a product image that satisfies the image quality detection, or the product image is re-generated according to the AIGC model.

[0033] According to an aspect of the present application, an advertising generation system is provided, comprising:

[0034] An artificial intelligence production content (AIGC) image generation engine is configured to obtain product information and scene information of at least one product, and generate a product image corresponding to the product information based on an AIGC model.

[0035] An image label system engine is configured to measure each dimension of the product image based on an image label model to obtain label information of each dimension of the product image.

[0036] The advertisement copy intelligent generation engine is configured to perform semantic combination on the label information of each dimension and the scene information based on a large language model, and generate an advertisement copy.

[0037] The material intelligent synthesis engine is configured to match an advertisement template from an advertisement template library according to the label information of each dimension, and synthesize the label image and the advertisement copy into the advertisement template to generate a label advertisement.

[0038] In an implementation manner, the label information includes label text description information or label category information.

[0039] The AIGC image generation engine is specifically configured to input the label text description information or the label category information into the AIGC model, and output a label image corresponding to the label text description information according to an interactive prompt-to-image algorithm or a text-to-image algorithm of the AIGC model.

[0040] In an implementation manner,

[0041] The label information includes label text description information or label category information and a label seed image, or the label information includes label text description information or label category information and a label contour image.

[0042] The AIGC image generation engine is specifically configured to input the label text description information or the label category information and the label seed image, or the label text description information or the label category information and the label contour image into the AIGC model, and output a label image corresponding to the label text description information or the label category information and the label seed image, or output a label image corresponding to the label text description information or the label category information and the label contour image according to a graph-to-image algorithm of the AIGC model.

[0043] In an implementation manner, the AIGC image generation engine is further configured to perform secondary training on a basis of an open-source stable diffusion model by using historical platform label data of a platform to which the label belongs, to obtain a base model of a local text-to-image model and a local graph-to-image model; perform low-order adaptation LORA of a large language model on the base model, to generate a corresponding LORA model for each category label; and use the base model and a plurality of LORA models to constitute the AIGC model.

[0044] In an implementation manner, the image label system engine is specifically configured to: convert visual information of the product image into semantic information based on an image single classification algorithm and / or an image multi-label classification algorithm of the image label model; and measure label information of each dimension based on the semantic information, the label information including any one or more of image color distribution information, image subject information, image category information, image name information, product name information, and product associated scene information.

[0045] In an implementation manner, the material intelligent synthesis engine is specifically configured to: match an advertisement template according to any one or more of image color distribution information, image subject information, image category information, image name information, product name information, and product associated scene information, and according to a matching effect of the matching.

[0046] In an implementation manner, the copy intelligent generation engine is specifically configured to: input the label information of each dimension and the scene information into the large language model, obtain syntax and semantics corresponding to the information, and obtain label semantic information and scene semantic information; and combine the label semantic information and the scene semantic information to obtain the advertisement copy.

[0047] In an implementation manner, the material intelligent synthesis engine is specifically configured to: arrange the product image in a main body position of the advertisement template, arrange the advertisement copy in a non-main body position of the advertisement template, or arrange the product image in the main body position of the advertisement template and highlight a short number or text extracted from the advertisement copy in the advertisement template.

[0048] In an implementation manner, the method further includes:

[0049] An image quality review engine configured to perform quality detection of the product image in at least one dimension and determine a quality detection result of each dimension, the at least one dimension including an image color dimension, a picture naturalness dimension, an image color value dimension, an image optical dimension, an image clarity dimension, a product main body integrity dimension, a font dimension, a special character dimension, and a watermark dimension.

[0050] The material intelligent synthesis engine is further configured to: in an advertisement synthesis process, determine whether the image quality detection result passes according to a comparison between the detection threshold value of each dimension and the quality of each dimension, and perform advertisement synthesis only on an image with a passed quality detection result.

[0051] In an implementation manner,

[0052] The material intelligent synthesis engine is further configured to: for the standard product image that does not pass the image quality detection result, perform image correction to obtain a standard product image that satisfies the image quality detection, or re-generate the standard product image according to the AIGC model.

[0053] According to an aspect of the present application, a storage medium is provided, and the storage medium stores a computer program, wherein the computer program is configured to execute the above-mentioned advertisement generation method when running.

[0054] According to an aspect of the present application, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the computer program to execute the above-mentioned advertisement generation method.

[0055] According to the above technical solution, the advertisement generation method, system, medium and device provided by the present application obtain a standard product image through AIGC text-to-image or image-to-image, establish a multi-dimensional label for the standard product image, generate an advertisement script based on the label information and the scene information, and finally synthesize the standard product image and the advertisement script into an advertisement template selected according to the multi-dimensional label to obtain a final personalized standard product advertisement. The entire advertisement generation scheme only needs to provide standard product related information, without the need for manual selection of pictures or templates and other operations, and is simple and efficient. In addition, since the AIGC model can provide a large number of high-quality images, the advertisement content and form can be enriched. In addition, the advertisement identity information can be determined from multiple angles through image label analysis, so as to match the most suitable advertisement template. Moreover, through the addition of scene information, a personalized advertisement script that meets the application scene of the standard product can be generated by using the LLM model, which helps to improve the conversion rate of the advertisement. In addition, in one way, in order to control the images generated by the AIGC model, the quality of the standard product image can be audited, and it can be determined whether to perform image synthesis according to the quality audit result. Thus, by adding the image quality detection operation, the image quality and the advertisement quality can be improved.

[0056] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the content of the specification can be implemented, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0057] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0058] Figure 1An implementation scenario schematic diagram of an advertisement generation method provided by an embodiment of the present application is shown.

[0059] Figure 2 A flowchart of an advertisement generation method provided by an embodiment of the present application is shown.

[0060] Figure 3 A structure schematic diagram of an advertisement generation system provided by an embodiment of the present application is shown.

[0061] Figure 4 A working principle schematic diagram of an AIGC image generation engine provided by an embodiment of the present application is shown.

[0062] Figure 5 An instance schematic diagram of an AIGC image generation engine provided by an embodiment of the present application is shown.

[0063] Figure 6 A logic schematic diagram of a base model in an AIGC model provided by an embodiment of the present application is shown.

[0064] Figure 7 A working principle schematic diagram of an image quality audit engine provided by an embodiment of the present application is shown.

[0065] Figure 8 A working principle schematic diagram of an image label system engine provided by an embodiment of the present application is shown.

[0066] Figure 9 A working principle schematic diagram of a script intelligent generation engine provided by an embodiment of the present application is shown.

[0067] Figure 10 An instance schematic diagram of a script intelligent generation engine provided by an embodiment of the present application is shown.

[0068] Figure 11 A working principle schematic diagram of a material intelligent synthesis engine provided by an embodiment of the present application is shown.

[0069] Figure 12 An instance schematic diagram of a material intelligent synthesis engine provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0070] In order to enable persons skilled in the art to better understand the present application scheme, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative labor should be within the scope of protection of the present application. It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0071] In the Internet (such as take-out APP) scenario, putting advertising materials into other external Internet channels as one of the important means to attract new customers plays an important role in user growth and product influence improvement. Image materials as one of the main external advertising material types, the original images mainly come from the APP local image library, and these materials have problems such as poor image quality, many infringing elements such as merchant logos or texts. In addition, current cannot automatically generate fixed category external advertising materials according to the scene requirements, which will lead to the inability to ensure the advertising materials of some best-selling goods in a short period of time when facing special seasons, seasons, events and scenes. In addition, in the process of synthesizing advertising materials, only a few factors such as image color, subject position and main tone are matched with the inherent template, and the synthesized materials lack concise and effective advertising language, resulting in a lot of waste of advertising cost.

[0072] Therefore, an embodiment of the present application provides an advertisement generation method. Referring to Figure 1 , an application scenario diagram of the advertisement generation method provided by an embodiment of the present application is shown. In the embodiment of the present application, only the product information (such as standard dishes) of the product to be generated and the scene information (such as seasons, seasons, events or other scenes) need to be provided, and then a plurality of high-definition, high-color-value, multi-element and rich external advertising material results can be automatically generated by the advertisement generation system, saving the time cost and labor cost of selecting gallery images. In addition, the present solution will also generate advertising language according to the season, season, event or other scene and the dish itself, match a variety of templates, and improve the quality of the generated materials. Finally, the generated advertisement is externally projected through an advertisement page.

[0073] Referring to Figure 2 , a flowchart of an advertisement generation method provided by an embodiment of the present application is shown. The advertisement generation method comprises the following steps S201-S203.

[0074] S201: Obtain product information and scene information of at least one product.

[0075] The product (standard product) refers to the product displayed or sold by the website or APP or applet, for example, each dish displayed on the take-out APP platform. Each product corresponds to a product category, for example, cola belongs to the beverage category, and Kung Pao chicken belongs to the dish category. It can be understood that under the large category, a plurality of sub-categories can be further divided.

[0076] In an embodiment of the present application, a product list including multiple products can be submitted to the advertisement generation system at one time. In the list, each product is distinguished by product information such as category number or product name. At the same time, in order to add a meaning that meets the scene requirements in the generated advertisement later, the product list can also be accompanied by scene information for each product. The scene information includes but is not limited to information about the time of year, season, event, or other scene. It is understood that multiple products belonging to the same category can be set in one product list. Since products belonging to the same category are generally suitable for the same or similar scenes, the scene information can be marked only once in the list.

[0077] In an embodiment of the present application, the standard product information can have multiple forms. One form is that the standard product information only includes standard product text description information or standard product category information. For example, the standard product text description information can be the standard product name or standard product descriptive text, and the standard product category information refers to the category identifier or category name to which the standard product belongs; another form is that in addition to the text description information or standard product category information, the standard product information also includes a standard product sub-image or a standard product outline image, wherein the standard product sub-image refers to an image corresponding to the standard product. For example, assuming that the standard product is "Kung Pao Chicken", the standard product sub-image is the dish image of "Kung Pao Chicken", and the standard product outline image refers to an image that only determines the image outline elements (image size and shape, etc.).

[0078] S202: Based on artificial intelligence content generation (AIGC), a standard product image corresponding to the standard product information is generated, each dimension of the standard product image is measured to obtain label information of each dimension of the standard product image, and the label information and scene information of each dimension are semantically combined to generate advertising copy.

[0079] In step S202 , a standard product image may be first generated based on the AIGC model, and then a multi-dimensional label may be established for the standard product image. Finally, an advertising copy may be generated based on the label information and the scene information.

[0080] In the implementation of generating standard product images based on the AIGC model, it can be based on text-generated images or on a "text + image" combination. As previously mentioned, in step S201, the input standard product information can include a single text form of text / category or a "text + image" form of "text / category + image". Therefore, for different forms of standard product information, standard product images are generated based on different algorithms of the AIGC model.

[0081] AIGC can be used to create images from text. By using text descriptions, text can be converted into images and displayed. AIGC-based text-to-image can provide the following functions: (1) Text-to-image conversion: Convert the input text into an image effect to make the text more vivid; (2) Image customization: Users can choose different colors, fonts, backgrounds, painting styles, etc. to customize their favorite images.

[0082] Therefore, in the case that the product information includes product text description information or product category information, the process of generating a product image corresponding to the product information based on the AIGC model can include: inputting the product text description information or the product category information into the AIGC model, and outputting a product image corresponding to the product text description information according to an interactive prompt-to-image algorithm or a text-to-image algorithm of the AIGC model. The interactive prompt-to-image algorithm may, for example, be a prompt-to-image algorithm, and the text-to-image algorithm may, for example, be a text-to-image algorithm of a latent diffusion model (LDM).

[0083] Based on AIGC, a new image can also be generated from an existing image by modifying the parameters of the existing image. Based on AIGC, a text-to-image function can be provided, in which a user can customize an image with different colors, fonts, backgrounds, drawing styles, etc., based on an input image.

[0084] Therefore, in the case that the product information includes product text description information or product category information and a product seed image, or product text description information or product category information and a product outline image, the process of generating a product image corresponding to the product information based on the AIGC model can include: inputting the product text description information or the product category information and the product seed image, or the product text description information or the product category information and the product outline image into the AIGC model, and outputting a product image corresponding to the product text description information or the product category information and the product seed image, or outputting a product image corresponding to the product text description information or the product category information and the product outline image according to a image-to-image algorithm of the AIGC model. The image-to-image algorithm may, for example, be a image-to-image algorithm based on a stable diffusion model, which supports generating a new image by modifying parameters.

[0085] It can be understood that before generating a product image by the AIGC model, a step of training the AIGC model can also be included. In an implementation, a stable diffusion model open source can be used as a basis, and historical platform product data of a platform to which the product belongs can be used for secondary training to obtain a base model of a local text-to-image model and a local image-to-image model; a low-order adaptation (LORA) of a large language model can be performed on the base model to generate a corresponding LORA model for each product category; and the base model and the plurality of LORA models can constitute the AIGC model.

[0086] In the process of generating image dimension label information based on an image label model, the visual information of the image is converted into semantic information by assigning appropriate labels to the image, which helps better understand and analyze the image. Image labels include image single classification and image multi-label classification. Image single classification is to find a classification label matching the image content from a fixed set of classification labels and assign it to the input image. In the real world, an image often contains rich semantic information, such as multiple targets, scenes, behaviors, etc. Image multi-label classification aims to assign multiple labels to the image to fully express the specific content contained in the image.

[0087] Therefore, based on the image label model, the dimensions of the sample image are measured to obtain label information of each dimension of the sample image, which can include: converting the visual information of the sample image into semantic information based on the image label model image single classification algorithm and / or image multi-label classification algorithm; based on the semantic information, the label information of each dimension is measured, wherein the label information includes any one or more of image color distribution information, image subject information, image category information, image name information, sample name information, and sample associated scene information. Wherein, the image single classification algorithm or the image multi-label classification algorithm can be implemented based on the image classification algorithm of the convolutional neural network.

[0088] In the process of generating an advertisement script based on a large language model combining label information and scene information of each dimension, the label information and scene information of each dimension are input into the large language model to obtain the corresponding syntax and semantics of the information, and the label semantic information and scene semantic information are obtained; the label semantic information and scene semantic information are combined to obtain an advertisement script. Wherein, the large language model (LLM) refers to a deep learning model trained using a large amount of text data, which can generate natural language text or understand the meaning of language text. In the embodiments of the present application, the multi-dimensional label can comprehensively analyze the "identity" information of each dimension of the advertisement, and further limit the corresponding scene of the sample by combining the scene information. Therefore, based on the input of the advertisement identity information and scene information into the LLM model, a highly profiled advertisement script, such as a one-sentence advertisement, can be output.

[0089] S203: According to the label information of each dimension, an advertisement template is matched from an advertisement template library, and the sample image and the advertisement script are synthesized into the advertisement template to generate a sample advertisement.

[0090] After generating the sample image and the advertisement script, the label information of each dimension can be understood, the most suitable advertisement template can be matched in the advertisement template library, and then the sample image and the advertisement script can be embedded into the advertisement template to generate the final advertisement.

[0091] As described previously, the label information of each dimension can be understood as the identity information of each dimension of the advertisement, including but not limited to image color distribution information, image subject information, image category information, image name information, label product name information, and label product associated scene information. Therefore, any one or more of the image color distribution information, the image subject information, the image category information, the image name information, the label product name information, and the label product associated scene information can be used to match the advertisement template according to the matching effect. The matching effect can be determined by machine automatic scoring or user selection.

[0092] After selecting the advertisement template, the label product image and the advertisement script are synthesized into the advertisement template. Specifically, the label product image can be arranged in the main position of the advertisement template, and the advertisement script can be arranged in the non-main position of the advertisement template, or the label product image can be arranged in the main position of the advertisement template, and a short number or text can be extracted from the advertisement script and highlighted in the advertisement template.

[0093] Therefore, the label product image is obtained by AIGC text-to-image or image-to-image, multi-dimensional label establishment is performed on the label product image, the advertisement script is generated based on the label information and the scene information, and finally the label product image and the advertisement script are synthesized into the advertisement template selected according to the multi-dimensional label to obtain the final personalized label product advertisement. The entire advertisement generation scheme only needs to provide label product related information, and does not need to manually select pictures or templates, etc., which is simple and efficient. In addition, since the AIGC model can provide a large number of high-quality images, the advertisement content and form can be enriched. In addition, the image label analysis can determine the advertisement identity information from multiple angles, so as to match the most suitable advertisement template. Moreover, by adding the scene information, the LLM model can generate a personalized advertisement script that meets the application scene of the label product, which helps to improve the conversion rate of the advertisement.

[0094] In an implementation manner, in order to control the image generated by the AIGC model, the quality of the label product image can be audited, and it is determined whether to perform image synthesis according to the quality audit result. Therefore, by adding the image quality detection operation, the image quality and the advertisement quality can be improved.

[0095] Therefore, after generating the product image corresponding to the product information, the following step can also be included: performing quality detection in at least one dimension on the product image to determine the quality detection result in each dimension; the at least one dimension includes image color dimension, picture naturalness dimension, image value dimension, image optical dimension, image definition dimension, product main body integrity dimension, font dimension, special character dimension, and watermark dimension; in the advertisement synthesis process, the quality of each dimension is compared according to the detection threshold of each dimension to determine whether the image quality detection result passes, and the advertisement synthesis is performed only on the image that passes the quality detection.

[0096] In addition, for the product image that does not pass the image quality detection result, image correction can also be performed until a product image that meets the image quality detection is obtained, or a new product image is generated according to the AIGC model again.

[0097] The following takes a dish as an example to illustrate the embodiments of the present application from the perspective of an advertisement generation system.

[0098] Referring to Figure 3 , a structural schematic diagram of an advertisement generation system provided by an embodiment of the present application is shown.

[0099] In the advertisement generation system provided by the embodiment of the present application, first, a user (advertisement generation party) will publish a standard dish list of a key category to be generated according to seasonal, festival, and other scene information, and then enter an automatic material generation process.

[0100] Figure 3 The process of the automatic outdoor material generation based on AIGC under the scene includes the following parts:

[0101] 1) AIGC image generation engine: generate a dish image for each standard dish;

[0102] 2) Image quality review engine: perform quality discrimination in each dimension for the AIGC generation result;

[0103] 3) Image label system engine: calculate each attribute label of the image for the AIGC generation result;

[0104] 4) Copywriting intelligent generation engine: generate an advertisement text using scene information, standard dish information, and related image information;

[0105] 5) Material intelligent synthesis engine: use the output results of the above steps 1) to 4), match a local outdoor advertisement material template, and output an outdoor material.

[0106] The following will be introduced respectively on the principles and examples of AIGC image generation engine, image quality review engine, image label system engine, copywriting intelligent generation engine, and material intelligent synthesis engine.

[0107] The AIGC image generation engine generates more original external material according to popular standard dishes and their pictures (if any) through intelligent prompt prompt generation, AIGC text-to-image, and image-to-image algorithms. Compared with the current selection method from the image library, this method has the characteristics of timeliness and style diversity.

[0108] Referring to Figure 4 , a working principle schematic diagram of the AIGC image generation engine provided by the embodiment of the application is shown. It can support multiple forms of input: pure text input, text + original seed image, text + outline Figure Three image. In addition, input conditions (such as custom generation ratio, custom background, custom perspective, custom style, etc.) can also be added. In the AIGC image generation engine, through shape control, LORA (LORA), perspective control, style control, background replacement, package circle generation, multi-size generation (intelligent cropping), etc. The AIGC generation result of the original image is obtained.

[0109] Referring to Figure 5 , an example schematic diagram of the AIGC image generation engine provided by the embodiment of the application is shown. In this example, the AIGC image generation engine is used to generate dish images of standard dishes. In this example, three input forms (pure text input, text + original seed image, and text + outline image) are shown, and the corresponding AIGC generation results are shown.

[0110] The AIGC image generation engine is composed of a local version of Stablediffusion base large model + multiple auxiliary generation capabilities, wherein the local version of Stablediffusion base large model is obtained by using local platform data for secondary training based on the open source Stablediffusion, which can greatly improve the localization effect.

[0111] Referring to Figure 6 , a logic schematic diagram of the base model in the AIGC model provided by the embodiment of the application is shown. According to Figure 6 , the Stablediffusion model mainly includes two inputs in the process of generating pictures: one is the prompt (prompt), and the other is the seed (the role is to generate a noise image). The noise image generated by the fixed seed and the fixed ratio is fixed, and the model generates pictures based on this. The noise image is not a picture, but a representation in the latent space.

[0112] Various auxiliary capabilities are further post-processing and optimization of large models to fit various landing scenarios. First, image LORA optimization is a model structure with very small parameters used to store fine-tuning parameter changes for a specific style or scene. After adding LORA, the generated results will have better performance in certain specific scenarios. By classifying purposes, a LORA model is trained for each category, achieving the form of 1 Stablediffusion base + N LORA.

[0113] For shape control of the generation process, controlnet can be used to control shape. Controlnet supports different shape information inputs such as canny edge detection, hed contour detection, and depth detection, and applies the results to the generation process. In addition, it also supports image background replacement and editing functions. Using the inpainting technical solution, the background outside the main body is masked, and then a new background image is generated using the diffusion method. It can also splice existing background information to achieve a unified style. On this basis, the relative position of the main body in the original image can be obtained through the main body positioning method, and then a variety of proportional generation results can be obtained through intelligent cropping.

[0114] Since AIGC image generation technology belongs to the generation model, the generation results of this type of model usually have divergence. In order to avoid the generation of images that violate rules, laws and rights from being put into use, a perfect image quality audit engine needs to be designed. In the embodiments of the present application, the image quality audit engine includes multiple dimensions of quality audit, such as image color score, picture natural score, image value score, image optical score, image clarity, commodity main body integrity, font detection, pseudo logo recognition, special character pattern detection, raw food detection, ugly picture detection, watermark detection, picture psoriasis detection, etc. Referring to Figure 7 , a schematic diagram of the working principle of the image quality audit engine provided by the embodiments of the present application is shown. Since the results of the generated model often have uncertainty, it is necessary to monitor the generation results from multiple dimensions, and prohibit the release of ugly pictures, infringing pictures, and pictures that do not conform to values. The role of the image quality audit engine is to judge the quality of all dimensions of the image. Usually, N discriminant models are used to obtain the corresponding results. The quality judgment of various dimensions can be controlled by setting different thresholds in the image synthesis process.

[0115] The image tagging system engine uses various image algorithms to tag various dimensions of an image. This is used for subsequent intelligent matching of advertising templates, intelligent generation of advertising copy, and generation of meal plans. It can also support the systematic construction of local image library tags, so that each image has clear information. The tagging system that the image tagging system engine can build mainly includes: image color distribution, image subject ratio, subject position, subject perspective, subject completeness, subject number, image category, image name, image standard dish, dish common name, dish-related seasonal scene, and other information. See Figure 8 , showing a schematic diagram of the working principle of the image labeling system engine provided by an embodiment of the present application. The image quality labeling system engine measures each dimension of the generated result image and obtains the "identity" information of all dimensions of the generated result. This information provides input for subsequent template synthesis, template color selection, cropping position, cropping size, text placement, and other dimensions.

[0116] Typically, external image advertising materials include not only the image itself, but also interest points such as red envelopes, advertising slogans or recommendations, etc. In the embodiment of the present application, the intelligent copywriting generation engine is used to generate recommendation slogans or advertising slogans, which can refer to scene information such as seasons; food information such as dishes, categories, and flavors; visual information such as image appearance and the number of subjects, and combine relevant information through technologies such as the LLM large language model to generate advertising copy. Figure 9 , shows a schematic diagram of the working principle of the copywriting intelligent generation engine provided by the embodiment of this application. The copywriting intelligent generation engine mainly generates recommended advertising slogans related to the delivery scene and delivery materials. Based on the label information of each dimension of the image, the standard dish name information, and the delivery seasonal scene information, a one-sentence advertising recommendation result is generated based on the large language model. See Figure 10 , shows a schematic diagram of an example of the intelligent copywriting generation engine provided in an embodiment of the present application.

[0117] The intelligent material generation engine analyzes the label system of the images that have passed the quality review, and uses color matching, aesthetic matching, intelligent cropping, edge extension and other technologies to combine the original materials that have passed the quality review with the advertising template, and then adds the advertising promotion copy of the image to synthesize the final external advertising materials. Figure 11 , showing a schematic diagram of the working principle of the material intelligent synthesis engine provided by the embodiment of the present application. The material intelligent synthesis engine is a process of inputting the results of the above steps 1 to 4 (original image, image quality results, image label analysis results, advertising copy), and matching templates from the template library, which mainly includes different analysis angles such as subject position analysis, subject integrity analysis, subject perspective analysis, extension direction analysis, subject color analysis, etc., and generates the final result according to the analysis results and templates. In addition, intelligent recommendations can also be obtained, such asFigure 12 As shown, an example of "spicy beef pizza" is shown, and a schematic diagram of material intelligent synthesis is performed.

[0118] So far, the advertisement generation scheme provided by the embodiments of the present application has technical advantages over the existing scheme. The current synthesis of outdoor advertising materials is mainly in the "material-driven" mode, that is, the generation range of the material mainly refers to the historical indicators such as the click rate and the login rate of the past materials. When facing scenarios such as festivals (Dragon Boat Festival), seasons (Winter Solstice), and events (football matches), it is difficult to effectively synthesize the advertising materials that users are most interested in in a timely manner; in addition, the original images of the current image advertising materials come from the local image library, and most of the pictures have problems such as low definition, poor lighting, mismatched images and texts, existence of merchant logos, existence of too much text, existence of watermarks, etc., which will greatly reduce the aesthetic appearance of the synthesized materials; and the current materials are only simple combinations of original pictures + templates + interest points, lack attractive brief advertising language and guide words, resulting in waste of advertising costs. Therefore, it can be seen that the advertisement generation system provided by the embodiments of the present application can effectively solve the defects of the above-mentioned existing scheme, wherein the AIGC model is used to generate outdoor picture advertising materials, which improves the efficiency and saves the advertising cost compared with the existing process; the AIGC technology is used to generate new image materials, which solves the problems of low definition, ugly pictures, and text and logo infringement elements compared with selecting images from the image library, and increases the diversity and operability of the generation process; multiple algorithms are used to audit image quality factors, and an image label system is established to analyze the quality of the generated image results and build the label system, which avoids excessive human auditing cost and improves efficiency; the LLM large language model is used to generate recommended advertising language of different lengths in combination with multiple dimensions such as outdoor scenarios, standard dish names, image labels, and interest prices, which is more in line with the scene demand and more targeted.

[0119] The embodiments of the present application also provide a storage medium having a computer program stored therein, wherein the computer program is configured to execute the steps in any of the method embodiments described above when running.

[0120] Optionally, in the present embodiment, the above-mentioned storage medium can be configured to store a computer program for executing the following steps:

[0121] Obtaining product information and scene information of at least one product;

[0122] Based on AIGC, generating a product image corresponding to the product information, measuring each dimension of the product image to obtain label information of each dimension of the product image, and performing semantic combination on the label information of each dimension and the scene information to generate an advertising script;

[0123] According to the label information of each dimension, an advertisement template is matched from an advertisement template library, and the product image and the advertisement script are synthesized into the advertisement template to generate a product advertisement.

[0124] Optionally, in the embodiment, the storage medium can include, but is not limited to, a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various storage media that can store computer programs.

[0125] Embodiments of the present application also provide an electronic device including a memory and a processor, the memory storing a computer program, and the processor being configured to execute the computer program to perform the steps in any of the above method embodiments.

[0126] Optionally, the electronic device can further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0127] Optionally, in the embodiment, the processor can be configured to execute the following steps through the computer program:

[0128] Obtaining product information and scene information of at least one product;

[0129] Based on AIGC, generating a product image corresponding to the product information, measuring each dimension of the product image to obtain label information of each dimension of the product image, and performing semantic combination on the label information of each dimension and the scene information to generate an advertisement script;

[0130] According to the label information of each dimension, an advertisement template is matched from an advertisement template library, and the product image and the advertisement script are synthesized into the advertisement template to generate a product advertisement.

[0131] Optionally, specific examples in the embodiment can refer to examples described in the above embodiments and optional implementation manners, and the embodiment will not be described here.

[0132] The serial numbers of the above embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0133] In the above embodiments of the present application, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0134] In several embodiments provided in the present application, it should be understood that the disclosed technology can be implemented by other ways. Among them, the above-described device embodiments are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed units can be indirect coupling or communication connection through some interfaces, units or modules, and can be electrical or other forms.

[0135] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or they can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0136] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0137] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art that contributes to the technical solutions or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0138] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.

Claims

1. A method for generating an advertisement, characterized in that: include: Obtaining product information and scene information of at least one standard product; Based on artificial intelligence content generation AIGC, a standard product image corresponding to the standard product information is generated, each dimension of the standard product image is measured to obtain label information of each dimension of the standard product image, and the label information of each dimension is semantically combined with the scene information to generate advertising copy; According to the label information of each dimension, an advertisement template is matched from an advertisement template library, and the standard product image and the advertisement copy are synthesized into the advertisement template to generate a standard product advertisement.

2. The method according to claim 1, characterized in that The standard product information includes standard product text description information or standard product category information; Based on AIGC, a standard product image corresponding to the standard product information is generated, including: inputting the standard product text description information or standard product category information into the AIGC model, and outputting the standard product image corresponding to the standard product text description information according to the interactive prompt image generation algorithm or text image generation algorithm of the AIGC model.

3. The method according to claim 1, characterized in that The standard product information includes standard product text description information or standard product category information and standard product sub-image, or the standard product information includes standard product text description information or standard product category information and standard product outline image; Based on AIGC, a standard product image corresponding to the standard product information is generated, including: inputting the standard product text description information or the standard product category information and the standard product sub-image, or the standard product text description information or the standard product category information and the standard product outline image, into the AIGC model; and outputting the standard product image corresponding to the standard product text description information or the standard product category information and the standard product sub-image, or outputting the standard product image corresponding to the standard product text description information or the standard product category information and the standard product outline image according to the image-to-image algorithm of the AIGC model.

4. The method according to claim 1, wherein Also includes: The AIGC model is trained, further including: On the basis of the open source stable diffusion model, the historical platform standard product data of the platform to which the standard product belongs is used for secondary training to obtain a base model composed of the local cultural graph model and the local map graph model; Perform low-level adaptation LORA of the large language model for the base model, and generate a corresponding LORA model for each category of target items; The base model and multiple LORA models constitute the AIGC model.

5. The method according to claim 1, wherein Measuring each label dimension of the standard product image to obtain label information of each dimension of the standard product image includes: Converting the visual information of the standard product image into semantic information based on the image single classification algorithm and / or the image multi-label classification algorithm of the image label model; Based on the semantic information, label information of each dimension is measured and obtained, and the label information includes any one or more of image color distribution information, image subject information, image category information, image name information, standard product name information, and standard product associated scene information.

6. The method according to claim 5, characterized in that The step of matching an advertisement template from an advertisement template library according to the tag information of each dimension includes: According to any one or more of the image color distribution information, image subject information, image category information, image name information, standard product name information, and standard product associated scene information, the various templates in the advertising template library are matched, and an advertising template is matched according to the matching effect.

7. The method according to claim 1, characterized in that The semantically combining the label information of each dimension and the scene information to generate an advertisement copy includes: Inputting the label information of each dimension and the scene information into a large language model, obtaining the syntax and semantics corresponding to the information, and obtaining label semantic information and scene semantic information; The tag semantic information and the scene semantic information are combined to obtain the advertising copy.

8. The method according to claim 1, characterized in that The step of synthesizing the standard product image and the advertising copy into the advertising template includes: The standard product image is placed in the main body position of the advertising template, and the advertising copy is placed in the non-main body position of the advertising template. Alternatively, the standard product image is placed in the main body position of the advertising template, and short numbers or text are extracted based on the advertising copy and highlighted in the advertising template.

9. The method according to any one of claims 1 to 8, characterized in that After generating the standard product image corresponding to the standard product information, the method further includes: Performing quality inspection on the standard product image in at least one dimension to determine quality inspection results for each dimension; the at least one dimension includes image color dimension, picture naturalness dimension, image appearance dimension, image optical dimension, image clarity dimension, product body integrity dimension, font dimension, special character dimension, and watermark dimension; During the advertisement synthesis process, the quality of each dimension is compared according to the detection threshold of each dimension to determine whether the image quality detection result passes. Advertisement synthesis is only performed on images that pass the quality detection.

10. The method according to claim 9, characterized in that Also includes: For standard product images that fail the image quality test, image correction is performed to obtain standard product images that meet the image quality test, or the standard product images are regenerated based on the AIGC model.

11. An advertisement generation system, characterized in that: include: An AIGC image generation engine is used to obtain product information and scene information of at least one standard product and generate a standard product image corresponding to the standard product information based on AIGC; An image labeling system engine is used to measure each dimension of the standard product image and obtain label information of each dimension of the standard product image; An intelligent copy generation engine, configured to semantically combine the tag information of each dimension and the scene information to generate advertising copy; The material intelligent synthesis engine is used to match the advertising template from the advertising template library according to the label information of each dimension, and synthesize the standard product image and the advertising copy into the advertising template to generate a standard product advertisement.

12. The system according to claim 11, wherein: The standard product information includes standard product text description information or standard product category information; The AIGC image generation engine is specifically used to: input the standard product text description information or standard product category information into the AIGC model, and output the standard product image corresponding to the standard product text description information according to the interactive prompt image generation algorithm or text image generation algorithm of the AIGC model.

13. The system according to claim 11, wherein: The standard product information includes standard product text description information or standard product category information and standard product sub-image, or the standard product information includes standard product text description information or standard product category information and standard product outline image; The AIGC image generation engine is specifically used to: input the standard product text description information or the standard product category information and the standard product sub-image, or the standard product text description information or the standard product category information and the standard product outline image, into the AIGC model, and output the standard product image corresponding to the standard product text description information or the standard product category information and the standard product sub-image, or output the standard product image corresponding to the standard product text description information or the standard product category information and the standard product outline image according to the image generation algorithm of the AIGC model.

14. The system according to claim 11, wherein: The AIGC image generation engine is also used to: conduct secondary training based on the open source stable diffusion model using the historical platform standard product data of the platform to which the standard product belongs, and obtain a base model composed of a local cultural image model and a local map image model; perform low-order adaptation LORA of the large language model on the base model to generate a corresponding LORA model for each category of target products; and use the base model and multiple LORA models to form the AIGC model.

15. The system according to claim 11, wherein: The image labeling system engine is specifically used to: convert the visual information of the standard product image into semantic information based on the image single classification algorithm and / or image multi-label classification algorithm of the image labeling model; based on the semantic information, measure and obtain label information of each dimension, and the label information includes any one or more of image color distribution information, image subject information, image category information, image name information, standard product name information, and standard product associated scene information.

16. The system according to claim 15, wherein: The material intelligent synthesis engine is specifically used to: match the various templates in the advertising template library according to any one or more of the image color distribution information, image subject information, image category information, image name information, standard product name information, and standard product associated scene information, and match the advertising template according to the matching effect.

17. The system according to claim 11, wherein: The intelligent copy generation engine is specifically used to input the label information of each dimension and the scene information into a large language model, obtain the syntax and semantics corresponding to the information, and obtain label semantic information and scene semantic information; combine the label semantic information and the scene semantic information to obtain the advertising copy.

18. The system according to claim 11, wherein: The material intelligent synthesis engine is specifically used to: place the standard product image in the main position of the advertising template, place the advertising copy in the non-main position of the advertising template, or place the standard product image in the main position of the advertising template, extract short numbers or text based on the advertising copy and highlight them in the advertising template.

19. The system according to any one of claims 11 to 18, characterized in that: Also includes: An image quality audit engine is configured to perform quality inspection on the standard product image in at least one dimension and determine the quality inspection results of each dimension; the at least one dimension includes image color dimension, picture naturalness dimension, image appearance dimension, image optical dimension, image clarity dimension, product body integrity dimension, font dimension, special character dimension, and watermark dimension; The material intelligent synthesis engine is also used to: during the advertisement synthesis process, compare the quality of each dimension according to the detection threshold of each dimension, determine whether the image quality detection result passes, and only synthesize advertisements for images that pass the quality detection.

20. The system according to claim 19, wherein: The material intelligent synthesis engine is also used to: perform image correction on standard product images that fail the image quality detection results to obtain standard product images that meet the image quality detection results, or regenerate standard product images based on the AIGC model.

21. A storage medium, characterized in that The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 10 when executed.

22. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 10.

Images