A large model driven adaptive intelligent marketing poster generation method

Abstract

The present application relates to the technical field of automatic poster generation, and particularly relates to a large model driven adaptive intelligent marketing poster generation method. The method comprises the following steps: obtaining a poster base map, preprocessing the poster base map to obtain a processed poster base map; inputting a title script and a product information script into a content recommendation module to obtain title content; extracting a title color according to the processed poster base map and the content recommendation module; obtaining a main title overlaid on the processed poster base map according to the title content and the title color, and obtaining a title poster; overlaying commodity detailed information on the title poster to obtain a detailed information poster; and overlaying icon element information on the detailed information poster to obtain a finally generated marketing poster. The present application introduces a mainstream deep learning visualization reasoning framework, and realizes the automatic generation of a marketing poster by a system without human intervention.

Description

Technical Field

[0001] This invention relates to the field of automatic poster generation technology, and in particular to a large model-driven adaptive intelligent marketing poster generation method. Background Technology

[0002] Currently, intelligent marketing poster generation mainly relies on the following technical solutions, but all of them have many limitations and are difficult to meet the needs of e-commerce scenarios for high-efficiency, high-quality, and highly consistent batch poster production: Template-based static poster generation systems offer preset layout templates, allowing users to create posters by replacing text and images. Their main drawbacks are: input images need to be manually cropped to a fixed ratio, otherwise stretching, blank spaces, or obscuring background content can easily occur; titles and subtitles need to be manually written, and color schemes rely on subjective judgment, making it difficult to guarantee brand consistency or marketing effectiveness; furthermore, each poster still requires manual intervention, failing to achieve an end-to-end process of "inputting product information → automatically outputting compliant posters."

[0003] Automated poster generation methods based on rule engines are used in some e-commerce platforms and design SaaS platforms. Typical implementations include: building a fixed keyword library and combining it with regular expressions or template slots, such as generating a title like "{Brand}{Category} direct discount {amount} yuan"; simultaneously matching preset color schemes and layouts based on structured tags such as product category and price range. For example, mainstream online design tools in China use this mechanism in their enterprise-level API services—after users input fields such as product name, price, and main image, the system automatically applies industry templates to assemble the poster. However, this method has limitations: the copywriting generation lacks semantic understanding and creative capabilities, only able to truncate, splice, or replace keywords in the input text, making it difficult to generate concise and attractive marketing language based on product characteristics; furthermore, there is a certain disconnect between visual elements and content, easily leading to poor readability or an overall inconsistent style.

[0004] While end-to-end image generation methods based on generative AI are innovative, they still have significant drawbacks: poor controllability, making it difficult to precisely control the embedding of text content, fonts, positions, and brand elements (such as logos and QR codes); the generated content may contain unauthorized materials, errors, or expressions that violate platform guidelines, posing a high risk of commercial compliance; and the reliance on GPU-accelerated inference results in high computational costs. Furthermore, this method cannot reliably guarantee hard constraints such as output size, text length, and color format, typically requiring additional post-processing.

[0005] Existing methods fail to effectively address a core issue: how to achieve end-to-end collaborative automation of copywriting creativity, visual aesthetics, and unobstructed text-image layout without human intervention. Summary of the Invention

[0006] This invention aims to at least solve one of the technical problems existing in related technologies. To this end, this invention provides a large-model-driven adaptive intelligent marketing poster generation method, which enables the system to automatically complete the marketing poster generation process without manual intervention.

[0007] This invention provides a large-model-driven adaptive intelligent marketing poster generation method, comprising: S1: Obtain the poster base image, preprocess the poster base image to obtain the processed poster base image; S2: Input the text of the title into the content recommendation module to obtain the main title content. Based on the processed poster background image and the content recommendation module, extract the main title color in a structured manner. Based on the main title content and the main title color, the main title is overlaid on the processed poster background image to obtain the main title poster. S3: Input the text of obtaining product information into the content recommendation module to obtain the subtitle content. Based on the processed poster background image and the content recommendation module, extract the subtitle color in a structured way. Based on the subtitle content and subtitle color, the subtitle is overlaid on the main title poster to obtain the subtitle poster. S4: Overlay the product details onto the subtitle poster to obtain the detailed information poster; S5: Overlay the icon element information onto the detailed information poster to obtain the final generated marketing poster.

[0008] According to the large model-driven adaptive intelligent marketing poster generation method provided by the present invention, step S1 includes: S11: Perform image normalization and size standardization processing on the poster base image to obtain a standardized image; S12: Encode and convert the standardized image, and input it into the large language model to obtain the top cropping value and the bottom cropping value; S13: Based on the top and bottom crop values, crop or fill to obtain the processed poster base image.

[0009] According to the large model-driven adaptive intelligent marketing poster generation method provided by the present invention, step S2 includes: S21: Obtain the text of the title and construct a conversational request message sequence; S22: Send the conversational request message sequence to the content recommendation module to obtain one or more responses, and take the first result as the main title content. The content recommendation module is a large language model. S23: Calculate the cutting height based on the original size of the processed poster base image, and perform a rectangular cutting operation based on the cutting height to obtain the cutting area; S24: Extract the average color value of the cropped area to obtain the background color, input the background color and the conversational request message sequence into the content recommendation module to obtain one or more responses, and take the first result as the main title color code; S25: Perform a matching operation on the main title color code. If the match is successful, the color corresponding to the main title color code is the main title color; if the match is not successful, the main title color code is specified as #FFFFFF, and the main title color is white. S26: Based on the main title content and main title color, the main title is overlaid on the processed poster background to obtain the main title poster.

[0010] According to the large model-driven adaptive intelligent marketing poster generation method provided by the present invention, step S3 includes: S31: Obtain the text of the product information, determine whether the string length of the product information text is greater than the preset length, if the string length of the product information is not greater than the preset length, the text of the product information is a subtitle, and execute S35; if the string length of the product information is greater than the preset length, execute step S32; S32: Copywriting to obtain product information, constructing a second conversational request message sequence; S33: Send the second conversational request message sequence to the content recommendation module to obtain one or more responses, and take the first result as a candidate for subtitle content. The content recommendation module is a large language model. S34: Determine whether the string length of the candidate subtitle content is greater than the preset length. If the string length of the candidate subtitle content is not greater than the preset length, the candidate subtitle content is the subtitle content, and proceed to S35; if the string length is greater than the preset length, return to step S32. S35: Calculate the subtitle cropping height based on the original size of the processed poster base image, and perform a rectangular cropping operation based on the subtitle cropping height to obtain the subtitle cropping area; S36: Extract the average color value of the subtitle cropping area to obtain the subtitle background color, input the subtitle background color and the second conversational request message sequence into the content recommendation module to obtain one or more responses, and take the first result as the subtitle color code; S37: Perform a matching operation on the subtitle color code. If the match is successful, the color corresponding to the main title color code is the subtitle color; if the match is not successful, the subtitle color code is specified as #2E94D4, and the subtitle color is blue.

[0011] According to the large model-driven adaptive intelligent marketing poster generation method provided by the present invention, step S4 includes: S41: Input the product details into the Text_Image_Multiline_Zho module to obtain the product information content; S42: Adjust the product information content using a linear displacement calculation method and cover it on the subtitle poster to obtain a detailed information poster.

[0012] According to the large model-driven adaptive intelligent marketing poster generation method provided by the present invention, the calculation formula of the linear displacement calculation method is as follows: in, The length of the price text in characters. The average width occupied by each numeric character. The extra fine-tuning spacing is reserved to compensate for gaps between characters or visual alignment deviations. As the initial reference position, This is the result after displacement.

[0013] According to the large model-driven adaptive intelligent marketing poster generation method provided by the present invention, step S5 uses the LayerUtility: ImageBlendAdvance module to perform the image overlay task.

[0014] This invention also provides a large model-driven adaptive intelligent marketing poster generation, including: Background image processing module: used to acquire the poster background image, preprocess the poster background image to obtain the processed poster background image; Main title generation module: This module is used to obtain the text of the title and input it into the content recommendation module to get the main title content. Based on the processed poster background image and the content recommendation module, the main title color is extracted in a structured way. Based on the main title content and the main title color, the main title is overlaid on the processed poster background image to obtain the main title poster. Subtitle generation module: The text of product information is input into the content recommendation module to obtain the subtitle content. Based on the processed poster background image and the content recommendation module, the subtitle color is extracted in a structured way. Based on the subtitle content and subtitle color, the subtitle is overlaid on the main title poster to obtain the subtitle poster. Product details generation module: used to overlay product details onto the subtitle poster to obtain a detailed information poster; Icon element generation module: Used to overlay icon element information onto the detailed information poster to obtain the final generated marketing poster.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a large model-driven adaptive intelligent marketing poster generation method as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a large model-driven adaptive intelligent marketing poster generation method as described above.

[0017] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects: This invention provides a large-model-driven adaptive intelligent marketing poster generation method. This method adopts an end-to-end design concept, providing users with product description text, original images of any size, product details, and icon elements. The system then automatically completes the marketing poster generation process without manual intervention in image cropping, copywriting, color matching, or layout adjustments. The entire process takes no more than a few seconds. A key feature is that this method innovatively utilizes some color features of the background image as input parameters for the large language model, ensuring that the generated title style and text color achieve harmony and unity with the background image in terms of visual aesthetics and semantic expression. This avoids low-readability combinations such as "high-saturation red background with light gray small font," enhancing user visual appeal and information reception.

[0018] This invention includes a main title generation module and a subtitle generation module. These two modules utilize a multimodal large model to accurately identify key visual elements in an image and effectively solve the problem of visual focus loss caused by traditional fixed-ratio cropping by dynamically calculating the optimal cropping ratio. This ensures that marketing posters always focus on the most commercially valuable visual focal point. Furthermore, fixed-height solid-color extension areas are intelligently added to the top and bottom of the image. The color of these extension areas is automatically generated based on the RGB average of the top and bottom 10% areas of the original image, achieving near-perfect integration with the original image, avoiding jarring colors, and enhancing the overall professionalism and aesthetics. The design of the extension areas also considers non-interference with the text display area, preventing reading difficulties caused by image and text overlap. An anomaly fallback mechanism within the system allows this method to robustly handle various types of input images, including but not limited to product images, portraits, landscapes, and images with complex backgrounds.

[0019] This invention constructs a large-scale model invocation and post-processing mechanism for marketing scenarios. This mechanism guides a large language model to generate copy content that conforms to marketing standards under strict constraints through configured system prompts: the main title is precisely limited to 4 to 6 Chinese characters, while the subtitle is intelligently compressed to within 16 characters, retaining the core selling points of the product, significantly improving the professionalism and dissemination efficiency of the copy. Furthermore, the solution innovatively integrates color coordination theory into the generation process—based on the analysis of specific areas of the background image, it automatically recommends foreground text colors and subtitle background colors that possess both high readability and aesthetic value. To ensure output reliability, the system uses regular expressions to perform structured parsing of the color codes returned by the model, supplemented by a default fallback mechanism, ensuring that all structured outputs are legal and robust. This design fully unleashes the semantic understanding and creative generation potential of the large model while effectively balancing the compliance, stability, and usability of the output content, thereby significantly improving the response speed and visual style consistency of the automated marketing system.

[0020] Furthermore, this solution achieves high-precision dynamic layout functionality. By proposing a lightweight linear displacement formula, the system can dynamically adjust the position of the price tag component based on the number of characters in the price numerals and pre-calibrated parameters, without relying on an additional UI layout engine. This mechanism effectively solves the problem of component overlap or misalignment caused by differences in price text width, ensuring clear and readable price information and precise alignment of UI elements. This enhances users' trust in promotional information and increases their willingness to click.

[0021] This solution is based on the mainstream deep learning visualization inference framework ComfyUI. Through its node-based workflow mechanism, it organically integrates modules such as image processing, large model invocation, color analysis, and layout rendering, supporting flexible debugging and efficient deployment.

[0022] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a large-model-driven adaptive intelligent marketing poster generation method provided by the present invention.

[0025] Figure 2 This is the actual effect diagram of step S1.

[0026] Figure 3 This is the actual effect diagram of step S2.

[0027] Figure 4 This is the actual effect diagram of step S3.

[0028] Figure 5 This is the actual effect diagram of step S4.

[0029] Figure 6 This is the actual effect diagram of step S5.

[0030] Figure 7 This is a sample diagram of the output poster of the method of the present invention.

[0031] Figure 8 This is a structural block diagram of a large model-driven adaptive intelligent marketing poster generation system provided by the present invention.

[0032] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention.

[0033] Figure label: 101. Base image processing module; 102. Main title generation module; 103. Subtitle generation module; 104. Product details generation module; 105. Icon element generation module; 810. Processor; 820. Communication interface; 830. Memory; 840. Communication bus. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but cannot be used to limit the scope of this invention.

[0035] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0036] The following is combined with Figures 1 to 9 This invention is described.

[0037] Example like Figure 1 As shown, Figure 1 A flowchart of a large model-driven adaptive intelligent marketing poster generation method, including: S1: Obtain the poster base image, preprocess the poster base image to obtain the processed poster base image; S2: Input the text of the title into the content recommendation module to obtain the main title content. Based on the processed poster background image and the content recommendation module, extract the main title color in a structured manner. Based on the main title content and the main title color, the main title is overlaid on the processed poster background image to obtain the main title poster. S3: Input the text of obtaining product information into the content recommendation module to obtain the subtitle content. Based on the processed poster background image and the content recommendation module, extract the subtitle color in a structured way. Based on the subtitle content and subtitle color, the subtitle is overlaid on the main title poster to obtain the subtitle poster. S4: Overlay the product details onto the subtitle poster to obtain the detailed information poster; S5: Overlay the icon element information onto the detailed information poster to obtain the final generated marketing poster.

[0038] Specifically, such as Figure 2 As shown, step S1 includes: S11: Perform image normalization and size standardization processing on the poster base image to obtain a standardized image; S12: Encode and convert the standardized image, and input it into the large language model to obtain the top cropping value and the bottom cropping value; S13: Based on the top and bottom crop values, crop or fill to obtain the processed poster base image.

[0039] Let the image tensor of the poster background be... , Where H is the original height, W is the original width, and C is the number of image channels. Here, C=3, corresponding to the three color channels: red, green, and blue, i.e., an RGB color image.

[0040] Convert the tensor to PIL (Python Imaging Library) image format and adjust the image size according to the target width. Perform proportional scaling: in The aspect ratio of the original image. This is the scaled width. This is the scaled height, all in pixels. This is the floor sign. Thus, the standardized image tensor is obtained. , .

[0041] This standardized image is converted into Base64 encoded data, and the corresponding multimodal large language model task is executed by calling the multimodal large language model.

[0042] Constructing a standard conversational request message sequence is , Among them: system role messages Used to set model behavior guidelines and task instructions; user role messages Provides the height parameter of the standardized image. .

[0043] The model outputs a cropping ratio that satisfies the constraints: Where: t is the proportion cut off from the top. This represents the percentage to be cut off from the bottom. If the model response does not meet the above constraints or the parsing fails, it will fall back to the default value (t,b)=(0,0) to ensure system robustness.

[0044] Calculate the top coordinates of the actual cropped pixels and bottom coordinates The unit is pixels: The cropping operation is defined as follows: Where x is the horizontal pixel coordinate in the image, and y is the vertical pixel coordinate in the image. For color channel indices, the range of values ​​for y is restricted, while the range for x is unrestricted. The final output is the cropped image tensor. , The formula for the height after cutting is: This step ensures that the core content is compactly preserved in the lower center of the screen, conforming to the distribution pattern of human visual focus (F-shaped browsing mode), while also providing space for the rendering of text above and below.

[0045] Based on edge color statistics, a monochrome extension fill is applied to the top and bottom of the cropped image, extending the monochrome region by a fixed height at the top and bottom, respectively, with the top region also extended by a certain height. ; Lower extension height Define the upper edge region For the top 10% area of ​​the image: It includes the coordinates of all pixels located in the top tenth of the image.

[0046] Similarly, define the lower edge region. This refers to the bottom 10% of the cropped image. Its meaning is symmetrical to that of the upper edge region, representing the pixel coordinates of the bottom tenth of the image.

[0047] Next, the average color values ​​of the upper and lower edge regions in the RGB three channels are calculated respectively. (For the color channel index...) (where k=0 corresponds to the red channel, k=1 corresponds to the green channel, and k=2 corresponds to the blue channel), defined as: in: Let x be the pixel value of the cropped image tensor at position (x, y) and channel k. This represents the total number of pixels contained in the upper edge region. This represents the total number of pixels contained in the lower edge region; The average pixel intensity of the upper edge region in the k-th color channel. This represents the average pixel intensity of the lower edge region on the k-th color channel.

[0048] Convert the above average value to an 8-digit integer color value, and round it to the nearest integer to obtain the fill color: in: To round to the nearest integer; The fill color for the extended area above. Use this as the fill color for the extended area below.

[0049] Subsequently, the processed poster base image is constructed, with a total height of [missing information]. for: Construct the processed poster base image , Its pixel values ​​are defined as follows: The colon ":" assigns values ​​to all three color channels simultaneously; this design matches the color statistical characteristics of the expanded monochrome fill area with the color of the original image's edges, thereby enhancing the overall visual harmony and professionalism of the generated poster.

[0050] like Figure 3 As shown, step S2 includes: S21: Obtain the text of the title and construct a conversational request message sequence; S22: Send the conversational request message sequence to the content recommendation module to obtain one or more responses, and take the first result as the main title content. The content recommendation module is a large language model. S23: Calculate the cutting height based on the original size of the processed poster base image, and perform a rectangular cutting operation based on the cutting height to obtain the cutting area; S24: Extract the average color value of the cropped area to obtain the background color, input the background color and the conversational request message sequence into the content recommendation module to obtain one or more responses, and take the first result as the main title color code; S25: Perform a matching operation on the main title color code. If the match is successful, the color corresponding to the main title color code is the main title color; if the match is not successful, the main title color code is specified as #FFFFFF, and the main title color is white. S26: Based on the main title content and main title color, the main title is overlaid on the processed poster background to obtain the main title poster.

[0051] Specifically, the title copy should take the following form: in: The original input text for the title provided to the user; The input text is a collection of natural language text of arbitrary length; it typically contains marketing-related descriptive information such as product name, core functional features, and usage scenarios; this text is passed in from the upstream processing node and serves as the original basis for subsequent large language model generation tasks.

[0052] This module calls a third-party Large Language Model (LLM) service. The specific configuration is as follows: API key, server endpoint, and construction of a standard conversational request message sequence. .

[0053] message sequence The message is sent to the aforementioned server endpoint via HTTPS, executing a forward inference call. The server returns a structured response object containing one or more candidate responses. This solution uses only the first candidate result, i.e.: in: The final main title text generated for the model; The complete response object returned by the server; This is the first (and highest priority) result generated in the response; This refers to the text content field in the result.

[0054] Get the original size of the image , This is an attribute in the PIL image processing library used to obtain the width and height of an image, returning a tuple (width, height).

[0055] Next, the cropping height used for color extraction is calculated, defined as one-quarter of the total image height (rounded down): in The height of the region cropped from the top of the image (in pixels).

[0056] Then perform the rectangle cropping operation: Where: P.crop() is a cropping method provided by the PIL library; parameters The top-left corner of the cropping area is (0,0), and the bottom-right corner is (0,0). That is, to retain the top quarter of the image.

[0057] Next, the Img2Color-Color Palette Extractor module is used to extract the cropped area. The analysis is performed. This module uses the k-means algorithm to cluster pixel colors and calculates the most representative average color value, denoted as . This serves as the input condition for subsequent large language models.

[0058] in, It is one of the legal colors in the standard six-digit hexadecimal RGB color space.

[0059] Next, using the same third-party large language model service configuration as the "content recommendation module", including model identifier, API key, and service endpoint, a conversational request message sequence is constructed.

[0060] The message sequence is sent to the large language model server, and an inference call is executed. The server returns a structured response object; this solution only uses the content of the first candidate response: in, The raw string returned by the model (which may contain color codes or natural language descriptions). This refers to the text content field in the color results.

[0061] Since large language models may contain additional explanations in their output (such as "recommended color: #333333"), this invention introduces a regular expression post-processing mechanism to extract the string... Extract the substring that conforms to the standard hexadecimal color format. Perform a matching operation; if the match is successful, extract the valid color code, which in this embodiment is #2A3B35; if the match fails, use the default safe fallback strategy and return white as the foreground color. This mechanism ensures that regardless of whether the output of the large language model is standardized, this module always returns a valid, standard hexadecimal color value, thereby guaranteeing the stability and robustness of the downstream text rendering and poster compositing modules.

[0062] like Figure 4 As shown, step S3 includes: S31: Obtain the text of the product information, determine whether the string length of the product information text is greater than the preset length, if the string length of the product information is not greater than the preset length, the text of the product information is a subtitle, and execute S35; if the string length of the product information is greater than the preset length, execute step S32; S32: Copywriting to obtain product information, constructing a second conversational request message sequence; S33: Send the second conversational request message sequence to the content recommendation module to obtain one or more responses, and take the first result as a candidate for subtitle content. The content recommendation module is a large language model. S34: Determine whether the string length of the candidate subtitle content is greater than the preset length. If the string length of the candidate subtitle content is not greater than the preset length, the candidate subtitle content is the subtitle content, and proceed to S35; if the string length is greater than the preset length, return to step S32. S35: Calculate the subtitle cropping height based on the original size of the processed poster base image, and perform a rectangular cropping operation based on the subtitle cropping height to obtain the subtitle cropping area; S36: Extract the average color value of the subtitle cropping area to obtain the subtitle background color, input the subtitle background color and the second conversational request message sequence into the content recommendation module to obtain one or more responses, and take the first result as the subtitle color code; S37: Perform a matching operation on the subtitle color code. If the match is successful, the color corresponding to the main title color code is the subtitle color; if the match is not successful, the subtitle color code is specified as #2E94D4, and the subtitle color is blue.

[0063] Specifically, let's assume the text for inputting product information is a string. , S is any set of natural language text. This is the string representing the product or event name you are entering. The system first calculates its character length (in Unicode characters, including Chinese characters, letters, numbers, punctuation, and spaces): And set the maximum number of allowed characters. .

[0064] in: It is a Python function used to calculate the length of a string, returning the number of characters. The maximum number of characters allowed; for The length of the characters.

[0065] like Then no processing is needed, just directly... As output; like This triggers the large language model compression process. The large language model service is also used, with the specific configuration and steps identical to those for the main title processing. A conversational request message sequence is then constructed. .

[0066] The user role message is the actual product name entered, and the first candidate response is retrieved: in, This is the text content field in the actual input product name result.

[0067] The final output is a string. : Background color recommendation module: Crop the background image area at the subtitle location, calculate the height of the cropped area, and define it as the range of 26.5% to 33.5% of the image area in the vertical direction (this range corresponds to the visual area where the subtitle is usually located in poster design), and then perform a rectangular cropping operation: in: The row number of the upper boundary pixels of the cropped area, starting from the top of the image; This is the row number of the lower boundary pixels of the cropped area, excluding this row; It is an empirical scaling constant, representing the position taken from the top of the image downwards at 26.5%. It is another empirical scaling constant, representing the position taken from the top of the image downwards at 33.5%.

[0068] Then, the Img2Color-Color Palette Extractor module is used to perform color statistics on the cropped area, calculate its average dominant color, and output it in a standard six-digit hexadecimal format, denoted as . This serves as the color input condition for subsequent large language models.

[0069] The background color for this section is in string format: ,in, The background color for the subtitle area is a valid color value in the standard six-digit hexadecimal RGB color space. Next, a third-party large language model service is used, with the specific configuration the same as in step S2.

[0070] Send the message sequence to the large language model server, obtain the response object, and extract the content of the first candidate reply: If a match is successful, the valid color code is extracted. In this embodiment, the color is #2A5B6F. If a match fails (i.e., no valid color format is found), the preset default safe color is enabled. The default value is a medium-saturation blue, which, according to actual testing, provides good contrast with white text on most backgrounds and is visually appealing.

[0071] like Figure 5 As shown, step S4 includes: S41: Input the product details into the Text_Image_Multiline_Zho module to obtain the product information content; S42: Adjust the product information content using a linear displacement calculation method and cover it on the subtitle poster to obtain a detailed information poster.

[0072] This invention employs a linear displacement calculation method to dynamically calculate the new position of the "number of characters" component based on the number of digits N in the price numerals. The calculation formula is as follows: in, The length of the price text in characters. The average width occupied by each numeric character. The extra fine-tuning spacing is reserved to compensate for gaps between characters or visual alignment deviations. As the initial reference position, This is the result after displacement.

[0073] like Figure 6 As shown, step S5 uses the LayerUtility: ImageBlendAdvance module to perform the image overlay task. This module can overlay multiple image layers with different blending modes and opacities. It supports a variety of advanced features, including: position adjustment: allowing users to specify the placement of icons, whether using absolute coordinates or relative to certain feature points of the background image; blending mode selection, providing a variety of image blending modes, such as Normal, Multiply, and Screen, to achieve the best visual effect.

[0074] like Figure 7 As shown, Figure 7 This is a sample image of the output poster of the method of the present invention.

[0075] like Figure 8 As shown, the present invention also provides a large model-driven adaptive intelligent marketing poster generation device, comprising the following modules: Background image processing module 101: used to acquire the poster background image, preprocess the poster background image to obtain the processed poster background image; Main title generation module 102: Used to obtain the text of the title and input it into the content recommendation module to obtain the main title content. Based on the processed poster background image and the content recommendation module, the main title color is extracted in a structured manner. Based on the main title content and the main title color, the main title is overlaid on the processed poster background image to obtain the main title poster. Subtitle generation module 103: The text input for obtaining product information is fed into the content recommendation module to obtain the subtitle content. The subtitle color is extracted in a structured manner based on the processed poster background image and the content recommendation module. The subtitle is then overlaid on the main title poster based on the subtitle content and subtitle color to obtain the subtitle poster. Product details generation module 104: used to overlay product details onto the subtitle poster to obtain a detailed information poster; Icon element generation module 105: used to overlay icon element information onto the detailed information poster to obtain the final generated marketing poster.

[0076] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a large model-driven adaptive intelligent marketing poster generation method, which includes: S1: Obtain the poster base image, preprocess the poster base image to obtain the processed poster base image; S2: Input the text of the title into the content recommendation module to obtain the main title content. Based on the processed poster background image and the content recommendation module, extract the main title color in a structured manner. Based on the main title content and the main title color, the main title is overlaid on the processed poster background image to obtain the main title poster. S3: Input the text of obtaining product information into the content recommendation module to obtain the subtitle content. Based on the processed poster background image and the content recommendation module, extract the subtitle color in a structured way. Based on the subtitle content and subtitle color, the subtitle is overlaid on the main title poster to obtain the subtitle poster. S4: Overlay the product details onto the subtitle poster to obtain the detailed information poster; S5: Overlay the icon element information onto the detailed information poster to obtain the final generated marketing poster.

[0077] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a 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 several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0078] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described large-model-driven adaptive intelligent marketing poster generation method, the method comprising: S1: Obtain the poster base image, preprocess the poster base image to obtain the processed poster base image; S2: Input the text of the title into the content recommendation module to obtain the main title content. Based on the processed poster background image and the content recommendation module, extract the main title color in a structured manner. Based on the main title content and the main title color, the main title is overlaid on the processed poster background image to obtain the main title poster. S3: Input the text of obtaining product information into the content recommendation module to obtain the subtitle content. Based on the processed poster background image and the content recommendation module, extract the subtitle color in a structured way. Based on the subtitle content and subtitle color, the subtitle is overlaid on the main title poster to obtain the subtitle poster. S4: Overlay the product details onto the subtitle poster to obtain the detailed information poster; S5: Overlay the icon element information onto the detailed information poster to obtain the final generated marketing poster.

[0079] While this disclosure has been described with reference to several specific embodiments, it should be understood that this disclosure is not limited to the specific embodiments disclosed. This disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A large-model-driven adaptive intelligent marketing poster generation method, characterized in that, include: S1: Obtain the poster base image, preprocess the poster base image to obtain the processed poster base image; S2: Input the text of the title into the content recommendation module to obtain the main title content. Based on the processed poster background image and the content recommendation module, extract the main title color in a structured manner. Based on the main title content and the main title color, the main title is overlaid on the processed poster background image to obtain the main title poster. S3: Input the text of obtaining product information into the content recommendation module to obtain the subtitle content. Based on the processed poster background image and the content recommendation module, extract the subtitle color in a structured way. Based on the subtitle content and subtitle color, the subtitle is overlaid on the main title poster to obtain the subtitle poster. S4: Overlay the product details onto the subtitle poster to obtain the detailed information poster; S5: Overlay the icon element information onto the detailed information poster to obtain the final generated marketing poster.

2. The large-model-driven adaptive intelligent marketing poster generation method according to claim 1, characterized in that, Step S1 includes: S11: Perform image normalization and size standardization processing on the poster base image to obtain a standardized image; S12: Encode and convert the standardized image, and input it into the large language model to obtain the top cropping value and the bottom cropping value; S13: Based on the top and bottom crop values, crop or fill to obtain the processed poster base image.

3. The large-model-driven adaptive intelligent marketing poster generation method according to claim 1, characterized in that, Step S2 includes: S21: Obtain the text of the title and construct a conversational request message sequence; S22: Send the conversational request message sequence to the content recommendation module to obtain one or more responses, and take the first result as the main title content. The content recommendation module is a large language model. S23: Calculate the cutting height based on the original size of the processed poster base image, and perform a rectangular cutting operation based on the cutting height to obtain the cutting area; S24: Extract the average color value of the cropped area to obtain the background color, input the background color and the conversational request message sequence into the content recommendation module to obtain one or more responses, and take the first result as the main title color code; S25: Perform a matching operation on the main title color code. If the match is successful, the color corresponding to the main title color code is the main title color; if the match is not successful, the main title color code is specified as white #FFFFFF. S26: Based on the main title content and main title color, the main title is overlaid on the processed poster background to obtain the main title poster.

4. The large-model-driven adaptive intelligent marketing poster generation method according to claim 1, characterized in that, Step S3 includes: S31: Obtain the text of the product information, determine whether the string length of the product information text is greater than the preset length, if the string length of the product information is not greater than the preset length, the text of the product information is a subtitle, and execute S35; if the string length of the product information is greater than the preset length, execute step S32; S32: Copywriting to obtain product information, constructing a second conversational request message sequence; S33: Send the second conversational request message sequence to the content recommendation module to obtain one or more responses, and take the first result as a candidate for subtitle content. The content recommendation module is a large language model. S34: Determine whether the string length of the candidate subtitle content is greater than the preset length. If the string length of the candidate subtitle content is not greater than the preset length, the candidate subtitle content is the subtitle content, and proceed to S35; if the string length is greater than the preset length, return to step S32. S35: Calculate the subtitle cropping height based on the original size of the processed poster base image, and perform a cropping operation based on the subtitle cropping height to obtain the subtitle cropping area; S36: Extract the average color value of the subtitle cropping area to obtain the subtitle background color, input the subtitle background color and the second conversational request message sequence into the content recommendation module to obtain one or more responses, and take the first result as the subtitle color code; S37: Perform a matching operation on the subtitle color code. If the match is successful, the color corresponding to the main title color code is the subtitle color; if the match is not successful, the subtitle color code is specified as blue #2E94D4.

5. The large-model-driven adaptive intelligent marketing poster generation method according to claim 1, characterized in that, Step S4 includes: S41: Input the product details into the Text_Image_Multiline_Zho module to obtain the product information content; S42: Adjust the product information content using a linear displacement calculation method and cover it on the subtitle poster to obtain a detailed information poster.

6. The method for generating adaptive intelligent marketing posters driven by a large model according to claim 5, wherein the calculation formula for the linear displacement calculation method is: in, The length of the price text in characters. The average width occupied by each numeric character. The extra fine-tuning spacing is reserved to compensate for gaps between characters or visual alignment deviations. As the initial reference position, This is the result after displacement.

7. The large-model-driven adaptive intelligent marketing poster generation method according to claim 1, characterized in that, Step S5 uses the LayerUtility: ImageBlendAdvance module to perform the image overlay task.

8. A large-model-driven adaptive intelligent marketing poster generation system, used to execute the large-model-driven adaptive intelligent marketing poster generation method as described in any one of claims 1 to 7, characterized in that, include: Background image processing module: used to acquire the poster background image, preprocess the poster background image to obtain the processed poster background image; Main title generation module: This module is used to obtain the text of the title and input it into the content recommendation module to get the main title content. Based on the processed poster background image and the content recommendation module, the main title color is extracted in a structured way. Based on the main title content and the main title color, the main title is overlaid on the processed poster background image to obtain the main title poster. Subtitle generation module: The text of product information is input into the content recommendation module to obtain the subtitle content. Based on the processed poster background image and the content recommendation module, the subtitle color is extracted in a structured way. Based on the subtitle content and subtitle color, the subtitle is overlaid on the main title poster to obtain the subtitle poster. Product details generation module: used to overlay product details onto the subtitle poster to obtain a detailed information poster; Icon element generation module: Used to overlay icon element information onto the detailed information poster to obtain the final generated marketing poster.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the large model-driven adaptive intelligent marketing poster generation method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the large model-driven adaptive intelligent marketing poster generation method as described in any one of claims 1 to 7.

Images