Image generation method, computer device, readable storage medium and program product

Abstract

The application relates to an image generation method, a computer device, a readable storage medium and a program product. The method comprises the following steps: identifying objects in image materials to obtain description information of the objects; at least one preset difference type corresponding to the image materials is acquired, and a to-be-modified object corresponding to each preset difference type is determined from the objects; an image modification prompt word of the image materials is generated according to the description information of the to-be-modified object and the preset difference type; the image modification prompt word and the image materials are input into an image redrawing model to obtain a target image corresponding to the image materials; an association relationship between the image materials and the target image is established; the association relationship is used for creating an image difference annotation task; and the image difference annotation task is used for enabling a user to find out the difference between the target image and the image materials. The method can improve the generation efficiency of the find-the-difference picture.

Description

Technical Field

[0001] This application relates to the field of game content generation technology, and in particular to an image generation method, computer device, readable storage medium, and program product. Background Technology

[0002] In the traditional game content production process, the creation of images for "spot the difference" games (also known as spot-the-difference games) relies on manual creation by artists. Specifically, artists first need to draw a basic scene image, and then manually modify several local details in that basic scene image to generate another comparison image. There are several differences between these two images, and players need to find these differences during gameplay.

[0003] During the creation of game images, artists, based on their creative concepts, manually manipulate the original images using image editing software, adding or removing objects, modifying colors, moving positions, and changing textures to generate "spot the difference" images. The entire process requires a significant investment of time and effort from the artists; a single, satisfactory "spot the difference" image often takes several hours or even longer to complete, resulting in low generation efficiency. Summary of the Invention

[0004] Therefore, it is necessary to provide an image generation method, apparatus, computer device, computer-readable storage medium, and computer program product that can improve the generation efficiency of spot-the-difference images, in response to the above-mentioned technical problems.

[0005] In a first aspect, this application provides an image generation method, including:

[0006] Objects in the image material are identified to obtain descriptive information for each object;

[0007] At least one preset difference type corresponding to the image material is obtained, and the object to be modified corresponding to each preset difference type is determined from each of the objects; the preset difference type is used to indicate the modification method of the object to be modified in the image material;

[0008] Based on the description information corresponding to the object to be modified and the preset difference type, generate image modification prompts for the image material;

[0009] The image modification prompt and the image material are input into the image redrawing model to obtain the target image corresponding to the image material; there is at least one difference between the target image and the image material.

[0010] Establish a relationship between the image material and the target image; the relationship is used to create an image difference annotation task; the image difference annotation task is used by the user to find the differences between the target image and the image material.

[0011] In one embodiment, generating image modification prompts for the image material based on the description information corresponding to the object to be modified and the preset difference type includes:

[0012] Based on the description information corresponding to the object to be modified and the preset difference type, difference point planning information of the image material is generated; the difference point planning information includes at least one difference point; each difference point includes an object to be modified, a preset difference type corresponding to the object to be modified, and the position information of the object to be modified in the image material.

[0013] The difference point planning information is input into the prompt word generation model to obtain the image modification prompt words for the image material.

[0014] In one embodiment, the step of inputting the image modification prompt and the image material into the image redrawing model to obtain the target image corresponding to the image material includes:

[0015] The image modification prompts and the image material are input into the image redrawing model to obtain local modified images corresponding to each difference point; there is one difference between each local modified image and the image material.

[0016] From each of the locally modified images, locally modified images that meet the preset quality conditions are selected as candidate modified images;

[0017] The differences contained in each of the candidate modified images are fused into the image material to obtain the target image.

[0018] In one embodiment, the step of selecting locally modified images that meet preset quality conditions from each of the locally modified images as candidate modified images includes:

[0019] Obtain quality quantification information for each of the locally modified images; the quality quantification information includes at least one of the following: difference significance assessment result, difference reasonableness assessment result, and image quality assessment result;

[0020] From each of the locally modified images, locally modified images whose quality quantification information meets the preset quality conditions are selected as candidate modified images.

[0021] In one embodiment, fusing the differences contained in each of the candidate modified images into the image material to obtain the target image includes:

[0022] Determine the number of candidate modified images;

[0023] If the number of images is greater than or equal to the preset number of differences required, a target modified image that matches the preset number of differences required is selected from each of the candidate modified images;

[0024] The differences contained in each of the target modified images are fused into the image material to obtain the target image.

[0025] In one embodiment, after determining the number of candidate modified images, the method further includes:

[0026] If the number of images is less than the preset difference requirement and the number of images is greater than or equal to the preset fault tolerance threshold, the difference points corresponding to the locally modified images that do not meet the preset quality conditions are optimized to obtain the updated difference point planning information. Then, the process returns to the step of inputting the difference point planning information into the prompt word generation model to obtain the image modification prompt words for the image material.

[0027] If the number of images is less than the preset fault tolerance threshold, the image material is discarded, the next image material is used as the updated image material, and the process returns to the step of identifying objects in the image material and obtaining description information of each object.

[0028] In one embodiment, after obtaining the target image corresponding to the image material, the method further includes:

[0029] From the target images, identify the rejection images whose quality review result is unsatisfactory. Count the first occurrence frequency of each preset difference type in each rejection image. Identify preset difference types whose first occurrence frequency is higher than a first frequency threshold as risk difference types, reducing the probability of assigning these risk difference types to the next image material; and / or,

[0030] The second occurrence frequency of each preset difference type in each target image is statistically analyzed, and the preset difference types whose second occurrence frequency is higher than the second frequency threshold are determined as high-frequency difference types, thus expanding the modification method of the high-frequency difference type indication; and / or,

[0031] From the target images, identify the rejection images whose quality audit result is not passed, and replace the preset difference type corresponding to the rejection image with the target difference type, wherein the frequency of the target difference type in historical rejection images is lower than that of the preset difference type.

[0032] Secondly, this application also provides an image generation apparatus, comprising:

[0033] The recognition module is used to identify objects in the image material and obtain descriptive information of each object;

[0034] The determination module is used to obtain at least one preset difference type corresponding to the image material, and determine the object to be modified corresponding to each preset difference type from each of the objects; the preset difference type is used to indicate the modification method of the object to be modified in the image material;

[0035] The generation module is used to generate image modification prompts for the image material based on the description information corresponding to the object to be modified and the preset difference type.

[0036] The redrawing module is used to input the image modification prompt and the image material into the image redrawing model to obtain the target image corresponding to the image material; there is at least one difference between the target image and the image material.

[0037] A module is established to establish a relationship between the image material and the target image; the relationship is used to create an image difference annotation task; the image difference annotation task is used by the user to find the differences between the target image and the image material.

[0038] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0039] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0040] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0041] The aforementioned image generation methods, apparatus, computer devices, computer-readable storage media, and computer program products can achieve automated understanding of image content by identifying objects in image materials and obtaining descriptive information; they can achieve intelligent planning of difference points by obtaining preset difference types and determining the objects to be modified; they can achieve automated creation of difference images by generating image modification prompts and using image redrawing models to generate target images; and they can apply the generated image pairs to spot-the-difference game scenarios by establishing a correlation between image materials and target images to create image difference annotation tasks. The entire process eliminates the need for manual drawing and modification of images, improving the generation efficiency of spot-the-difference game images and enabling the generation of a large number of diverse and high-quality spot-the-difference images to meet the needs of large-scale production of game content. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is an application environment diagram of an image generation method in one embodiment;

[0044] Figure 2 This is a flowchart illustrating an image generation method in one embodiment;

[0045] Figure 3 This is a logic diagram of an image generation method in one embodiment;

[0046] Figure 4 This is a flowchart illustrating an image generation method in another embodiment;

[0047] Figure 5 This is a structural block diagram of an image generation apparatus in one embodiment;

[0048] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] The image generation method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server.

[0051] Terminal 102 identifies objects in the image material and obtains descriptive information for each object; Terminal 102 acquires at least one preset difference type corresponding to the image material, and determines the object to be modified corresponding to each preset difference type from each object; the preset difference type is used to indicate the modification method for the object to be modified in the image material; Terminal 102 generates image modification prompts for the image material based on the descriptive information corresponding to the object to be modified and the preset difference type; Terminal 102 inputs the image modification prompts and the image material into the image redrawing model to obtain the target image corresponding to the image material; there is at least one difference between the target image and the image material; Terminal 102 establishes a correlation between the image material and the target image; the correlation is used to create an image difference annotation task; the image difference annotation task is used by the user to find the differences between the target image and the image material.

[0052] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. Portable wearable devices can be smartwatches, smart bracelets, head-mounted displays, etc. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0053] In one exemplary embodiment, such as Figure 2 As shown, an image generation method is provided, which can be applied to... Figure 1 Taking terminal 102 as an example, the explanation includes:

[0054] Step S202: Identify the objects in the image material to obtain description information of each object.

[0055] Here, image assets refer to the basic scene images used to generate the spot-the-difference game. Optionally, image assets can be scene images generated from text prompts using a text-based image model, or they can be image assets pre-stored in a database.

[0056] In practical implementation, to generate image materials in batches, a batch of preset text prompts describing diverse scenes (such as scene type, main objects, environmental atmosphere, visual style, etc.) can be input. Then, the text-to-image model is loaded all at once, continuously generating a large number of high-resolution, high-quality base scene images as image materials, thus obtaining well-composed and clear "canvases." The text-to-image model refers to an artificial intelligence model capable of generating corresponding images based on text descriptions. It parses the semantic information in the text prompts, understands the scene content and visual requirements described by the prompts, and generates corresponding high-resolution images accordingly. By using the text-to-image model, a large number of diverse, high-quality image materials can be generated quickly, improving the efficiency and scale of material acquisition.

[0057] Here, "object" refers to the various identifiable visual elements contained in the image material, including objects, regions, backgrounds, etc.; "descriptive information" refers to the information used to describe the characteristics of the object, including attributes such as the object's category, color, position, size, shape, and texture.

[0058] For example, the terminal can invoke a visual understanding model to perform scene analysis on image materials. A visual understanding model is an artificial intelligence model that can recognize image content and understand the semantic information of each object in the image. After the terminal inputs the image materials into the visual understanding model, the model analyzes the image materials, identifies each object contained in the image materials, and generates corresponding descriptive information for each object. For example, for an image material of an indoor scene containing a table, a cup, and a book, the visual understanding model can identify the "table" object and generate the descriptive information "table, white, coordinate region (x, y)".

[0059] Step S204: Obtain at least one preset difference type corresponding to the image material, and determine the object to be modified corresponding to each preset difference type from each object.

[0060] The preset difference type is used to indicate the modification method for the object to be modified in the image material.

[0061] Among them, preset difference types refer to predefined operation types used to modify image materials to produce differences. Preset difference types include, but are not limited to: change attribute type, add object type, delete object type, and move position type. Specifically, the change attribute type is used to indicate the modification of a certain attribute of an object, such as changing the object's color, texture, or shape; the add object type is used to indicate the addition of a new object to a certain area of ​​the image material; the delete object type is used to indicate the removal of an object from the image material; and the move position type is used to indicate the change of the object's position coordinates within the image material.

[0062] The object to be modified refers to the object selected to perform the modification operation corresponding to the preset difference type. Optionally, the object to be modified can be randomly selected from various objects in the image material.

[0063] In practice, the terminal can assign at least one logically reasonable preset difference type to image materials from a pre-established difference type library. The difference type library stores various available preset difference types along with their corresponding assignment probabilities, applicable scenarios, constraints, and other information. The terminal can select at least one suitable preset difference type from the library based on factors such as the content characteristics and object distribution of the image materials. For example, for a complex scene containing multiple objects, three preset difference types can be assigned to the image material: changing attribute type, adding object type, and deleting object type.

[0064] For example, after obtaining the preset difference type, the terminal needs to identify the corresponding object to be modified for each preset difference type from the various objects identified in the image material.

[0065] As an example, the object to be modified can be determined based on its characteristics, location, and predefined constraints of the difference type. For instance, to change the attribute type, an object with obvious color characteristics can be selected as the object to be modified; to delete an object type, a non-structural object that will not disrupt the overall rationality of the scene after removal can be selected as the object to be modified; to add an object type, a blank area or background area can be selected as the target location for the object to be added.

[0066] As another example, the corresponding object to be modified can also be randomly selected from each object for each preset difference type.

[0067] Step S206: Based on the description information corresponding to the object to be modified and the preset difference type, generate image modification prompts for the image material.

[0068] Image modification prompts are text instructions used to guide the image redrawing model in modifying image materials. Image modification prompts include a specific description of the desired effect, such as the visual characteristics the modified object should exhibit, including color, shape, and texture.

[0069] In practice, the terminal generates image modification prompts based on the description of the object to be modified and the preset difference type. These prompts accurately guide the image redrawing model to make modifications. For example, for the object to be modified, "blue ceramic cup," and the preset difference type, "change attribute type - change color," the image modification prompt "a bright red ceramic cup" can be generated. For the preset difference type, "add object type," and a specified area, the image modification prompt "a green potted plant on the corner of the table" can be generated. The generated image modification prompts clearly and accurately describe the visual effect after modification, enabling the image redrawing model to generate high-quality modification results.

[0070] Step S208: Input the image modification prompt and image material into the image redrawing model to obtain the target image corresponding to the image material.

[0071] There is at least one difference between the target image and the image material.

[0072] The image redrawing model refers to an artificial intelligence model that can locally modify the original image based on the input original image and modification prompts to generate a new image. The target image refers to the image obtained after modification by the image redrawing model that differs from the original image. This target image, together with the original image, will form the image pair in the spot-the-difference game.

[0073] In practice, the terminal inputs both the image modification prompt and the image material into the image redrawing model. Based on the prompt, the model modifies the image material accordingly to generate the target image. For example, upon receiving the image material and the prompt "a bright red ceramic cup," the model will change the original blue ceramic cup in the image material to a red one, while keeping the rest of the image unchanged, thus generating the target image. The generated target image must have at least one obvious visual difference from the original image; this difference is what the player in the spot-the-difference game needs to find.

[0074] Step S210: Establish the association between the image material and the target image.

[0075] Among them, the association relationship is used to create image difference annotation tasks; the image difference annotation tasks are used to help users find the differences between the target image and the image material.

[0076] In this context, the association relationship refers to a data structure used to pair image materials with target images and record their correspondence. The association relationship may include metadata such as the identification information of the image materials, the identification information of the target images, the location coordinates of the differences, and the type description of the differences.

[0077] Among them, the image difference annotation task refers to the task presented to players in the "spot the difference" game, where players need to compare the image material and the target image to find the differences between them.

[0078] In practice, to establish the association between image assets and target images, unique identifiers are assigned to each image asset and target image. These identifiers and difference information are stored in a database, forming an association record. Based on this association, the game system can create image difference annotation tasks. When a user starts the game, the system retrieves the corresponding image assets and target image according to the association, displaying the two images side-by-side or alternately. The user needs to carefully observe and click to annotate the differences between the two images. The game system compares the user's annotations with the actual difference locations recorded in the association to determine if the user has found the correct differences and provides corresponding game feedback.

[0079] In the above image generation method, by identifying objects in the image material and obtaining descriptive information, automated understanding of image content can be achieved; by obtaining preset difference types and determining the objects to be modified, intelligent planning of difference points can be achieved; by generating image modification prompts and using image redrawing models to generate target images, automated creation of difference images can be achieved; by establishing the association between image material and target image to create image difference annotation tasks, the generated image pairs can be applied to spot-the-difference game scenarios. The entire process does not require manual participation in the manual drawing and modification of images, improving the generation efficiency of spot-the-difference game images, and generating a large number of diverse and high-quality spot-the-difference images to meet the needs of large-scale production of game content.

[0080] In another embodiment, generating image modification prompts for image materials based on the description information corresponding to the object to be modified and a preset difference type includes: generating difference point planning information for image materials based on the description information corresponding to the object to be modified and a preset difference type; and inputting the difference point planning information into the prompt generation model to obtain image modification prompts for image materials.

[0081] The difference point planning information refers to a structured description of all modification operations required for the image material. The difference point planning information includes at least one difference point; each difference point includes an object to be modified, a preset difference type corresponding to the object, and the object's location information within the image material.

[0082] Here, the difference point refers to an independent modification operation and its related parameters; the location information refers to the spatial coordinates of the object to be modified in the image material, which can be represented by the coordinates of the upper left and lower right corners of the bounding box, or by the coordinates of the four corner points of the bounding box, or by the coordinates of the center point and the width and height.

[0083] In practice, the terminal generates difference point planning information for image materials based on the description information of the object to be modified and the preset difference type. The difference point planning information organizes multiple difference points in a structured manner. Each difference point explicitly specifies an object to be modified, the preset difference type corresponding to that object, and the object's location information within the image material. For example, for an image material of an indoor scene, difference point planning information containing two difference points can be generated: the first difference point is "to change the attribute type of the blue ceramic cup located in the coordinate region (100, 150) to (200, 250), changing its color to red"; the second difference point is "to delete the object type of the book located in the coordinate region (300, 100) to (400, 200). By generating difference point planning information, the modification methods of image materials can be managed in a structured and traceable manner.

[0084] Among them, the prompt word generation model refers to an artificial intelligence model that can convert structured difference point planning information into image modification prompt words that are described in natural language. The prompt word generation model can be a large language model that can understand the semantic content in the difference point planning information and generate accurate prompt word text that meets the input requirements of the image redrawing model.

[0085] In practice, the difference point planning information is input into the prompt word generation model. The model parses each difference point in the difference point planning information, understanding the object to be modified, the preset difference type, and the location information contained in each difference point. It then converts this structured information into image modification prompt words in natural language. For example, for the difference point planning information "change the blue ceramic cup to red," the prompt word generation model can generate the image modification prompt word "a brightred ceramic cup." The image modification prompt words generated by the prompt word generation model accurately convey the modification intention and contain sufficient detail to enable the image redrawing model to generate high-quality, visually natural modification results.

[0086] The technical solution of this embodiment, by generating difference point planning information, can record the difference points corresponding to image materials in a structured form. At the same time, by utilizing the language generation capability of the prompt word generation model, it generates diverse prompt word descriptions for each difference point, improving the richness and naturalness of the image modification results. This solves the problem that the limited types of differences designed by humans result in low fun in spot-the-difference games.

[0087] In another embodiment, inputting image modification prompts and image materials into an image redrawing model to obtain a target image corresponding to the image materials includes: inputting image modification prompts and image materials into an image redrawing model to obtain local modification images corresponding to each difference point; there is one difference between each local modification image and the image material; selecting local modification images that meet preset quality conditions from each local modification image as candidate modification images; and fusing the differences contained in each candidate modification image into the image material to obtain the target image.

[0088] In this context, a locally modified image refers to an intermediate image obtained by modifying an image source material only at one point of difference. For difference point planning information containing multiple points of difference, the terminal generates a locally modified image for each point of difference. Each locally modified image differs from the image source material only in the region corresponding to the point of difference, while other regions remain consistent with the image source material.

[0089] In practice, image modification prompts and image materials are input into the image redrawing model. The model processes each difference point separately, generating multiple locally modified images. The model employs local redrawing technology, regenerating only the specified regions within the difference point planning information while keeping other parts of the image unchanged. For example, for difference point planning information containing two differences, the model generates three locally modified images: the first image simply changes the blue ceramic cup to a red one, keeping everything else the same as the original image; the second image only removes the book object, keeping everything else the same. By generating locally modified images separately for each difference point, the modification quality of each point can be independently evaluated, and qualified differences points can be flexibly selected for subsequent compositing based on the quality evaluation results.

[0090] The preset quality conditions refer to the standards used to judge whether a locally modified image is acceptable. These preset quality conditions can include requirements in multiple dimensions, such as requirements for the significance of differences, the reasonableness of differences, and image quality requirements. Candidate modified images are locally modified images that have passed quality screening and meet the preset quality conditions; the differences in these images will be used in the synthesis of the final target image.

[0091] In practice, the quality of each generated local modification image is evaluated, and those that meet preset quality conditions are selected as candidate modification images. Optionally, a multi-dimensional quality evaluation strategy can be adopted to give each local modification image a comprehensive score. For example, the difference between the local modification image and the original image can be evaluated to ensure that the player can identify the differences; the logical rationality of the modified content in the local modification image can be evaluated, such as the absence of physically unreasonable occlusion relationships or inconsistent lighting effects; the overall visual quality of the local modification image can also be evaluated, such as whether the boundaries of the modified area blend naturally, whether the texture is clear, and whether there are obvious generation defects. Only local modification images that meet the preset quality conditions in all evaluation dimensions will be selected as candidate modification images. Local modification images that do not meet the preset quality conditions will be marked as unqualified and temporarily excluded.

[0092] After identifying candidate images for modification, the differences contained in each candidate image can be fused into the original image. Difference fusion refers to the operation of merging the difference points from multiple candidate images into a single image. Since each candidate image contains only one difference point, image fusion technology is needed to integrate the difference points from multiple candidate images into the original image, generating a final target image containing multiple difference points.

[0093] For example, image compositing techniques can be used to fuse the differences contained in each candidate modified image into the original image material. This can be achieved by extracting the regions in each candidate modified image that have changed relative to the original image material, and then sequentially overlaying these changed regions onto the original image material according to the location information of the difference points. After fusion, a target image containing multiple difference points can be obtained, which, together with the original image material, constitutes a complete "spot the difference" image.

[0094] The technical solution of this application embodiment generates the difference points as independent local modified images, and then performs quality screening on each local modified image to ensure that each difference point in the target image meets the quality requirements. This improves the quality of the spot-the-difference images, solves the problem of inconsistent quality caused by the difficulty and rationality of objectively defining each difference point in manually generated spot-the-difference images, and enhances the gaming experience of spot-the-difference games.

[0095] In another embodiment, selecting locally modified images that meet preset quality conditions from each locally modified image as candidate modified images includes: obtaining quality quantization information of each locally modified image; and selecting locally modified images whose quality quantization information meets preset quality conditions from each locally modified image as candidate modified images.

[0096] Quality quantification information refers to the numerical or graded results obtained after quantitatively evaluating the quality of a locally modified image. Quality quantification information includes at least one of the following: difference significance assessment results, difference reasonableness assessment results, and image quality assessment results.

[0097] The significance of difference assessment refers to the evaluation metric used to measure whether the difference between the locally modified image and the original image is sufficiently obvious and easily identifiable. Significance of difference can be quantified by calculating parameters such as the degree of pixel difference, color contrast, and texture variation in the difference regions between the two images. Optionally, a pre-trained deep learning model can be used to measure the degree of difference between the locally modified image and the original image to obtain the significance of difference assessment result.

[0098] The difference rationality assessment result refers to the evaluation index used to measure whether the modified content in a locally modified image conforms to the scene logic and whether it is natural and consistent. Optionally, the image material, the locally modified image, and the image modification prompts can be input into a multimodal large model such as a visual language model. The multimodal large model performs semantic understanding and rationality judgment on the modified scene to obtain the difference rationality assessment result.

[0099] The image quality assessment result refers to the evaluation metrics used to measure whether the visual quality of the locally modified image meets the standards, including evaluation metrics such as image sharpness, color fidelity, naturalness of boundary blending, and the presence of generation defects. Optionally, a pre-trained deep learning model can be used to evaluate the consistency of appearance and style in the locally modified image to generate the image quality assessment result.

[0100] In practice, one or more dimensions can be selected from the difference significance assessment results, difference rationality assessment results, and image quality assessment results as quality quantification information.

[0101] Optionally, if each dimension (difference significance assessment result, difference rationality assessment result, and image quality assessment result) included in the quality quantification information meets the corresponding threshold requirements, then the quality quantification information can be determined to meet the preset quality conditions; or, if the difference significance assessment result indicates that the modification is significantly different, the difference rationality assessment result indicates that the modified difference is reasonable, and the image quality assessment result indicates that the difference is qualified, then the quality quantification information can be determined to meet the preset quality conditions.

[0102] The technical solution of this application embodiment achieves multi-dimensional quantitative evaluation of locally modified images by acquiring quality quantification information including at least one of difference significance evaluation results, difference rationality evaluation results, and image quality evaluation results. By presetting quality conditions, high-quality candidate modified images can be automatically screened, avoiding the subjectivity and inconsistency of manual evaluation, and ensuring that the differences in the final target image reach a high level in terms of significance, rationality, and visual quality, thereby improving the quality and user experience of the spot-the-difference game.

[0103] In another embodiment, the differences contained in each candidate modified image are fused into the image material to obtain a target image, including: determining the number of candidate modified images; if the number of images is greater than or equal to a preset difference requirement, selecting a target modified image from each candidate modified image that matches the preset difference requirement; and fusion of the differences contained in each target modified image into the image material to obtain the target image.

[0104] The number of images refers to the total number of locally modified images that have passed quality screening and been identified as candidate modified images. Since not all generated locally modified images can meet the preset quality conditions, the number of candidate modified images may be less than the total number of difference points planned in the difference point planning information.

[0105] The preset difference requirement refers to the number of difference points that the final target image needs to contain. This number is usually determined in advance based on the game difficulty setting and product requirements. For example, for a beginner-level spot-the-difference game, the preset difference requirement can be set to 3, meaning that the final target image needs to contain 3 difference points; for an advanced-level spot-the-difference game, the preset difference requirement can be set to 5 or more.

[0106] The target modified image refers to the locally modified image that is finally selected from the candidate modified images and will be used to synthesize the target image.

[0107] If the number of candidate images for modification is greater than or equal to the preset number of differences required, it means that there are enough candidate images for modification, and the target image for modification that matches the preset number of differences required can be selected from the candidate images.

[0108] Specifically, the target modified image that matches the preset number of differences required is selected from each candidate modified image. This can be done randomly or by sorting the candidate modified images according to their scores based on quality quantification information, and then selecting the preset number of candidate modified images in descending order of the sorting.

[0109] The process of fusing the differences contained in each target modified image into the image material to obtain the target image can refer to the difference fusion step in the aforementioned embodiment. The terminal extracts the regions in each target modified image that have changed relative to the image material, and superimposes these changed regions onto the image material according to the location information of the difference points.

[0110] The technical solution of this application improves the quality of the final target image and enhances the visual effects and user experience of the spot-the-difference game by determining the number of candidate modified images and selecting the optimal target modified image for fusion when the number is sufficient. Furthermore, by flexibly setting the preset number of difference requirements, target images containing different numbers of difference points can be generated according to different game difficulties and product needs, enhancing the applicability and scalability of the solution.

[0111] In another embodiment, after determining the number of candidate modified images, the method further includes: if the number of images is less than the preset difference requirement and the number of images is greater than or equal to the preset fault tolerance threshold, optimizing the difference points corresponding to the local modified images that do not meet the preset quality conditions to obtain updated difference point planning information, and returning to the step of inputting the difference point planning information into the prompt word generation model to obtain image modification prompt words for the image material; if the number of images is less than the preset fault tolerance threshold, discarding the image material, using the next image material as the updated image material, and returning to the step of identifying objects in the image material to obtain description information for each object.

[0112] The preset fault tolerance threshold refers to the minimum number of acceptable differences required to determine whether an image is worth retrying for optimization. The preset fault tolerance threshold is less than the preset difference requirement but greater than 0. For example, if the preset difference requirement is 5, the preset fault tolerance threshold can be set to 3. This means that if the number of candidate images for modification is between 3 and 4, the image is considered salvageable and worth retrying for unacceptable differences; if the number of candidate images for modification is less than 3, the success rate of the image is considered too low, and it is not worthwhile to continue investing resources in optimization retrying.

[0113] In practice, after determining the number of candidate images to be modified, the terminal checks whether this number is less than the preset difference requirement. If the number of candidate images to be modified is insufficient to meet the preset difference requirement, it further checks whether the number of images is greater than or equal to a preset fault tolerance threshold. If the number of candidate images to be modified is between the preset fault tolerance threshold and the preset difference requirement, it indicates that some difference points have been successfully generated in the image material, and it still has optimization value. The terminal then initiates a difference point optimization retry process.

[0114] The terminal identifies the difference points in the locally modified image that do not meet the preset quality conditions and optimizes these difference points. Optimization strategies may include: adjusting the preset difference type of the difference points, such as replacing the original "deleted object" type with "changed attribute"; adjusting the object to be modified, such as selecting another object in the scene as the new object to be modified; and adjusting the description style of the image modification prompts, such as adding more detailed visual feature descriptions or changing the description style. Through these optimizations, updated difference point planning information is generated, and the process returns to the step of generating image modification prompts based on the difference point planning information. The locally modified image is then regenerated for the optimized difference points and a quality assessment is performed.

[0115] If the number of candidate modified images is less than the preset fault tolerance threshold, it indicates that the success rate of generating difference points for the image material is too low, and continuing to optimize and retry is not very beneficial. The terminal determines that the image material is low-quality material, chooses to discard the image material, and no longer invests additional computing resources in it. The terminal extracts the next image material from the queue of images to be processed, uses it as the updated image material, returns to the step of recognizing objects in the image material, and restarts the entire image generation process.

[0116] For the convenience of those skilled in the art, Figure 3 An example of a logic diagram for an image generation method is provided.

[0117] The technical solution of this application embodiment, by introducing a preset fault tolerance threshold and an intelligent fault tolerance retry mechanism, improves the utilization efficiency of computing resources and the overall production efficiency in the process of generating spot-the-difference images while ensuring the quality of the target image.

[0118] In another embodiment, after obtaining the target image corresponding to the image material, the method further includes: identifying rejection images from the target images whose quality review result is unsuccessful; counting the first occurrence frequency of each preset difference type in each rejection image; identifying preset difference types whose first occurrence frequency is higher than a first frequency threshold as risk difference types; reducing the probability of assigning risk difference types to the next image material; and / or counting the second occurrence frequency of each preset difference type in each target image; identifying preset difference types whose second occurrence frequency is higher than a second frequency threshold as high-frequency difference types; expanding the modification method of high-frequency difference type indication; and / or identifying rejection images from the target images whose quality review result is unsuccessful; replacing the preset difference type corresponding to the rejection image with the target difference type, wherein the occurrence frequency of the target difference type in historical rejection images is lower than that of the preset difference type.

[0119] Quality auditing refers to the process of manually sampling and inspecting the generated target images, and having auditors determine whether the target images meet the usability standards.

[0120] Among them, a rejected image refers to a target image that is determined by the auditors to be unacceptable or not meeting the usage requirements during the quality audit.

[0121] Among them, the first occurrence frequency refers to the number of times a certain preset difference type appears in the rejection image; the first frequency threshold refers to the threshold value for the number of occurrences used to determine whether a certain preset difference type is a high-risk type; the risk difference type refers to the preset difference type that appears frequently in the rejection image and is likely to cause generation failure or quality problems.

[0122] In practice, after generating a batch of target images, the terminal submits these images to the review system for quality auditing. Auditors inspect the sampled target images, determining whether each image meets quality standards, and rejecting any images that fail, providing the reasons for rejection. The terminal then filters out all rejected images from the audit results and performs data analysis on these rejected images. The terminal extracts the difference information corresponding to each rejected image, counting the frequency of each preset difference type in each rejected image to obtain the first occurrence frequency. For example, in 100 rejected images, the "delete object" type appears 45 times, the "change attribute type - change color" type appears 30 times, and the "add object" type appears 25 times. The terminal sets a first frequency threshold, for example, 40 times, and identifies preset difference types with a first occurrence frequency higher than 40 times as risky difference types. In this example, the first occurrence frequency of the "delete object" type is 45 times, exceeding the first frequency threshold of 40 times, and is therefore identified as a risky difference type.

[0123] When assigning preset difference types to new image materials, the terminal reduces the probability of assigning risky difference types. For example, the terminal can reduce the assignment probability of the "delete object" type from 20% to 10%, reducing the number of difference points generated for that type and thus lowering the risk of the target image being rejected due to such difference points. By dynamically adjusting the assignment probability of preset difference types, the terminal can adaptively optimize the difference point generation strategy and improve the overall pass rate of the target image.

[0124] The second frequency of occurrence refers to the total number of times a certain preset difference type appears in all generated target images; the second frequency threshold is the threshold value for determining whether a certain preset difference type is a high-frequency use type; the high-frequency difference type refers to the preset difference type that is frequently used in the target image generation process.

[0125] Among them, the expansion and modification method refers to adding more diverse specific implementation methods and prompt word templates to a certain preset difference type, so as to enrich the expression of the difference points of that type.

[0126] In the specific implementation, all target images generated within a certain period of time are counted, and the total number of times each preset difference type appears in these target images is calculated to obtain the second occurrence frequency; the preset difference types whose second occurrence frequency is higher than the second frequency threshold are determined as high-frequency difference types.

[0127] Because high-frequency difference types are widely used, if their modification methods are monotonous and their presentation fixed, the generated differences will easily lack originality, reducing the game's fun and appeal. The terminal expands the modification methods for high-frequency difference types. For example, for changing the attribute type—changing color—the terminal can add more diverse color modification methods such as "gradient color replacement" and "pattern color replacement" to the existing "solid color replacement" modification method. It can also expand the prompt word template library for this type, preparing richer color description words for different objects and scenes. For example, for the cup object, it can add diverse prompt word templates such as "coral pink ceramic cup with gradient effect" and "deep sea blue glass cup with matte effect." By expanding the modification methods, while maintaining the use of high-frequency difference types, it is possible to generate more diverse and novel difference point presentations, enhancing the richness of the target image and the game experience.

[0128] The target difference type refers to the preset difference type that has historically performed better and is used to replace the original preset difference type in the rejection image. Historical rejection images refer to the collection of target images that were generated in the past and whose quality review results were not satisfactory.

[0129] In practice, images that fail quality review are filtered from the target images. For each rejected image, its preset difference types are analyzed, and the frequency of these preset difference types in historical rejected images is queried. The terminal selects a target difference type from the difference type library that has a lower frequency and higher success rate in historical rejected images to replace the original preset difference type in the rejected image. Then, new difference point planning information can be generated, and the corresponding local modified image and target image can be regenerated. By replacing with a target difference type with a higher success rate, the success probability of retry generation can be improved, and resource waste can be reduced.

[0130] The technical solution of this application embodiment, by establishing a data optimization process based on quality review feedback, can identify risk difference types from rejected images and reduce their usage frequency, discover innovative expansion needs from high-frequency difference types and enrich their expression forms, and improve the retry success rate through intelligent replacement strategies. This continuous learning and adaptive optimization mechanism enables the image generation process to continuously improve the difference point generation strategy, thereby enhancing the quality stability and content diversity of the target image.

[0131] In practical applications, the terminal can also employ a phased batch processing strategy for multiple stages of the entire image generation process. The terminal divides the image generation process into several independent stages, such as image material generation, prompt word generation, and local modification image generation. Within each stage, the terminal batches all pending tasks, loading the corresponding model only at the beginning of the stage and unloading it at the end. For example, in the image material generation stage, the terminal loads the text-based image model all at once, generates all image materials in batches, and then unloads the model; in the prompt word generation stage, the terminal loads the prompt word generation model all at once, generates all the image modification prompt words for the differences in the images in batches, and then unloads the model; in the local modification image generation stage, the terminal loads the image redrawing model all at once, generates the local modification images corresponding to each difference in batches, and then unloads the model. This phased batch processing strategy reduces the number of model scheduling operations, improves resource utilization and image generation throughput, and enables efficient large-scale production of spot-the-difference images, meeting the needs of large-scale production and rapid iteration of game content.

[0132] In another embodiment, such as Figure 4 As shown, an image generation method is provided, which can be applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps:

[0133] Step S402: Identify the objects in the image material and obtain description information for each object.

[0134] Step S404: Obtain at least one preset difference type corresponding to the image material, and determine the object to be modified corresponding to each preset difference type from each object.

[0135] The preset difference type is used to indicate the modification method for the object to be modified in the image material.

[0136] Step S406: Generate difference point planning information for image materials based on the description information corresponding to the object to be modified and the preset difference type.

[0137] The difference point planning information includes at least one difference point; each difference point includes an object to be modified, a preset difference type corresponding to the object to be modified, and the location information of the object to be modified in the image material.

[0138] Step S408: Input the difference point planning information into the prompt word generation model to obtain the image modification prompt words for the image material.

[0139] Step S410: Input the image modification prompts and image materials into the image redrawing model to obtain the local modified images corresponding to each difference point.

[0140] Each locally modified image has one difference from the original image.

[0141] Step S412: Obtain the quality quantization information of each local modified image, and select the local modified images whose quality quantization information meets the preset quality conditions from each local modified image as candidate modified images.

[0142] In one embodiment, the step of selecting locally modified images that meet preset quality conditions from each of the locally modified images as candidate modified images includes: obtaining quality quantification information of each of the locally modified images; the quality quantification information includes at least one of difference significance assessment results, difference reasonableness assessment results, and image quality assessment results; and selecting locally modified images whose quality quantification information meets preset quality conditions from each of the locally modified images as candidate modified images.

[0143] Step S414: The differences contained in each candidate modified image are fused into the image material to obtain the target image.

[0144] There is at least one difference between the target image and the image source.

[0145] In one embodiment, the step of fusing the differences contained in each of the candidate modified images into the image material to obtain the target image includes: determining the number of candidate modified images; if the number of images is greater than or equal to a preset difference requirement, selecting a target modified image from each of the candidate modified images that matches the preset difference requirement; and fusing the differences contained in each target modified image into the image material to obtain the target image.

[0146] In one embodiment, after determining the number of candidate modified images, the method further includes: if the number of images is less than the preset difference requirement and the number of images is greater than or equal to a preset fault tolerance threshold, optimizing the difference points corresponding to the local modified images that do not meet the preset quality conditions to obtain updated difference point planning information, and returning to the step of inputting the difference point planning information into the prompt word generation model to obtain the image modification prompt words for the image material; if the number of images is less than the preset fault tolerance threshold, discarding the image material, using the next image material as the updated image material, and returning to the step of identifying objects in the image material to obtain description information of each object.

[0147] Step S416: Establish the association between the image material and the target image.

[0148] The association relationship is used to create an image difference annotation task; the image difference annotation task is used by users to find the differences between the target image and the image material.

[0149] Step S418: Identify the rejection images from the target images whose quality review results are not passed, count the first occurrence frequency of each preset difference type in each rejection image, identify the preset difference types whose first occurrence frequency is higher than the first frequency threshold as risk difference types, and reduce the probability of assigning risk difference types to the next image material.

[0150] Step S420: Count the second occurrence frequency of each preset difference type in each target image, determine the preset difference type whose second occurrence frequency is higher than the second frequency threshold as a high-frequency difference type, and expand the modification method of the high-frequency difference type indication.

[0151] Step S422: Identify the rejection images from the target images whose quality audit results are unsuccessful, and replace the preset difference type corresponding to the rejection images with the target difference type.

[0152] It should be noted that the specific limitations of the above steps can be found in the specific limitations of an image generation method described above.

[0153] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0154] Based on the same inventive concept, this application also provides an image generation apparatus for implementing the image generation method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more image generation apparatus embodiments provided below can be found in the limitations of the image generation method described above, and will not be repeated here.

[0155] In one exemplary embodiment, such as Figure 5 As shown, an image generation apparatus is provided, comprising:

[0156] The recognition module 510 is used to recognize objects in the image material and obtain description information of each object;

[0157] The determining module 520 is used to obtain at least one preset difference type corresponding to the image material, and determine the object to be modified corresponding to each preset difference type from each of the objects; the preset difference type is used to indicate the modification method of the object to be modified in the image material;

[0158] The generation module 530 is used to generate image modification prompts for the image material based on the description information corresponding to the object to be modified and the preset difference type.

[0159] The redrawing module 540 is used to input the image modification prompt and the image material into the image redrawing model to obtain the target image corresponding to the image material; there is at least one difference between the target image and the image material.

[0160] The module 550 is used to establish the association between the image material and the target image; the association is used to create an image difference annotation task; the image difference annotation task is used to allow the user to find the differences between the target image and the image material.

[0161] In one embodiment, the generation module 530 is specifically used to generate difference point planning information for the image material based on the description information corresponding to the object to be modified and the preset difference type; the difference point planning information includes at least one difference point; each difference point includes an object to be modified, a preset difference type corresponding to the object to be modified, and the position information of the object to be modified in the image material; the difference point planning information is input into the prompt word generation model to obtain the image modification prompt word for the image material.

[0162] In one embodiment, the redrawing module 540 is specifically used to input the image modification prompt and the image material into the image redrawing model to obtain local modified images corresponding to each difference point; there is a difference between each local modified image and the image material; local modified images that meet preset quality conditions are selected from each local modified image as candidate modified images; the differences contained in each candidate modified image are fused into the image material to obtain the target image.

[0163] In one embodiment, the redrawing module 540 is specifically used to obtain quality quantification information of each of the local modified images; the quality quantification information includes at least one of difference significance assessment result, difference reasonableness assessment result and image quality assessment result; and selects local modified images from each of the local modified images whose quality quantification information meets preset quality conditions as candidate modified images.

[0164] In one embodiment, the redrawing module 540 is specifically used to determine the number of candidate modified images; if the number of images is greater than or equal to a preset difference requirement, select a target modified image from each of the candidate modified images that matches the preset difference requirement; and fuse the differences contained in each target modified image into the image material to obtain the target image.

[0165] In one embodiment, the redrawing module 540 is specifically used to optimize the difference points corresponding to locally modified images that do not meet the preset quality conditions when the number of images is less than the preset difference requirement and the number of images is greater than or equal to the preset fault tolerance threshold, to obtain updated difference point planning information, and return to the step of inputting the difference point planning information into the prompt word generation model to obtain the image modification prompt words for the image material; when the number of images is less than the preset fault tolerance threshold, discard the image material, use the next image material as the updated image material, and return to the step of identifying objects in the image material to obtain description information of each object.

[0166] In one embodiment, the image generation device further includes a modification module, specifically configured to: determine, from the target images, a rejection image whose quality review result is unsuccessful; count the first occurrence frequency of each preset difference type in each rejection image; determine the preset difference type whose first occurrence frequency is higher than a first frequency threshold as a risk difference type, thereby reducing the probability of assigning the risk difference type to the next image material; and / or, count the second occurrence frequency of each preset difference type in each target image; determine the preset difference type whose second occurrence frequency is higher than a second frequency threshold as a high-frequency difference type, thereby expanding the modification method indicated by the high-frequency difference type; and / or, determine, from the target images, a rejection image whose quality review result is unsuccessful; replace the preset difference type corresponding to the rejection image with a target difference type, wherein the occurrence frequency of the target difference type in historical rejection images is lower than that of the preset difference type.

[0167] Each module in the aforementioned image generation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0168] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements an image generation method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0169] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0170] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0171] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0172] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0173] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0174] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0175] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0176] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An image generation method, characterized in that, The method includes: Objects in the image material are identified to obtain descriptive information for each object; Obtain at least one preset difference type corresponding to the image material, and determine the object to be modified corresponding to each preset difference type from each of the objects; the preset difference type is used to indicate the modification method of the object to be modified in the image material; Based on the description information corresponding to the object to be modified and the preset difference type, generate image modification prompts for the image material; The image modification prompt and the image material are input into the image redrawing model to obtain the target image corresponding to the image material; there is at least one difference between the target image and the image material. Establish a relationship between the image material and the target image; the relationship is used to create an image difference annotation task; the image difference annotation task is used by the user to find the differences between the target image and the image material.

2. The method according to claim 1, characterized in that, The step of generating image modification prompts for the image material based on the description information corresponding to the object to be modified and the preset difference type includes: Based on the description information corresponding to the object to be modified and the preset difference type, difference point planning information of the image material is generated; the difference point planning information includes at least one difference point; each difference point includes an object to be modified, a preset difference type corresponding to the object to be modified, and the position information of the object to be modified in the image material. The difference point planning information is input into the prompt word generation model to obtain the image modification prompt words for the image material.

3. The method according to claim 2, characterized in that, The step of inputting the image modification prompt and the image material into the image redrawing model to obtain the target image corresponding to the image material includes: The image modification prompts and the image material are input into the image redrawing model to obtain local modified images corresponding to each difference point; there is one difference between each local modified image and the image material. From each of the locally modified images, locally modified images that meet the preset quality conditions are selected as candidate modified images; The differences contained in each of the candidate modified images are fused into the image material to obtain the target image.

4. The method according to claim 3, characterized in that, The step of selecting locally modified images that meet preset quality conditions from each of the locally modified images as candidate modified images includes: Obtain quality quantification information for each of the locally modified images; the quality quantification information includes at least one of the following: difference significance assessment result, difference reasonableness assessment result, and image quality assessment result; From each of the locally modified images, locally modified images whose quality quantification information meets the preset quality conditions are selected as candidate modified images.

5. The method according to claim 3, characterized in that, The step of fusing the differences contained in each of the candidate modified images into the image material to obtain the target image includes: Determine the number of candidate images for modification; If the number of images is greater than or equal to the preset number of differences required, a target modified image that matches the preset number of differences required is selected from each of the candidate modified images; The differences contained in each of the target modified images are fused into the image material to obtain the target image.

6. The method according to claim 5, characterized in that, After determining the number of candidate modified images, the method further includes: If the number of images is less than the preset difference requirement and the number of images is greater than or equal to the preset fault tolerance threshold, the difference points corresponding to the locally modified images that do not meet the preset quality conditions are optimized to obtain the updated difference point planning information. Then, the process returns to the step of inputting the difference point planning information into the prompt word generation model to obtain the image modification prompt words for the image material. If the number of images is less than the preset fault tolerance threshold, the image material is discarded, the next image material is used as the updated image material, and the process returns to the step of identifying objects in the image material and obtaining description information of each object.

7. The method according to claim 1, characterized in that, After obtaining the target image corresponding to the image material, the method further includes: From the target images, identify the rejection images whose quality review result is unsatisfactory. Count the first occurrence frequency of each preset difference type in each rejection image. Identify preset difference types whose first occurrence frequency is higher than a first frequency threshold as risk difference types, reducing the probability of assigning these risk difference types to the next image material; and / or, The second occurrence frequency of each preset difference type in each target image is statistically analyzed, and the preset difference types whose second occurrence frequency is higher than the second frequency threshold are determined as high-frequency difference types, thus expanding the modification method of the high-frequency difference type indication; and / or, From the target images, identify the rejection images whose quality audit result is not passed, and replace the preset difference type corresponding to the rejection image with the target difference type, wherein the frequency of the target difference type in historical rejection images is lower than that of the preset difference type.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images