Picture album generation method

Abstract

The present application discloses a picture album generation method. The method comprises: acquiring a scene picture, and extracting story description information from the scene picture; generating target story text corresponding to the story description information, and determining story plots corresponding to the target story text; and on the basis of a prompt word in each story plot, generating a character story picture corresponding to each story plot, and constructing a picture album on the basis of the character story picture corresponding to each story plot. In the present application, story text is generated by acquiring story description information from a scene picture, and a character story picture corresponding to each story plot is generated on the basis of a prompt word of each story plot in the story text, so that a picture album is constructed on the basis of the character story picture corresponding to each story plot, that is, the picture album can be generated in a one-click manner by means of the scene picture, thereby reducing the generation time of the picture album, and improving the generation efficiency of the picture album.

Description

Album generation method

[0001] Cross-reference to related applications

[0002] This disclosure claims priority to Chinese Patent Application No. 202411979533.9, filed with the Chinese Patent Office on December 30, 2024, entitled “Method for Generating Brochures”, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application belongs to the field of data processing technology, and in particular relates to a method for generating picture albums. Background Technology

[0004] A picture book is a special form of visual art that combines real photographs of people with comic strips, telling a story through a series of continuous images of people and narrative text, which has a strong sense of realism and vividness.

[0005] In an exemplary technique, when creating a picture book, a story text is first written. Referring to the scenario content of the story text, and using traditional photography techniques, real-life images with the same pose and position are taken. The real-life figures and background images are then merged by computer to create a series of live-action sequential images with a storyline. The picture book is then synthesized based on these live-action sequential images.

[0006] However, the writing of the story text takes a long time, and the time required to photograph the real people and background images using traditional photography techniques is also long, resulting in low production efficiency for the picture book. Summary of the Invention

[0007] This application provides a method for generating a picture album to solve the problem of low production efficiency of picture albums.

[0008] In a first aspect, embodiments of this application provide a method for generating a picture album, including:

[0009] Acquire scene images and extract story description information from the scene images;

[0010] Generate the target story text corresponding to the story description information, and determine the story plot corresponding to the target story text;

[0011] Based on the clues in each storyline, generate character story images corresponding to each storyline, and construct a picture album based on the character story images corresponding to each storyline.

[0012] Secondly, embodiments of this application provide a picture album generation apparatus, including:

[0013] The acquisition module is configured to acquire scene images and extract story description information from the scene images.

[0014] The first generation module is configured to generate the target story text corresponding to the story description information and determine the story plot corresponding to the target story text;

[0015] The second generation module is configured to generate character story images corresponding to each storyline based on the prompt words in each storyline, and to construct a picture album based on the character story images corresponding to each storyline.

[0016] Thirdly, embodiments of this application provide a catalog generation device, which includes: a processor and a memory storing computer program instructions;

[0017] When the processor executes the computer program instructions, it implements a picture album generation method as described in one aspect.

[0018] Fourthly, embodiments of this application provide a computer storage medium on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the brochure generation method as described in one aspect is implemented.

[0019] Fifthly, embodiments of this application provide a computer program product, wherein instructions in the computer program product, when executed by a processor of an electronic device, cause the electronic device to perform a picture album generation method as described in one aspect.

[0020] The album generation method provided in this application generates story text by obtaining story description information from scene images, and generates character story images for each story plot based on the prompt words of each story plot in the story text. Thus, an album is constructed based on the character story images corresponding to each story plot. In other words, an album can be generated with one click using scene images, reducing the album generation time and improving the album generation efficiency. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 is a schematic diagram of an application scenario that can realize the album generation method provided in the embodiments of this application;

[0023] Figure 2 is a flowchart illustrating the first embodiment of the brochure generation method of this application;

[0024] Figure 3 is a flowchart illustrating the second embodiment of the brochure generation method of this application;

[0025] Figure 4 is a flowchart illustrating the third embodiment of the brochure generation method of this application;

[0026] Figure 5 is a schematic diagram of the process for obtaining the target story text in this application;

[0027] Figure 6 is a flowchart illustrating the fourth embodiment of the brochure generation method of this application;

[0028] Figure 7 is a schematic diagram of the process for obtaining compositional features in this application;

[0029] Figure 8 is a schematic diagram of the mapping adapter in this application;

[0030] Figure 9 is a schematic diagram of the multimodal joint coding process in this application;

[0031] Figure 10 is a schematic diagram of the workflow of subspace transformation from the attention layer in this application;

[0032] Figure 11 is a flowchart illustrating the fifth embodiment of the brochure generation method of this application;

[0033] Figure 12 is a flowchart illustrating the sixth embodiment of the brochure generation method of this application;

[0034] Figure 13 is a schematic diagram of the process for obtaining the training model of this application;

[0035] Figure 14 is a flowchart illustrating the seventh embodiment of the brochure generation method of this application;

[0036] Figure 15 is a simplified flowchart of the album generation process in this application;

[0037] Figure 16 is a schematic diagram of the functional modules of the brochure generation device involved in this application;

[0038] Figure 17 is a schematic diagram of the brochure generation device provided in an embodiment of this application. Detailed Implementation

[0039] A picture book is a special form of visual art that combines real photographs of people with comic strips, telling a story through a series of continuous images of people and narrative text, which has a strong sense of realism and vividness.

[0040] In an exemplary technique, when creating a picture book, a story text is first written. Referring to the scenario content of the story text, and using traditional photography techniques, real-life images with the same pose and position are taken. The real-life figures and background images are then merged by computer to create a series of live-action sequential images with a storyline. The picture book is then synthesized based on these live-action sequential images.

[0041] The inventors of this application discovered that, however, the writing of the story text takes a long time, and the time required to photograph real people and background images using traditional photography techniques is also long, resulting in low production efficiency for the picture book.

[0042] The inventors of this application therefore conceived of generating story text by obtaining story description information from scene images, and generating character story images for each story plot based on the prompts for each story plot in the story text. Thus, a picture album can be constructed based on the character story images corresponding to each story plot. In other words, a picture album can be generated with one click using scene images, reducing the generation time of the picture album and improving the generation efficiency of the picture album.

[0043] This application provides a method for generating brochures. The following examples illustrate application scenarios where the brochure generation method provided in this application can be applied.

[0044] As shown in Figure 1, the user uploads scene images to the album generation device 200 via mobile terminal 100. The album generation device 200 extracts story description information from the scene images and generates story text corresponding to the story description information. Based on the prompts for each plot point in the story text, the album generation device 200 generates character story images corresponding to each plot point. The album generation device 200 assembles the various character story images to obtain the album and outputs it.

[0045] The following describes the brochure generation method provided in the embodiments of this application with reference to Figure 1.

[0046] Figure 2 shows a flowchart of a first embodiment of the brochure generation method provided in this application. As shown in Figure 2, the brochure generation method includes:

[0047] Step S201: Obtain scene images and extract story description information from the scene images.

[0048] In this embodiment, the executing entity is a brochure generation device. For ease of description, the term "device" will be used to refer to the brochure generation device below. The device can be a server or any terminal device with brochure generation functionality.

[0049] Users can select an image from the cloud drive or local photo album and upload it to the device as a scene image to generate a photo album.

[0050] After acquiring a scene image, the device extracts story description information from it. For example, the device may have an image understanding model pre-configured. This model extracts key information from the image to obtain the story description. Key information includes, for example, the scene and the task. The image understanding model incorporates a mixture of encoder-decoder framework, integrating understanding and generation capabilities. This mixture of encoder-decoder framework combines multiple encoder and decoder models, each handling a specific task or data type. The image understanding model is trained using captioning and filtering techniques. This training method improves the model's data cleaning and data augmentation capabilities, enabling it to effectively extract key information from scene images as story description information.

[0051] Step S202: Generate the target story text corresponding to the story description information, and determine the story plot corresponding to the target story text.

[0052] After obtaining the story description information, the device generates the target story text corresponding to the story description information. For example, the device can utilize a large language model to generate the target story text corresponding to the story description information. The device inputs the story description information into the large language model, and the large language model outputs the story text as the target story text.

[0053] After obtaining the target story text, the device determines the individual story segments within the target story text. For example, the device divides the target story text into multiple story segments, with each segment being a unit. A minimum segment can be, for example, a paragraph in the target story text, or the text content between two adjacent periods in the target story text.

[0054] Step S203: Generate character story images corresponding to each storyline based on the clue words in each storyline, and construct a picture album based on the character story images corresponding to each storyline.

[0055] The device generates corresponding character story images for each storyline. For example, the device searches for clues within the storyline; these clues are divided into global clues and local clues. Global clues include descriptive terms for scenes within the storyline, such as "sunny" or "gloomy environment." Local clues can include characters, objects, facial expressions, positions, actions, shapes, distances between objects and characters, and placement of objects within the storyline.

[0056] The device extracts prompts from the storyline and generates corresponding character story images based on these prompts. For example, if the storyline describes character A having a meal on a ship, and the prompts are "character A," "ship," and "having a meal," then a character story image will be generated that reads: "A ship is sailing on the ocean. Character A is sitting at a table on the deck of the ship. There is food on the table, and character A's action is picking up the food."

[0057] The above method generates character story images corresponding to each storyline, and the device constructs a picture album based on these images. Specifically, the picture album is obtained by assembling the character story images according to their positions within the target story text.

[0058] In this embodiment, story text is generated by acquiring story description information from scene images, and character story images for each story plot are generated based on the prompts for each plot in the story text. Thus, a picture album is constructed based on the character story images corresponding to each plot. In other words, a picture album can be generated with one click using scene images, without the need for manual writing of story text or manual taking of images of the scenes in the story text. This simplifies the picture album generation process, reduces the generation time, and improves the efficiency of picture album generation.

[0059] Referring to Figure 3, which shows a flowchart of the second embodiment of the brochure generation method of this application, based on the first embodiment, step S202 includes:

[0060] Step S301: Obtain the story generation instruction and parse the story generation instruction to obtain the target parameters. The target parameters include at least one of the story type and text style type.

[0061] In this embodiment, the device can generate story text based on the user's personalized needs. For example, the user inputs an auxiliary command into the device. The auxiliary command contains the user's personalized parameters. The device parses the auxiliary command to obtain the personalized parameters as target parameters. The target parameters include at least one of story type and text style type. Story types include, but are not limited to, fairy tales, inspirational stories, and touching stories; text style types include, but are not limited to, tragedies, comedies, lyrical stories, and lighthearted stories.

[0062] Step S302: Generate target story text that matches the target parameters based on the story description information.

[0063] After obtaining the target parameters, the device generates the target story text based on the target parameters and story description information. This means the target story text contains a story plot with story description information, and the target story text is either the text style type desired by the user, or the target story is the story type desired by the user. Specifically, the story description information and target parameters are input into the large language model to obtain the target story text output by the large language model.

[0064] In this embodiment, the device generates target story text information that matches the user's preferences based on the user's personalized needs for story text, thereby improving the user experience.

[0065] Referring to Figure 4, which shows a flowchart of the third embodiment of the brochure generation method provided in this application, based on the embodiment shown in Figure 2 or Figure 3, step S202 includes:

[0066] Step S401: Generate multiple first story texts based on the story description information.

[0067] In this embodiment, the device inputs story description information into a large language model, and the large language model outputs multiple story texts, each of which is defined as a first story text.

[0068] In addition, when users have personalized needs, the target parameters and story description information are input into the large language model to obtain multiple first story texts output by the large language model.

[0069] Step S402: Determine the score for each first story text.

[0070] Because large language models cannot understand linguistic ambiguity and cannot perform sentiment analysis or judgment on the generated story texts, they generate multiple first-story texts. An automatic evaluation metric is then used to score each of these first-story texts, thereby selecting the most suitable one. This automatic evaluation metric requires no reference answer and determines whether a story text aligns with people's preferences based on dimensions such as repetition, coherence, and logic of the plot. The automatic evaluation metric is a score; the higher the score, the closer the story text is to people's preferences.

[0071] To this end, the device determines a score for each first story text. For example, a model can be used to determine the score of the first story text; that is, the first story text is input into the model to obtain its score. The model is trained using training story texts labeled with scores. The score of the first story text: S 故事 (I) = Sigmoid(Wv + b), where W and b are model parameters, and S... 故事 (I) represents the scoring value.

[0072] Step S403: Determine a first target score among the various scoring scores, and determine the first story text corresponding to the first target score as the target story text. The first target score is used to indicate a scoring score that is greater than a first preset score.

[0073] After determining the score for each first story text, the device identifies a first target score among these scores. The first target score is the score greater than a first preset score; that is, the score greater than the first preset score is determined as the first target score. If there are multiple scores greater than the first preset score, the largest score is determined as the first target score.

[0074] After determining the first target score, the device can use the first story text corresponding to the first target score as the target story text.

[0075] In this embodiment, the device acquires multiple first story texts for each storyline and determines the rating score for each first story text, thereby selecting a story text that resonates with people's emotions based on the rating score.

[0076] In one embodiment, after the device generates a second story text corresponding to the story description information, the device displays the second story text. Users can edit the plot, characters, and other content of the second story text according to their personal preferences; editing includes, but is not limited to, adding, deleting, and modifying. Upon detecting an editing operation on the story text, the device modifies the second story text based on the editing operation to obtain the target story text.

[0077] It should be noted that the second story text can be the first story text corresponding to the aforementioned first target score, and the first story text can be generated based on the user's story generation instructions and story description parameters. Specifically, referring to Figure 5, the generation method of the target story text is as follows: 1. Select scene images from the cloud; 2. The image understanding model extracts story description information from the scene images; 3. The large language model generates multiple first story texts based on the story description information and story generation instructions; 4. Evaluate the multiple first story texts to obtain a score, and obtain the best story text based on the score. The best story text is the first story text corresponding to a score greater than the first preset score; 5. Edit the best story text to obtain the target story text.

[0078] In this embodiment, by having the user edit the story text, a target story text that matches the user's preferences can be obtained, thus improving the user experience.

[0079] Referring to Figure 6, which shows a flowchart of the fourth embodiment of the brochure generation method provided in this application, step S203 includes:

[0080] Step S601: Obtain the clue words for the storyline and generate the story illustrations corresponding to the storyline.

[0081] In this embodiment, after the target story text is input into the large language model, the device outputs K story plots and a global cue word p corresponding to each story plot. i And story illustrations corresponding to each storyline. i .

[0082] Each cue word [p1, p2, ..., p] k ] = G[F T2P [T, K], where G is the large language model, F T2P The auxiliary instructions for generating the storyline describe the requirements of the differential storyline and are input by the user; T is the target story text and K is the number of prompt words.

[0083] Step S602: Determine the layout parameters of the target object in the story illustration. The target object includes characters and / or items in the story illustration. The layout parameters are used to indicate the bounding rectangle of the target object.

[0084] To improve the quality of story illustrations and reduce the complexity of generating character story images, a large language model can be used to add corresponding layout parameters to each story illustration. These layout parameters are essentially the bounding rectangles of the target objects within the illustration. The target objects refer to the characters and / or items in the story illustration, and the bounding rectangle of a target object is the box containing the target object. The device inputs the target story text into the story illustrations generated by the large language model, which contain the bounding rectangles of the target objects. These layout parameters act as control signals to guide the generation of character story images.

[0085] Assume σ i Let F be the layout parameter for the i-th story segment. P2L F is an auxiliary instruction for generating layout parameters. P2L The requirements for the layout parameters for generating story illustrations are described, with the layout parameters for each story segment being: [σ1, σ2, ..., σ...]. k ] = G[F P2L [p1, p2, ..., p] k ]).

[0086] Set global prompt word p i Based on the number of targets k described in the sentence i Disassemble into k i These are local prompts representing different target objects. That is, k iThis represents the number of local cue words in the i-th story illustration and the number of bounding rectangles of the target object, with each local cue word corresponding to one target object. and Let represent the bounding rectangles of the local cue words and the target object, respectively. Therefore, the layout parameters of the target object in the story illustration take the following form:

[0087] Step S603: Extract compositional features from the story illustrations based on layout parameters.

[0088] Dense compositional features such as the outline key points of the target object represent the overall compositional distribution of the image. Compositional features, as control conditions, will help generate high-quality image content.

[0089] To address this, the device extracts the compositional features of the target object's contour key points based on layout parameters. For example, suppose the j-th target object in the i-th story segment is I. ij The local cue words for the target object are First, local cue words are input into the diffusion model to generate an image of a single target object. Then, a general detection model is used to obtain the bounding box parameters of the target object in the image. Then As input to the segmentation algorithm, the segmentation mask m is obtained. ij Finally, a pixel difference network is used to extract the contour key points E of the target object. ij Therefore, the set of key outline points of the target object in the i-th story segment can be obtained as E. i E i This refers to compositional features. Referring to Figure 7, the steps for obtaining compositional features are as follows:

[0090] 1. Input local prompt words into a diffusion model to generate a target, which is the image of the target object; 2. Obtain the bounding box (outer rectangle) of the target object image; 3. Segment the target, that is, segment the image; 4. Extract contour points from the segmented image.

[0091] Step S604: Based on the compositional features, prompts, story illustrations, and diffusion model corresponding to the storyline, the preset noise map corresponding to the storyline is denoised to obtain the character story image corresponding to the storyline.

[0092] In this embodiment, a diffusion model is incorporated into the device. The diffusion model is a probabilistic generative model that excels in image synthesis. It performs a two-stage diffusion calculation in the latent space. The first stage, the forward diffusion process, uses a T-step Markov chain to progressively add Gaussian noise to the input data x0 to obtain the confused data x. T Let q(x0) be the original distribution of the data, and its prior distribution is:

[0093] Where t represents the number of steps, N represents a Gaussian distribution, I is the variance of the Gaussian noise, and β t This is a hyperparameter.

[0094] The second stage of backdiffusion utilizes the noise estimation model ε. θ A reverse denoising process is fitted to reconstruct the samples. Assume ε is the value of sample x. t Real noise, ε θ (x t Let ε be the predicted noise, and C be the introduced category or image, etc., used to guide the generation of the target image. Then the noise estimation model ε θ The mean squared error loss function is:

[0095] The device acquires a preset noise map corresponding to each storyline. The preset noise map can be randomly generated. The device inputs the composition features, prompts, story illustrations, and preset noise map of the storyline into the diffusion model. The diffusion model can then output character story images corresponding to the storyline. In other words, the composition features, prompts, and story illustrations of the storyline guide the diffusion model to denoise the preset noise map, thereby obtaining character story images that match the storyline.

[0096] In this embodiment, the device acquires the clue words of the story plot, generates story illustrations of the story plot, and determines the layout parameters of the characters and / or objects in the story illustrations. Based on the layout parameters, it extracts composition features, and then uses the story plot composition features, clue words, and story illustrations to guide the diffusion model to denoise the preset noise map, thereby obtaining a character story picture that matches the story plot.

[0097] In one embodiment, the diffusion model includes a composition adapter for processing compositional features corresponding to the storyline. Specifically, compositional features such as the outline keypoints of the target object are input as parameters to the composition adapter. The composition adapter aligns the internal knowledge and external information of the diffusion model and provides more refined structural control, generating high-quality story illustrations. As shown in Figure 8, the composition adapter comprises multiple units, each consisting of a deformable convolutional layer and two residual blocks. The introduced deformable convolutional layer can flexibly extract target features of arbitrary shapes, better representing the region of interest in the image, and extracting a total of N scales of compositional features F. E The input is concatenated with the features of N sub-units of the noise estimation network, and the mapping features at N scales are extracted as the outputs of each unit of the mapping adapter.

[0098] In this embodiment, a composition adapter is set in the diffusion model. The composition adapter uses the obtained outline key points of the target object as additional conditions to guide and control the diffusion model to generate stable and detailed images.

[0099] In one embodiment, the diffusion model includes an encoder that processes the cue words and story illustrations to unify their text and image content. Specifically, the encoder can be a multimodal co-encoder, which encodes text and images into a unified multimodal latent control, effectively eliminating ambiguity between modalities and ensuring consistency between text and images.

[0100] Referring to Figure 9, which is a flowchart of the multimodal co-encoder, the text prompts include global prompts and local prompts. The bounding rectangle of the target object and the prompts are denoted as y, and the single target image set is denoted as x. s A single-target image refers to an image of the target object extracted from the story illustrations. The position of the target image in the cue word is denoted as O. Let y represent Rose on the boat, x... s This represents the character Rose and the boat, with positions O = 0, 2.

[0101] The multimodal co-encoder will input parameters ( y x s O) is encoded into the joint latent space to obtain the joint encoded sequence h. u As shown in Figure 9, firstly, for image x... s Features are extracted and input into the multilayer perceptron (MLP) to be trained to obtain pseudo-word features. Then, the word features at the corresponding text encoding position O are replaced to obtain the recombined encoding sequence W. r The recombined coding sequence W r After being encoded by CLIP (Contrastive Language-Image Pre-training, a multimodal pre-trained neural network), the reconstructed hidden layer sequence h is obtained. r .

[0102] Finally, the recombined hidden layer sequence h r The sequence at position O in the middle layer is replaced with the original hidden layer sequence h. y The value of is used to obtain the joint encoded sequence h. u =ITUE(y, x) s O).

[0103] Furthermore, the diffusion model can set a mapping adapter and an encoder. Therefore, the diffusion model needs to train the parameters of the multimodal joint encoder and the mapping adapter. Thus, the optimization objective function of the diffusion model is:

[0104] In this embodiment, the encoder in the diffusion model can encode text and images into a unified multimodal latent control, effectively eliminating ambiguity between modalities and ensuring the consistency of text and images.

[0105] In one example, the diffusion model includes a noise estimation network that denoises a pre-defined noise map based on compositional features, cue words, and story illustrations.

[0106] Visual self-attention networks offer greater scalability and flexibility compared to traditional denoising networks, and possess superior modeling advantages. To reduce the computational complexity of the diffusion model and extract representative global information, a customized design was implemented using a subspace transformation self-attention network. This design replaced the original attention mechanism with a subspace transformation self-attention layer. The entire network consists of N sub-units used to fit the noise distribution; in other words, the noise estimation network is a subspace transformation self-attention noise estimation network.

[0107] Referring to Figure 10, since the computational burden of the feature space dimension is several times that of the channel dimension, this embodiment employs transformative self-attention to enhance feature learning, thereby reducing computational burden and backsampling time. Three overlapping convolutional layers are used to perform subspace projection (SP) on the attention model parameters Q, K, and V, respectively, and then the self-attention is represented on the feature subspace. Assuming the feature F has a length of A, a dimension of B, a subspace transformation parameter of r, and a learnable scale parameter of α, the specific calculation of the subspace transformation self-attention ST is as follows:

[0108] To compare the computational cost of different self-attention calculation methods, we set r = 0.5, feature length dimension A = 64 × 64, and feature dimension B = 32, and statistically analyzed the total computational cost of multiplication and addition for different self-attention structures. Table 1 shows that the subspace transformation self-attention method has the lowest computational cost, proving that our method effectively reduces computational overhead.

[0109] Table 1. Computational complexity of different self-attention calculation methods

[0110] In this embodiment, the diffusion model can reduce its computational complexity by setting a subspace transformation self-attention noise estimation network.

[0111] Referring to Figure 11, which shows a flowchart of the fifth embodiment of the brochure generation method provided in this application, based on the embodiment shown in Figure 6, before step S604, the method further includes:

[0112] Step S1101: Obtain multiple training samples and train the preset model based on each training sample to obtain an intermediate model.

[0113] In this embodiment, the device needs to train the diffusion model. Diffusion models typically use CLIP to extract image features, but CLIP is trained using poorly aligned data, resulting in a wide range of extracted features but poor expressiveness for facial features. To extract high-fidelity facial information, original images of people from a face database are used as target images, and a face model is used as a feature encoder to encode the target images to obtain training samples for training the diffusion model.

[0114] After obtaining each training sample, the device trains the preset model based on each training sample to obtain an intermediate model.

[0115] Step S1102: Obtain face training images and train the intermediate model based on the face training images to obtain a diffusion model. The face training images include the face images of the target person specified by the user.

[0116] To further improve facial similarity and refine facial details, a fine-tuning technique is employed during the training phase. This involves fusing training images into the intermediate model, ensuring that the images generated by the diffusion model possess the facial contour features of the user-specified target person. The training images include facial images of the target person specified by the user.

[0117] To address this, the device acquires face training images and trains an intermediate model based on these images to obtain the diffusion model.

[0118] In this embodiment, an intermediate model is obtained by training a preset model using training samples, and then a diffusion model is obtained by training the intermediate model using face training images. Thus, the image generated by the diffusion model has the facial contour features of the person specified by the user.

[0119] Referring to Figure 12, which shows a flowchart of the sixth embodiment of the brochure generation method provided in this application, based on the embodiment shown in Figure 11, step S1102 includes:

[0120] Step S1201: Obtain the first image uploaded by the user. The first image includes the face of the target person specified by the user from a first angle.

[0121] In this embodiment, the user stores a large amount of image data in the cloud. The user can select the avatar of the person for whom a digital avatar is to be created from the cloud image data. The image containing this avatar is defined as the first image, which is the first angle of the target person's face specified by the user. The first angle can be a frontal view, that is, the first image is a clear frontal image of the target person's face. The user uploads the first image to the device, that is, the device obtains the first image uploaded by the user.

[0122] Step S1202: Display multiple second images, each second image including a second angle of the target person's face, the second angle being different from the first angle.

[0123] After acquiring the first image, the device performs inspection on the first image to determine whether it meets the quality requirements. For example, the device analyzes the face, face angle, and face attributes using a face detection model and a face attribute model. If the analysis results indicate that the first image meets the quality requirements, multiple second images are displayed. Each second image shows the target person's face from a second angle, and the second angle is different from the first angle.

[0124] For example, after obtaining the first image, the device retrieves all images containing the face of the person in the first image from the database, and crops the face region of the retrieved images to obtain an image of the person containing the face region, which is then displayed as the second image. The user can select images from each of the second images as secondary images to supplement the outline of the target person from other angles.

[0125] Step S1203: In response to the detection of a selected operation, a face training image is constructed based on the second image selected by the selected operation and the first image.

[0126] After detecting a selection operation, the device constructs a face training image based on the second image selected by the selection operation and the first image. Multiple second images can be selected, and the more second images selected, the better the effect on the generated digital persona.

[0127] In addition, the selected second image also needs to be detected by a face detection model and a face attribute model to obtain a second image that meets the requirements of the instruction.

[0128] The first image is used as the main image, and the selected second image is used as the secondary image. Based on the image rotation model of orientation judgment and the face detection and key point model, the main image and the secondary image are processed by the face fine rotation method. Then, the human body analysis model and the portrait beautification model are used to obtain high-quality face images. Then, the face attribute model and the text annotation model are combined with the label post-processing method to generate fine labels for the face images, thus obtaining the face training image. The face training image is the task digital image clone.

[0129] Referring to Figure 13, which is a schematic diagram of the process for acquiring face training images, the specific steps are as follows: 1. The user uploads the main image and performs face, face angle, and face attribute detection on the main image to determine whether the main image meets the quality requirements; 2. When the main image meets the quality requirements, the user uploads a secondary image and performs face recognition, face queuing, and image quality analysis on the secondary image to determine whether the secondary image meets the quality requirements; 3. If the secondary image meets the quality requirements, the secondary image and the main image are processed based on human body analysis, portrait beautification, image quality analysis, facial key points, text annotation, and image rotation to obtain a digital image clone of the person.

[0130] In this embodiment, the device acquires a first image uploaded by the user and displays multiple second images based on the first image, thereby constructing a face training image corresponding to the person specified by the user based on the first image and the selected second images.

[0131] Referring to Figure 14, which shows a flowchart of the seventh embodiment of the brochure generation method provided in this application, based on the embodiment shown in Figure 6 or Figure 11, step S604 includes:

[0132] Step S1401: Based on the compositional features, prompts, story illustrations, and diffusion model corresponding to the storyline, the preset noise map corresponding to the storyline is denoised to obtain multiple intermediate images corresponding to the storyline.

[0133] In this embodiment, in order to select high-quality illustrations, each storyline has multiple story illustrations, that is, each storyline generates a story illustration candidate set, which contains multiple illustrations for that storyline, and the device selects high-quality illustrations from the story illustration candidate set.

[0134] For example, the device inputs the compositional features, prompts, story illustrations, and preset noise maps corresponding to the storyline into the diffusion model. The diffusion model outputs multiple intermediate images corresponding to the storyline, which constitute a pre-selection set of story illustrations.

[0135] Step S1402: Extract the first image features and text features of each intermediate image, determine the first similarity between the first image features and text features of the intermediate image, and determine the second similarity between the first image features of the intermediate image and the second image features in the template face image.

[0136] After obtaining the intermediate images, their quality needs to be evaluated. Specifically, the first image feature F of each intermediate image is extracted. i and text features F t The device calculates the cosine similarity between the first image features and the text features of the intermediate image as the first similarity.

[0137] Suppose the cue word p in the i-th story segment i There are M candidate images x m Then the first similarity is: S 图，文 (p i ,x m )=cos(F t (p i ),F i (x m ))

[0138] After determining the first similarity, the device acquires the second image feature F associated with the template face image. n Calculate the first image feature Fi and the second image feature F for each intermediate image. n The second similarity score is calculated between them. The cue word p in the i-th stage. i The m-th candidate image x m The second similarity is: S 人脸 (p i ,x m )=cos(F m ,F n )

[0139] It should be noted that the template face image is obtained from the training samples. That is, the face quality assessment model is used to evaluate the face images in each training sample and select the face with the best quality as the template face image.

[0140] Step S1403: Determine the quality score of the intermediate image based on the first similarity, second similarity, first image features, and text features corresponding to the intermediate image.

[0141] After determining the second similarity, the device determines the quality score corresponding to the intermediate image based on the first similarity, the second similarity, the first image features, and the text features corresponding to the intermediate image.

[0142] For example, suppose the cue word p in the i-th story segment... i There are M candidate images x m Enc t and Enc i Let S represent the text features and image features extracted by the encoder of the human preference classifier, respectively. Then, the text feature of the middle image is represented by the value corresponding to the first image feature: S 图 (p i ,x m ) = Enc t (p i )·Enc t (x m )×100; and the quality score is: S 插图 =a1S图 +a2S 图,文 +a3S 人脸 a1, a2, and a3 are the corresponding weights, which can be set manually.

[0143] Step S1404: Determine the second target score among the various quality scores, and determine the character story picture corresponding to the story plot based on the intermediate picture corresponding to the second target score. The second target score is a quality score that is greater than the second preset score.

[0144] After obtaining the quality score of each intermediate image, the device determines a second target score among the various quality scores. The second target score is a quality score that is greater than a second preset score. If there are multiple quality scores that are greater than the second preset score, the largest quality score is taken as the second target score.

[0145] After determining the second target score, the device determines the character story images corresponding to the storyline based on the intermediate images corresponding to the second target score.

[0146] In one example, the device directly uses the intermediate image corresponding to the second target score as the character story image corresponding to the storyline.

[0147] In another example, the face in the intermediate image corresponding to the second target score is fused with the face in the template face image to obtain the character story image corresponding to the storyline.

[0148] Referring to Figure 15, which is a schematic diagram of the workflow of the diffusion model trained in this embodiment, specifically, the target contour key point set is input into the mapping adapter; global cue words, local cue words, and single-target image sets (the set of images of the target objects obtained in the story illustrations) are input into the multimodal joint encoder to unify the text and image in the multimodal space; the subspace transformation from the attention noise estimation network, after being trained with training samples and face training images, is input into the noise map; the subspace transformation from the attention noise estimation network, through the input of the mapping adapter and the output of the multimodal latent space, denoises the noise map to obtain a pre-selection set of story illustrations. The intermediate image corresponding to the second target score is selected from the pre-selection set of story illustrations, and this intermediate image is fused with the template face image to obtain a candidate set of story illustrations, which includes character story images corresponding to each story segment.

[0149] In this embodiment, the intermediate images in the pre-selected set of story illustrations for each story segment are automatically scored and sorted across multiple dimensions, thereby selecting images with higher quality scores as character story images for the storyline.

[0150] In one embodiment, to enhance the appeal of the picture book, speech synthesis can be performed on the flower city to generate a video from the picture book. For example, after constructing the picture book based on character and story images corresponding to each storyline, the device generates speech corresponding to the target story text, thereby generating a story video based on the rain and character / story images in the picture book.

[0151] In this embodiment, it supports the synthesis of short videos of digital character stories, provides a variety of speech synthesis styles and sound effects, and supports the adjustment of speech speed and reading style to adapt to the preferences and personalized needs of different users.

[0152] In one embodiment, the album can be shared, for example, by sharing the album or story videos to social media platforms such as mobile cloud storage, third-party software, etc., to enhance interaction and communication between users and friends, relatives, or other members in their social circles, and to analyze their stories and feelings in a timely manner. Specifically, after the album is constructed based on the character story images corresponding to each storyline, the album is displayed. When a sharing operation for the album is detected, the target terminal corresponding to the sharing operation is determined, and the album is sent to the target terminal.

[0153] Furthermore, to provide users with better personalized services and optimize model performance, a user feedback option was added to the story book's display page. User ratings were collected from multiple dimensions, including character similarity, text-image consistency, story satisfaction, and illustration satisfaction. A Likert scale of 1-5 was used for rating, with higher scores indicating greater satisfaction, as shown in the table below. These user ratings can then be converted into training data, and reinforcement learning based on human preferences can be used to optimize the model, making the generated stories more aligned with user tastes.

[0154] The device acquires comment information about the catalog, identifies the elements in the catalog that need to be modified based on the comment information, and modifies the elements in the catalog based on the comment information.

[0155] For example, a user rating feedback option is set on the display page of the album. User ratings are collected from multiple dimensions such as character similarity, consistency between text and images, story satisfaction, and illustration satisfaction. The rating uses a 1-5 level rating, with higher scores indicating greater satisfaction, as shown in Table 2 below.

[0156] Table 2 User Evaluation Form

[0157] The user reviews can then be converted into training data, and reinforcement learning based on human preferences can be used to optimize the model, making the generated stories more in line with user preferences. Elements to be modified include, for example, digital characters, story content, and story illustrations. If the digital character does not resemble the real person, the element to be modified is the character; if the story content is inconsistent with the illustrations, the character story image is the element to be modified; if the story's coherence, logic, or plot is of low satisfaction, the story text is the element to be modified; if the illustrations in the booklet are of low consistency or quality, the character story image is the element to be modified.

[0158] In this embodiment, user cloud drive data is combined with a generation model. Users only need to upload a picture and specify the people in the album. Combined with technologies such as speech synthesis, a digital character story album with rich plots and emotional warmth can be automatically generated, providing cloud drive users with intelligent and fun applications. It is characterized by convenient operation, low cost and strong innovation.

[0159] Furthermore, considering the need for storybooks to generate detailed compositions while ensuring strict consistency between the storyline and the visuals, a story illustration generator was designed, consisting of modules such as a multimodal composition conditional diffusion model, a subspace transformation self-attention noise estimation network, and facial feature preservation. This generator maps text and images to a unified space, generates detailed story illustrations, effectively reduces computational overhead, and achieves high-fidelity digital character effects.

[0160] The cloud-based digital character storybook generation method proposed in this embodiment can automatically generate high-quality story text and illustrations, minimize user interaction, evaluate the quality of the booklet from multiple dimensions, and provide functions such as personalized settings and social sharing.

[0161] As shown in Figure 16, this application embodiment also provides a picture album generation device 1600, which includes: an acquisition module 1610, used to acquire scene images and extract story description information from the scene images; a first generation module 1620, used to generate target story text corresponding to the story description information and determine the story plot corresponding to the target story text; and a second generation module 1630, used to generate character story images corresponding to each story plot based on the prompt words in each story plot, and construct a picture album based on the character story images corresponding to each story plot.

[0162] In one embodiment, the picture book generation device 1600 is specifically configured to: acquire a story generation instruction and parse the story generation instruction to obtain target parameters, wherein the target parameters include at least one of story type and text style type; and generate target story text that matches the target parameters based on story description information.

[0163] In one embodiment, the picture book generation device 1600 is specifically configured to: generate multiple first story texts based on story description information; determine the score of each first story text; determine a first target score among the various score values, and determine the first story text corresponding to the first target score as the target story text, wherein the first target score is used to indicate a score value greater than a first preset score.

[0164] In one embodiment, the picture book generation device 1600 is specifically configured to: generate second story text corresponding to story description information and display the second story text; respond to an editing operation on the second story text, modify the second story text according to the editing operation, and obtain the target story text.

[0165] In one embodiment, the picture book generation device 1600 is specifically configured to: acquire clue words for the story plot and generate story illustrations corresponding to the story plot; determine the layout parameters of target objects in the story illustrations, the target objects including characters and / or items in the story illustrations, and the layout parameters are used to indicate the bounding rectangle of the target objects; extract compositional features in the story illustrations based on the layout parameters; and denoise the preset noise map corresponding to the story plot based on the compositional features, clue words, story illustrations, and diffusion model to obtain character story pictures corresponding to the story plot.

[0166] In one embodiment, the diffusion model includes a composition adapter for processing compositional features corresponding to storylines.

[0167] In one embodiment, the diffusion model is equipped with an encoder that processes the cue words and story illustrations to unify the text and image content of the cue words and story illustrations.

[0168] In one embodiment, the diffusion model includes a subspace transformation self-attention noise estimation network, which denoises a preset noise map based on composition features, cue words, and story illustrations.

[0169] In one embodiment, the album generation device 1600 is specifically configured to: acquire multiple training samples, and train a preset model based on each training sample to obtain an intermediate model; acquire face training images, and train the intermediate model based on the face training images to obtain a diffusion model, wherein the face training images include the face images of the target person specified by the user.

[0170] In one embodiment, the album generation device 1600 is specifically configured to: acquire a first image uploaded by a user, the first image including a first angle of the face of a target person specified by the user; display multiple second images, each second image including a second angle of the face of the target person, the second angle being different from the first angle; and in response to a detected selection operation, construct a face training image based on the second image selected by the selection operation and the first image.

[0171] In one embodiment, the picture book generation device 1600 is specifically configured as follows: denoising a preset noise image corresponding to the storyline based on the composition features, prompts, story illustrations, and diffusion model corresponding to the storyline to obtain multiple intermediate images corresponding to the storyline; extracting a first image feature and text feature of each intermediate image, determining a first similarity between the first image feature and the text feature of the intermediate image, and determining a second similarity between the first image feature of the intermediate image and the second image feature in the template face image; determining a quality score corresponding to the intermediate image based on the first similarity, the second similarity, the first image feature, and the text feature; determining a second target score among the various quality scores, and determining a character story image corresponding to the storyline based on the intermediate image corresponding to the second target score, wherein the second target score is a quality score greater than the second preset score.

[0172] In one embodiment, the picture book generation device 1600 is specifically configured to: fuse the face in the intermediate image corresponding to the second target score with the face in the template face image to obtain a character story picture corresponding to the story plot.

[0173] In one embodiment, the picture book generation device 1600 is specifically configured to: generate audio corresponding to the target story text; and generate a story video based on the audio and the character story pictures in the picture book.

[0174] In one embodiment, the album generation device 1600 is specifically configured to: display an album; in response to a sharing operation on the album, determine the target terminal corresponding to the analysis operation; and send the album to the target terminal.

[0175] In one embodiment, the album generation device 1600 is specifically configured to: acquire comment information on the album; determine the elements to be modified in the album based on the comment information; and modify the elements to be modified in the album based on the comment information.

[0176] It should be noted that the album generation device is the device corresponding to the album generation method described above. In embodiments where all implementation methods in the above method embodiments are applicable, the same technical effect can also be achieved.

[0177] Figure 17 shows a schematic diagram of the hardware structure of the album generation device provided in an embodiment of this application.

[0178] The brochure generation apparatus may include a processor 1701 and a memory 1702 storing computer program instructions.

[0179] Specifically, the processor 1701 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0180] Memory 1702 may include mass storage for data or instructions. For example, and not limitingly, memory 1702 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 1702 may include removable or non-removable (or fixed) media. Where appropriate, memory 1702 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 1702 is non-volatile solid-state memory.

[0181] In certain embodiments, memory 1702 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Thus, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the methods according to one aspect of this disclosure.

[0182] The processor 1701 reads and executes computer program instructions stored in the memory 1702 to implement any of the album generation methods in the above embodiments.

[0183] In one example, the brochure generation device may also include a communication interface 1703 and a bus 1710. As shown in Figure 17, the processor 1701, memory 1702, and communication interface 1703 are connected via the bus 1710 and communicate with each other.

[0184] The communication interface 1703 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0185] Bus 1710 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 1710 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0186] Furthermore, in conjunction with the brochure generation methods described in the above embodiments, this application embodiment can provide a computer storage medium for implementation. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the brochure generation methods described in the above embodiments.

[0187] This application also provides a computer program product, including a computer program, which, when executed, implements any of the brochure generation methods described in the above embodiments.

[0188] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0189] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0190] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0191] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable terminal device brochure generation apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable terminal device brochure generation apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0192] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for generating a catalog, the method comprising: Acquire scene images and extract story description information from the scene images; Generate the target story text corresponding to the story description information, and determine the story plot corresponding to the target story text; Based on the clues in each storyline, generate character story images corresponding to each storyline, and construct a picture album based on the character story images corresponding to each storyline.

2. The method according to claim 1, wherein, The generation of the target story text corresponding to the story description information includes: Obtain the story generation instruction and parse the story generation instruction to obtain the target parameters, wherein the target parameters include at least one of story type and text style type; Based on the story description information, generate target story text that matches the target parameters.

3. The method according to claim 1, wherein, The generation of the target story text corresponding to the story description information includes: Based on the story description information, generate multiple first story texts; Determine the score for each of the first story texts; A first target score is determined among the various rating scores, and the first story text corresponding to the first target score is determined as the target story text. The first target score is used to indicate the rating score that is greater than a first preset score.

4. The method according to claim 1, wherein, The generation of the target story text corresponding to the story description information includes: Generate the second story text corresponding to the story description information, and display the second story text; In response to an editing operation on the second story text, the second story text is modified according to the editing operation to obtain the target story text.

5. The method according to claim 1, wherein, The step of generating character story images corresponding to each storyline based on the clue words in each storyline includes: Obtain the clue words for the storyline and generate the corresponding story illustrations for the storyline; Determine the layout parameters of a target object in the story illustration, the target object including characters and / or items in the story illustration, the layout parameters being used to indicate the bounding rectangle of the target object; Extract the compositional features from the story illustrations based on the layout parameters; Based on the compositional features corresponding to the storyline, the prompt words, the story illustrations, and the diffusion model, the preset noise map corresponding to the storyline is denoised to obtain the character story images corresponding to the storyline.

6. The method according to claim 5, wherein, The diffusion model is equipped with a composition adapter, which is used to process the composition features corresponding to the storyline.

7. The method according to claim 5, wherein, The diffusion model is equipped with an encoder, which processes the prompt words and the story illustrations to unify the text and image content of the prompt words and the story illustrations.

8. The method according to claim 5, wherein, The diffusion model is equipped with a subspace transformation self-attention noise estimation network, which denoises the preset noise map based on the composition features, the cue words, and the story illustrations.

9. The method according to claim 5, wherein, Before denoising the preset noise map corresponding to the storyline based on the compositional features, the cue words, the story illustrations, and the diffusion model, the method further includes: Multiple training samples are obtained, and an intermediate model is obtained by training the preset model based on each training sample. Acquire face training images, and train the intermediate model based on the face training images to obtain a diffusion model. The face training images include face images of target persons specified by the user.

10. The album generation method according to claim 9, wherein, The acquisition of face training images includes: Obtain the first image uploaded by the user, the first image including the face of the target person specified by the user from a first angle; Display multiple second images, each of which includes a second angle of the target person's face, which is different from the first angle; Upon detecting a selected operation, a face training image is constructed based on the second image selected by the selected operation and the first image.

11. The method according to claim 5, wherein, The step of denoising the preset noise map corresponding to the storyline based on the compositional features corresponding to the storyline, the cue words, the story illustrations, and the diffusion model includes: Based on the compositional features corresponding to the storyline, the prompt words, the story illustrations, and the diffusion model, the preset noise map corresponding to the storyline is denoised to obtain multiple intermediate images corresponding to the storyline. Extract the first image features and text features of each intermediate image, determine the first similarity between the first image features and text features of the intermediate image, and determine the second similarity between the first image features of the intermediate image and the second image features in the template face image; The quality score corresponding to the intermediate image is determined based on the first similarity, the second similarity, the first image feature, and the text feature corresponding to the intermediate image. A second target score is determined among the various quality scores, and the character story image corresponding to the storyline is determined based on the intermediate image corresponding to the second target score. The second target score is the quality score that is greater than the second preset score.

12. The method according to claim 11, wherein, The step of determining the character story image corresponding to the storyline based on the intermediate image corresponding to the second target score includes: The face in the intermediate image corresponding to the second target score is fused with the face in the template face image to obtain the character story image corresponding to the storyline.

13. The method according to claim 1, wherein, After constructing the album based on the character story images corresponding to each of the aforementioned storylines, it also includes: Generate the audio corresponding to the target story text; A story video is generated based on the audio and the character story pictures in the album.

14. The method according to claim 1, wherein, After constructing the album based on the character story images corresponding to each of the aforementioned storylines, it also includes: Display the brochure; In response to a sharing operation on the brochure, the target terminal corresponding to the analysis operation is determined; The brochure is sent to the target terminal.

15. The method according to claim 14, wherein, After displaying the brochure, the following is also included: Obtain comment information for the aforementioned catalog; Based on the comment information, identify the elements in the album that need to be modified; Modify the elements in the album that need to be modified based on the comment information.

Images