Input method skin resource generation method and system

Abstract

The present invention provides an input method skin resource generation method and system, the method comprising: obtaining a target texture image output by a texture image generation model, wherein the texture image generation model is trained based on a sample texture image marked with a texture type label; obtaining a preset input method key pattern image, and generating target input method skin resource data according to the preset input method key pattern image and the target texture image. The present invention improves the generation efficiency of input method skin resources.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to an input method skin resource generation method and system. Background Art

[0002] With the rapid development of the Internet and the increasing integration of artificial intelligence technology into daily life, the mode of content production is undergoing a profound transformation. Despite this, the skin decoration in the input method, as a traditional and important element, still firmly occupies a place in the user's input method experience.

[0003] The current design of input method skins is extremely dependent on manual operations, and this model has significant limitations. The design quality of input method skins is limited by the designer's personal ability, the length of the creation cycle, and the level of technology. At the same time, it is difficult to achieve greater breakthroughs in production efficiency.

[0004] Therefore, there is an urgent need for an input method skin resource generation method and system to solve the above problems. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides an input method skin resource generation method and system.

[0006] The present invention provides an input method skin resource generation method, comprising: Acquire a target texture image output by a texture image generation model, wherein the texture image generation model is trained based on sample texture images marked with texture type labels; A preset input method key style image is obtained, and target input method skin resource data is generated according to the preset input method key style image and the target texture image.

[0007] According to an input method skin resource generation method provided by the present invention, the texture image generation model is trained by the following steps: Determining the texture type corresponding to each of the sample images according to the material information and style information in the sample images; According to the texture type, marking each of the sample images with a corresponding texture type label to obtain a sample texture image marked with the texture type label; Constructing training sample sets corresponding to the texture type labels, wherein the texture type labels of the sample texture images in the same training sample set are the same; According to each of the training sample sets, the large language model corresponding to each of the training sample sets is trained to obtain the texture image generation model corresponding to each of the texture type labels.

[0008] According to a method for generating input method skin resources provided by the present invention, before determining the texture type corresponding to each sample image according to the material information and style information in the sample image, the method further includes: Based on preset image resolution information, preprocessing each of the sample images to obtain a preprocessed sample image, wherein the resolutions of the preprocessed sample images are the same; The step of determining the texture type corresponding to each sample image according to the material information and style information in the sample image includes: According to the material information and the style information in the preprocessed sample images, the texture type corresponding to each of the preprocessed sample images is determined.

[0009] According to a method for generating input method skin resources provided by the present invention, the step of obtaining a preset input method key style image includes: Based on a preset input method key pattern database, the preset input method key pattern image is obtained, wherein the preset input method key pattern database is constructed from preset input method key pattern images.

[0010] According to an input method skin resource generation method provided by the present invention, the step of obtaining a target texture image output by a texture image generation model includes: Get the input method skin resource prompt word data; According to the keyword information in the input method skin resource prompt word data, obtaining the target texture type corresponding to the input method skin resource data to be generated, wherein the keyword information includes texture description information and input method skin texture description information; Determining a target texture image generation model from a plurality of texture image generation models according to the texture type label corresponding to the target texture type; The target texture image output by the target texture image generation model is obtained.

[0011] According to a method for generating input method skin resources provided by the present invention, generating target input method skin resource data according to the preset input method key style image and the target texture image includes: Constructing target input method skin resource description information according to the texture description information and the input method skin texture description information; The target input method skin resource description information is used as the image generation guidance information of the trained deep learning model, and the preset input method key style image and the target texture image are input into the trained deep learning model to obtain the target input method skin resource data output by the trained deep learning model.

[0012] According to a method for generating input method skin resources provided by the present invention, after generating target input method skin resource data according to the preset input method key style image and the target texture image, the method further includes: Receiving a first input instruction, wherein the first input instruction includes an operation instruction to adjust the target input method skin resource data; In response to the first input instruction, the text content, icon content or punctuation content corresponding to the target input method skin resource data is adjusted to obtain adjusted target input method skin resource data.

[0013] The present invention also provides an input method skin resource generation system, comprising: A texture image generation module, used to obtain a target texture image output by a texture image generation model, wherein the texture image generation model is trained based on sample texture images marked with texture type labels; The input method skin resource generation module is used to obtain a preset input method key style image, and generate target input method skin resource data according to the preset input method key style image and the target texture image.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-described input method skin resource generation methods is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the input method skin resource generation method described in any one of the above methods is implemented.

[0016] The input method skin resource generation method and system provided by the present invention determine the target texture image generation model from multiple texture image generation models, and then construct the target input method skin resource data according to the preset input method key style image and the target texture image output by the target texture image generation model, thereby improving the generation efficiency of the input method skin resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1A schematic diagram of the flow of the input method skin resource generation method provided by the present invention; Figure 2 A schematic diagram of a texture image output by the texture image generation model provided by the present invention; Figure 3 A schematic diagram of a preset input method key pattern image provided by the present invention; Figure 4 A schematic diagram showing a preview of the input method skin resource data provided by the present invention; Figure 5 A schematic diagram of the training process of the texture image generation model provided by the present invention; Figure 6 The overall schematic diagram of the input method skin resource generation process provided by the present invention; Figure 7 A schematic diagram of the structure of the input method skin resource generation system provided by the present invention; Figure 8 This is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] In the current input method skin design workflow, the operation team usually initiates the design idea, the designer is responsible for realizing the idea, and finally the operation team configures and deploys it online. However, the existing input method skin design process is limited by the separation of job roles and professional fields. Specifically, although the operation team plays an active role in creative conception, its creativity is often limited to the framework of existing skin design, and it is difficult to propose more forward-looking and innovative ideas from the perspective of professional design. This limitation may cause the design results to be out of touch with the current popular trends in the design industry, and fail to keenly capture the latest design trends and creative inspirations.

[0021] In addition, the entire process from the initial conception of the idea to the completion of the skin design and successful application often takes as long as 1 to 2 weeks. In this process, the communication and coordination costs incurred are considerable and cannot be ignored. More importantly, when the skin design finally comes out after going through many production links, its design style may no longer be the mainstream trend of the market, and the popularity has also dissipated. This time delay directly weakens the market appeal and expected benefits of the online skin, resulting in the actual results failing to fully achieve the established goals.

[0022] In view of the current application status of the input method skin design process in the field of costume resource production, although it can meet the basic needs of daily sales, it still faces significant limitations in efficiency improvement, creative innovation and cost control. The present invention provides an input method skin resource generation method based on artificial intelligence (AI), which not only significantly reduces the frequency and depth of manual intervention, but also breaks the constraints of professional skills on innovation ability, achieves a leap-forward improvement in skin production efficiency (specifically, the increase is as high as 900%), and greatly reduces labor costs.

[0023] Figure 1 A schematic diagram of the flow of the input method skin resource generation method provided by the present invention, such as Figure 1 As shown, the present invention provides an input method skin resource generation method, comprising: Step 101 : obtaining a target texture image output by a texture image generation model, wherein the texture image generation model is trained based on sample texture images marked with texture type labels.

[0024] In the present invention, the texture image generation model is built based on the LoRA (Low-Rank Adaptation) model. The LoRA model adapts (or "fine-tunes") the pre-trained model by introducing a low-rank matrix, thereby optimizing a specific task or data set. In the present invention, a large number of sample texture images marked with texture type labels are collected. These images cover a variety of texture types from simple to complex, from metal to cloth, and each image has a corresponding texture type label to guide the model training process.

[0025] Next, using these labeled sample texture images, the LoRA model is trained to learn how to generate images with corresponding textures based on input instructions or conditions (for example, specified texture types). Due to the lightweight nature of the LoRA model, this training process is relatively efficient, reducing the consumption of computing resources and time while maintaining model performance.

[0026] Once the training is completed, the model that best meets the target texture characteristics is selected from multiple trained texture image generation models according to actual needs or task requirements as the target texture image generation model. Subsequently, by inputting specific instructions or conditions to the model (for example, requiring the generation of an image with a "metal texture"), the model will output a target texture image that meets the target texture characteristics.

[0027] Step 102, obtaining a preset input method key pattern image, and generating target input method skin resource data according to the preset input method key pattern image and the target texture image.

[0028] In the present invention, the preset input method key style image refers to the basic layout image of the input method, which represents the fixed style of the input method, such as the common 9-grid, full keyboard, numeric keyboard, etc. These images usually include the basic framework of the keyboard and the preliminary design of the keys, and may not include specific details such as color, texture or texture.

[0029] The target input method skin resource description information is a set of data or parameters that describe the specific requirements and characteristics of the target input method skin. It may include color schemes, texture types, changes in key shapes, and special effects (such as animations when keys are pressed), etc. This information is the guiding principle for generating the target input method skin resource data.

[0030] The target texture image refers to the image output by the texture image generation model after specific training, which contains a certain texture effect, such as metal, leather, glass, simple and retro textures. These texture images will be used to enhance the beauty and realism of the input method skin. Figure 2 This is a schematic diagram of a texture image output by the texture image generation model provided by the present invention. For details, please refer to Figure 2 As shown, it should be noted that the texture image output by the texture image generation model can directly output the texture image of the existing input method skin style (such as Figure 2 As shown), texture images without input method skin style can also be generated.

[0031] Furthermore, in the present invention, a preset key style image matching the required input method style is first retrieved from the preset input method key style database. For example, if a skin needs to be designed for the 9-grid input method, the corresponding 9-grid key style image is selected. Figure 3 This is a schematic diagram of the preset input method key pattern image provided by the present invention, for details, please refer to Figure 3 shown.

[0032] Next, the target input method skin resource description information is read and parsed to obtain the color, texture, texture and other detailed information required by the input method skin to be generated, which will guide the subsequent image processing and synthesis work. In the present invention, the target input method skin resource description information can exist in the form of text, configuration file or structured data.

[0033] The ControlNet model is a conditional generation model that can generate images under the guidance of given conditions (such as text descriptions, images or other forms of guidance information). In the present invention, the ControlNet model is used to generate target input method skin resource data based on the target input method skin resource description information, the preset input method key style image and the target texture image.

[0034] Specifically, the target input method skin resource description information is input as a condition into the ControlNet model, and the ControlNet model will parse this information and understand the required skin characteristics. Next, the preset input method key style image and the target texture image are input as images into the ControlNet model. These images provide the basis for the keyboard layout and texture effects. Based on the conditional input and image input, the ControlNet model will combine the texture and texture effects in the target texture image with the keyboard layout in the preset input method key style image, and adjust the details according to the guidance in the target input method skin resource description information. Finally, the ControlNet model will output the target input method skin resource data, usually one or more images, which includes the required keyboard layout, color scheme, texture and texture effects, etc. Figure 4 This is a preview diagram of the input method skin resource data provided by the present invention. The specific display image of the target input method skin resource data can be referred to as shown in FIG4.

[0035] Optionally, in the present invention, the generated target input method skin resource data may need to undergo some subsequent processing before being used in actual input method applications, including image format conversion, resolution adjustment, key coordinates and response area marking, etc.

[0036] The input method skin resource generation method provided by the present invention determines a target texture image generation model from multiple texture image generation models, and then constructs target input method skin resource data according to a preset input method key style image and a target texture image output by the target texture image generation model, thereby improving the generation efficiency of input method skin resources.

[0037] Figure 5 This is a schematic diagram of the training process of the texture image generation model provided by the present invention, which can be referenced Figure 5 As shown, The texture image generation model is trained by the following steps: Determining the texture type corresponding to each of the sample images according to the material information and style information in the sample images; According to the texture type, marking each of the sample images with a corresponding texture type label to obtain a sample texture image marked with the texture type label; Constructing training sample sets corresponding to the texture type labels, wherein the texture type labels of the sample texture images in the same training sample set are the same; According to each of the training sample sets, the large language model corresponding to each of the training sample sets is trained to obtain the texture image generation model corresponding to each of the texture type labels.

[0038] In the present invention, sample images of different materials (such as metal, wood, and glass) and different styles (such as simple, retro, and futuristic) are first collected and screened from a wide range of image resources. These images should have high definition and diversity to ensure the accuracy and generalization ability of subsequent training. Further, the collected sample images are analyzed, and the texture type corresponding to each sample image is determined based on the material information and style information in the image. This process can be assisted by the judgment of professionals or by using existing image classification algorithms.

[0039] After determining the texture type, each sample image is labeled, that is, a corresponding texture type label is assigned to it, which will be used in the subsequent training and evaluation process. Then the labeled sample images are classified and sorted according to the texture type to ensure that each texture type label contains a certain number of sample images.

[0040] Then, based on the classified and sorted sample images, training sample sets corresponding to each texture type label are constructed, and these training sample sets will be used for the subsequent training of the large language model. Specifically, based on these sample texture images marked with the same texture type label, the present invention constructs training sample sets corresponding to each texture type label. For example, sample texture images of metal texture will be constructed into one training sample set, and similarly, sample texture images of glass texture will also form another training sample set.

[0041] After constructing these training sample sets, the corresponding large language model (the large language model consists of a basic large model and a LoRA model) is trained according to each training sample set. For each training sample set, the present invention trains a special texture image generation model. These texture image generation models will learn the key features of their respective texture types during the training process, and can accurately reproduce these features in subsequent generation tasks. For example, the metal texture image generation model will learn the texture, gloss, and reflection characteristics of the metal texture, and accurately reflect them when generating images. Similarly, the glass texture image generation model will also learn the transparency, refraction, and reflection characteristics of the glass texture, and apply them in the generation process.

[0042] Before training the texture image generation model, it is necessary to select a pre-trained image generation model as a basic large model. In the present invention, several basic large models of general styles can be used, for example, AWPainting_v1.3safetensors, basic algorithm_XL.safetensors, etc. These models already have certain image generation capabilities and can be further adapted to specific texture types through training.

[0043] During the training process, the LoRA model adapts to specific tasks or data sets by adjusting a small number of parameters (usually low-rank matrices) in the base large model. When training with the LoRA model, only the parameters in the LoRA module are updated, while most of the parameters of the base large model (usually "frozen") remain unchanged. Once the training is completed, the LoRA module can also be applied to other similar base large models without retraining the entire model.

[0044] In the present invention, for each training sample set corresponding to the texture type label, the selected basic large model is used for training. During the training process, it is necessary to set appropriate hyperparameters (such as learning rate, batch size, number of training rounds, etc.) to ensure that the model can fully learn the texture features in the sample image.

[0045] After the training is completed, a batch of sample images are generated using the trained texture image generation model. These images should be able to reflect the learned texture features. The generated sample images are then evaluated for texture to check whether they meet the expected texture type. In the present invention, this can be done by manual inspection or using quantitative indicators (such as FID scores) to measure the distance between the generated image and the real image in the feature space.

[0046] Through the above training process, the present invention can determine the texture type according to the material information and style information in the sample image, and construct a training sample set to train the texture image generation model. Finally, the performance of the model is verified through the sample generation and evaluation process to ensure that the texture image generation model can generate texture images that meet expectations.

[0047] Based on the above embodiment, before determining the texture type corresponding to each sample image according to the material information and style information in the sample image, the method further includes: Based on preset image resolution information, preprocessing each of the sample images to obtain a preprocessed sample image, wherein the resolutions of the preprocessed sample images are the same; The step of determining the texture type corresponding to each sample image according to the material information and style information in the sample image includes: According to the material information and the style information in the preprocessed sample images, the texture type corresponding to each of the preprocessed sample images is determined.

[0048] In the present invention, firstly, according to the design specification of the input method skin, a preset image resolution is determined, and this resolution can be set based on a comprehensive consideration of factors such as task requirements, computing resources or model performance, etc. For example, all images are adjusted to 1024x1024 pixels, or other resolutions that are more suitable for task requirements.

[0049] Then, each sample image is processed using image processing techniques, such as scaling, cropping or interpolation, to ensure that the resolution of all images is consistent, thereby avoiding errors caused by resolution differences. After obtaining the preprocessed sample images, the present invention can more accurately obtain the texture type corresponding to each preprocessed sample image based on the material information and style information in the image, wherein the material information generally refers to the physical properties of the surface of the object in the image, such as the gloss of metal, the texture of wood, etc.; the style information is more related to the overall visual effect of the image, such as simple style, technological style, etc.

[0050] Based on the above embodiment, the step of obtaining a preset input method key pattern image includes: Based on a preset input method key pattern database, the preset input method key pattern image is obtained, wherein the preset input method key pattern database is constructed from preset input method key pattern images.

[0051] In the process of input method skin design, the selection and customization of key styles is a key link. The present invention constructs a rich and comprehensive key style library, namely, a preset input method key style database. This preset input method key style database not only contains existing popular and classic preset key styles, but also incorporates a variety of general key styles carefully drawn by designers. In addition to the preset key styles, the preset input method key style database also has a high degree of scalability and customizability. For example, the size, shape and other attributes of the preset key style can be adjusted.

[0052] Specifically, in the present invention, existing key styles are collected and sorted, and these styles come from various popular input method applications, user feedback, design communities and other channels. By screening and classifying these styles, it is ensured that they are both representative and in line with design aesthetics. At the same time, the common key styles drawn by designers are obtained, which include but are not limited to line draft style, plane style, 2.5D stereo style and special style, etc. Each style can show different visual effects and user experience.

[0053] Finally, these common key style images are integrated with the key styles collected online into a complete preset input method key style database. This database not only contains a rich selection of styles, but also provides high-quality image resources, providing support for subsequent key style generation and customization.

[0054] In one embodiment, when a specific key pattern needs to be generated for a certain input method application, the required key pattern image can be obtained based on the preset input method key pattern database. Through the steps of searching, screening and customizing, a key pattern matching the user's needs can be found. At the same time, by using technical means such as ControlNet control, it can be ensured that the generated key pattern is consistent with the key structure, further improving the user experience.

[0055] Based on the above embodiment, the step of obtaining the target texture image output by the texture image generation model includes: Get the input method skin resource prompt word data; According to the keyword information in the input method skin resource prompt word data, obtaining the target texture type corresponding to the input method skin resource data to be generated, wherein the keyword information includes texture description information and input method skin texture description information; Determining a target texture image generation model from a plurality of texture image generation models according to the texture type label corresponding to the target texture type; The target texture image output by the target texture image generation model is obtained.

[0056] In the present invention, in order to provide rich and diverse skin resources to meet the personalized needs of different users, it is first necessary to obtain the prompt word data of the input method skin resources. These prompt word data can come from market research, user feedback, popular trend analysis, and designer's creative inspiration, etc., and exist in text form, including users' various expectations and descriptions of input method skins, such as color, pattern, style, and texture.

[0057] After obtaining these prompt word data, we conduct in-depth mining and extraction of the keyword information. These keyword information not only includes the user's specific description of the input method skin texture (such as "smooth", "rough", "metal texture", etc., which can intuitively determine the texture type), but also covers the descriptive information related to the input method skin texture (such as "detail level", "tactile feedback", etc.). This information provides an important basis for the subsequent determination of the target texture type corresponding to the input method skin resource data to be generated.

[0058] Furthermore, after determining the target texture type, the target texture image generation model is screened and determined from multiple texture image generation models. These texture image generation models are obtained through a large amount of training and optimization in the early stage, and they can generate images with corresponding textures according to different input conditions (such as texture type, color, pattern, etc.). In order to find the model that best matches the target texture type, the present invention compares and matches the target texture type with the training data of each model, so as to determine the most suitable texture image generation model.

[0059] Finally, by inputting the corresponding parameters and conditions (such as color, pattern, size, etc.), the target texture image generation module will generate the expected texture images. These texture images not only have the expected texture type, but also can visually present various details and features related to the texture, which can be used in the subsequent generation process of input method skin resources.

[0060] On the basis of the above embodiment, the target input method skin resource data is generated according to the preset input method key pattern image and the target texture image, including: Constructing target input method skin resource description information according to the texture description information and the input method skin texture description information; The target input method skin resource description information is used as the image generation guidance information of the trained deep learning model, and the preset input method key style image and the target texture image are input into the trained deep learning model to obtain the target input method skin resource data output by the trained deep learning model.

[0061] In the present invention, the texture description information is usually an abstract description of the appearance effect of the input method skin, such as "silky smooth", "metal texture", "simple style" and "retro style". The input method skin texture description information includes some detailed description indicators, such as skin texture details, glossiness, color gradient, etc., as well as image requirement indicators, such as resolution, file format, color mode, etc. These information together constitute a complete description of the target input method skin resource, providing a clear direction and basis for subsequent skin generation.

[0062] After determining the description information of the target input method skin resource, the trained deep learning model is used to generate the actual input method skin resource data. In the present invention, the trained deep learning model is a ControlNet model, which is a model that can generate images that meet the expected style and style based on the input control picture and specific guidance information.

[0063] Specifically, the present invention uses versions such as xinsir_anime_painter or xinsir_controlnetcanny-sdxl_V2 in the ControlNet model. These versions have been trained and optimized in large quantities and can handle various input conditions well and generate high-quality images. When generating the target input method skin resource, the previously determined texture description information and the input method skin texture description information need to be input into the model as image generation guidance information. At the same time, it is also necessary to provide a preset input method key style image as a control picture, and a target texture image as a reference.

[0064] During the model operation, the setting of control parameters is also very critical. For the ControlNet model, the present invention sets the value of the control parameter to be usually between 0.8 and 1.2. This parameter determines the degree of reliance of the model on the control image and guidance information when generating images. By adjusting the value of the control parameter, input method skin resources of different styles and details can be obtained.

[0065] Finally, after the model's calculation and processing, the target input method skin resource data output by the trained deep learning model is obtained. These data not only meet the previously determined texture description information and texture index requirements, but also visually present a style and features similar to the preset key style image and the target texture image.

[0066] On the basis of the above embodiment, after generating the target input method skin resource data according to the preset input method key pattern image and the target texture image, the method further includes: Receiving a first input instruction, wherein the first input instruction includes an operation instruction to adjust the target input method skin resource data; In response to the first input instruction, the text content, icon content or punctuation content corresponding to the target input method skin resource data is adjusted to obtain adjusted target input method skin resource data.

[0067] In the present invention, the generated target input method skin resource data can also be adjusted to a certain extent, and such adjustment may involve the modification of elements such as text, icons or punctuation marks. When the user terminal makes such adjustments, a corresponding instruction is input through a specific interface (such as a setting menu, an editing interface, etc.), i.e., a first input instruction, which usually includes the specific content of the adjustment, such as modifying the text color, size, or replacing the icon, or adjusting the display mode of punctuation marks, etc.

[0068] After receiving the first input instruction, it is necessary to first parse the first input instruction to clarify which part of the content is currently being adjusted (text content, icon content or punctuation content), as well as the specific adjustment method (such as color change, size adjustment, style change, etc.). Then, according to the parsed instruction, the corresponding part in the target resource data is adjusted. For example, if the text color is required to be changed, find the corresponding text element and modify its color attribute. If the icon is changed, the original icon resource will be replaced. During the adjustment process, the present invention can provide a real-time preview effect, so that the user end can instantly obtain the adjusted effect for further adjustment or confirmation.

[0069] Figure 6 The overall schematic diagram of the input method skin resource generation process provided by the present invention can be referred to Figure 6As shown, taking the generation of a 2.5D input method skin with a metallic texture as an example, the specific steps are as follows: First, select the basic model style as "Normal 2.5D_V1.0.safetensors" or other basic models with three-dimensional texture in WebUI, select the trained Lora model that can generate metal texture images, and select the general keyboard draft with 2.5D effect as the image control guidance information of the ControlNet model. Then, according to the actual input method skin generation requirements, construct the corresponding prompt words, for example, the positive prompt word "C4D, best, Metallic texture, more details, excellent, high quality", can be changed according to the current effect; the negative prompt word "ng_deepnegative_v1_75t, (badhandv4:1.2), EasyNegative, (worstquality:2)".

[0070] At the same time, modify the size of the output image, which is consistent with the key draft size in the input method skin design specification, and the maximum cannot exceed 1024 pixels. Sampling method selections: Euler, DPM++ 2M, Euler A. Furthermore, enable the ControlNet model, select lineart standard (standard line draft extraction - white background with black lines inverted) for the preprocessor, select xinsir_anime_painter for the processing model, and select 1 for the control weight. The final result generated by the ControlNet model is a 2.5D metal texture key skin with a metallic texture, and a specific effect preview image is generated. In addition, in the present invention, the generated input method skin resources can be fine-tuned according to the effect preview image, and the fine-tuning content includes text, icons, punctuation, etc.

[0071] The present invention introduces and trains the Lora model for the generation technology of input method skin resources, and uses its powerful processing power to accurately present the design effect. By accurately adjusting specific parameters (such as image size 1024×768), the size and range of the output image are effectively constrained. Through the input method key style base map, the ControlNet model is used to ensure that the final generated result meets the key specification standard of the input method skin, especially the adaptability of the key area. Optionally, in the present invention, image processing technology is used to finely cut the design draft to improve the processing efficiency and ensure the accurate separation and positioning of the skin elements. Subsequently, these processed images will be automatically uploaded to the existing skin production platform, and the one-key generation function of the platform will be used to quickly complete the production and release of skin resources. This optimization process not only significantly reduces the frequency and depth of manual intervention, breaks the constraints of professional skills on innovation ability, achieves a leap in skin production efficiency (specifically up to 900%), but also greatly reduces labor costs.

[0072] The input method skin resource generation system provided by the present invention is described below. The input method skin resource generation system described below and the input method skin resource generation method described above can be referenced to each other.

[0073] Figure 7 The structure diagram of the input method skin resource generation system provided by the present invention is as follows: Figure 7 As shown, the present invention provides an input method skin resource generation system, including a texture image generation module 701 and an input method skin resource generation module 702, wherein the texture image generation module 701 is used to obtain a target texture image output by a texture image generation model, wherein the texture image generation model is trained based on a sample texture image marked with a texture type label; the input method skin resource generation module 702 is used to obtain a preset input method key style image, and generate target input method skin resource data based on the preset input method key style image and the target texture image.

[0074] The input method skin resource generation system provided by the present invention determines a target texture image generation model from multiple texture image generation models, and then constructs target input method skin resource data according to a preset input method key style image and a target texture image output by the target texture image generation model, thereby improving the generation efficiency of input method skin resources.

[0075] On the basis of the above embodiments, the input method skin resource generation system also includes a model training module, which is specifically used to determine the texture type corresponding to each sample image according to the material information and style information in the sample image; according to the texture type, mark the corresponding texture type label for each sample image to obtain a sample texture image marked with the texture type label; construct a training sample set corresponding to each texture type label, wherein the texture type labels of the sample texture images in the same training sample set are the same; according to each training sample set, train the large language model corresponding to each training sample set to obtain the texture image generation model corresponding to each texture type label.

[0076] On the basis of the above embodiment, the model training module is also used to preprocess each of the sample images based on preset image resolution information to obtain a preprocessed sample image, wherein the resolutions of the preprocessed sample images are the same; and determine the texture type corresponding to each of the preprocessed sample images according to the material information and the style information in the preprocessed sample images.

[0077] Based on the above embodiment, the input method skin resource generation module is specifically used to obtain the preset input method key style image based on the preset input method key style database, wherein the preset input method key style database is constructed from the preset input method key style image.

[0078] On the basis of the above embodiments, the texture image generation module is specifically used to obtain input method skin resource prompt word data; according to the keyword information in the input method skin resource prompt word data, obtain the target texture type corresponding to the input method skin resource data to be generated, wherein the keyword information includes texture description information and input method skin texture description information; according to the texture type label corresponding to the target texture type, determine the target texture image generation model from multiple texture image generation models; obtain the target texture image output by the target texture image generation model.

[0079] Based on the above embodiment, the input method skin resource generation module is also used to construct target input method skin resource description information based on the texture description information and the input method skin texture description information; use the target input method skin resource description information as image generation guidance information of the trained deep learning model, and input the preset input method key style image and the target texture image into the trained deep learning model to obtain the target input method skin resource data output by the trained deep learning model.

[0080] Based on the above embodiment, the input method skin resource generation system is also used to receive a first input instruction, wherein the first input instruction includes an operation instruction to adjust the target input method skin resource data; in response to the first input instruction, the text content, icon content or punctuation content corresponding to the target input method skin resource data is adjusted to obtain the adjusted target input method skin resource data.

[0081] The system provided in the embodiment of the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific process and detailed contents, which will not be repeated here.

[0082] Figure 8 A schematic diagram of the structure of an electronic device provided by the present invention, such as Figure 8 As shown, the electronic device may include: a processor (Processor) 801, a communication interface (Communications Interface) 802, a memory (Memory) 803 and a communication bus 804, wherein the processor 801, the communication interface 802, and the memory 803 communicate with each other through the communication bus 804. The processor 801 may call the logic instructions in the memory 803 to execute the input method skin resource generation method, the method comprising: obtaining a target texture image output by a texture image generation model, wherein the texture image generation model is obtained by training based on a sample texture image marked with a texture type label; obtaining a preset input method key style image, and generating target input method skin resource data according to the preset input method key style image and the target texture image.

[0083] In addition, the logic instructions in the above-mentioned memory 803 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk.

[0084] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the input method skin resource generation method provided by the above-mentioned methods, and the method includes: obtaining a target texture image output by a texture image generation model, wherein the texture image generation model is trained based on a sample texture image marked with a texture type label; obtaining a preset input method key style image, and generating target input method skin resource data based on the preset input method key style image and the target texture image.

[0085] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the input method skin resource generation method provided in the above-mentioned embodiments, the method comprising: obtaining a target texture image output by a texture image generation model, wherein the texture image generation model is trained based on a sample texture image marked with a texture type label; obtaining a preset input method key style image, and generating target input method skin resource data based on the preset input method key style image and the target texture image.

[0086] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0087] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating input method skin resources, characterized in that: include: Acquire a target texture image output by a texture image generation model, wherein the texture image generation model is trained based on sample texture images marked with texture type labels; A preset input method key style image is obtained, and target input method skin resource data is generated according to the preset input method key style image and the target texture image.

2. The input method skin resource generation method according to claim 1, characterized in that: The texture image generation model is trained by the following steps: Determining the texture type corresponding to each of the sample images according to the material information and style information in the sample images; According to the texture type, marking each of the sample images with a corresponding texture type label to obtain a sample texture image marked with the texture type label; Constructing training sample sets corresponding to the texture type labels, wherein the texture type labels of the sample texture images in the same training sample set are the same; According to each of the training sample sets, the large language model corresponding to each of the training sample sets is trained to obtain the texture image generation model corresponding to each of the texture type labels.

3. The input method skin resource generation method according to claim 2, characterized in that: Before determining the texture type corresponding to each of the sample images according to the material information and style information in the sample images, the method further includes: Based on preset image resolution information, preprocessing each of the sample images to obtain a preprocessed sample image, wherein the resolutions of the preprocessed sample images are the same; The step of determining the texture type corresponding to each sample image according to the material information and style information in the sample image includes: According to the material information and the style information in the preprocessed sample images, the texture type corresponding to each of the preprocessed sample images is determined.

4. The input method skin resource generation method according to claim 1, characterized in that: The step of obtaining a preset input method key pattern image includes: Based on a preset input method key pattern database, the preset input method key pattern image is obtained, wherein the preset input method key pattern database is constructed from preset input method key pattern images.

5. The input method skin resource generation method according to claim 2 or 4, characterized in that: The step of obtaining a target texture image output by the texture image generation model includes: Get the input method skin resource prompt word data; According to the keyword information in the input method skin resource prompt word data, obtaining the target texture type corresponding to the input method skin resource data to be generated, wherein the keyword information includes texture description information and input method skin texture description information; Determining a target texture image generation model from a plurality of texture image generation models according to the texture type label corresponding to the target texture type; The target texture image output by the target texture image generation model is obtained.

6. The input method skin resource generation method according to claim 5, characterized in that: The step of generating target input method skin resource data according to the preset input method key style image and the target texture image includes: Constructing target input method skin resource description information according to the texture description information and the input method skin texture description information; The target input method skin resource description information is used as the image generation guidance information of the trained deep learning model, and the preset input method key style image and the target texture image are input into the trained deep learning model to obtain the target input method skin resource data output by the trained deep learning model.

7. The input method skin resource generation method according to claim 1, characterized in that: After generating the target input method skin resource data according to the preset input method key pattern image and the target texture image, the method further includes: Receiving a first input instruction, wherein the first input instruction includes an operation instruction to adjust the target input method skin resource data; In response to the first input instruction, the text content, icon content or punctuation content corresponding to the target input method skin resource data is adjusted to obtain adjusted target input method skin resource data.

8. An input method skin resource generation system, characterized in that: include: A texture image generation module, used to obtain a target texture image output by a texture image generation model, wherein the texture image generation model is trained based on sample texture images marked with texture type labels; The input method skin resource generation module is used to obtain a preset input method key style image, and generate target input method skin resource data according to the preset input method key style image and the target texture image.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the input method skin resource generation method as claimed in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the input method skin resource generation method as claimed in any one of claims 1 to 7 is implemented.

Images