Information processing method, information processing device, and program

Abstract

To easily set an appropriate prompt for obtaining desired content by a generative AI.SOLUTION: An information processing device for generating content by a generative AI comprises: deriving means for deriving a specific character string on the basis of information obtained about a user; and setting means for setting the specific character string derived by the deriving means as a negative prompt which specifies what content should not be generated for the generative AI.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present disclosure relates to prompt settings for image generation AI. [Background technology]

[0002] One of the functions of software for creating posters, flyers, etc. is to allow a user to select a desired template from a variety of pre-prepared templates and insert any text or an image taken by the user into the template. In this regard, Patent Document 1 discloses a technology that enables a search for an image suitable for the template from a group of separately prepared images by extracting words from property information of objects included in the template and creating search keywords. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2017-37557 Summary of the Invention [Problem to be solved by the invention]

[0004] For example, some software for creating posters and other content now features the ability to generate images using generative AI (artificial intelligence). This image generation function allows users to input words or phrases as prompts into the AI, which then automatically generates an image based on the input prompt (words or phrases). Suppose the poster template to be used is an illustration-style template. When inserting an image into an illustration-style template, the user expects the image to be similarly illustration-style, and inputs an arbitrary prompt into the AI. However, if the user inputs an inappropriate prompt, the AI ​​may generate a photorealistic image that does not match the style or atmosphere of the template. In such cases, the user must repeatedly generate the image using the AI, for example by inputting a different prompt, which is time-consuming for the user. Thus, it is difficult and time-consuming for users to find the appropriate prompt to input into the AI ​​to obtain the desired content. [Means for solving the problem]

[0005] The information processing device according to the present disclosure is generated by Content Get the embedded artifacts and based on information obtained about a user, A template selected by the user from a plurality of templates prepared for the deliverable is pre-assigned. The system is characterized by comprising a derivation means for deriving a specific character string, and a setting means for setting the specific character string derived by the derivation means as a negative prompt that specifies to the generation AI what kind of content it should not generate. [Effects of the Invention]

[0006] According to the present disclosure, it is possible to easily set an appropriate prompt to obtain desired content. [Brief explanation of the drawings]

[0007] [Figure 1]FIG. 1 is a diagram showing an example of the configuration of a client-server system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a diagram showing an example of a hardware configuration of an information processing apparatus. [Figure 3] FIG. 2 is a functional block diagram showing an example of the software configuration of a front-end application and a back-end application. [Figure 4] FIG. 10 is a diagram showing an example of a negative prompt. [Figure 5] 10 is a flowchart showing the flow of operations in a client. [Figure 6] 10A and 10B are diagrams showing an example of a poster editing screen. [Figure 7] 10A and 10B are diagrams showing an example of a poster editing screen. [Figure 8] 10A and 10B are diagrams showing an example of a poster editing screen. [Figure 9] 10 is a flowchart showing the flow of operations in the server. [Figure 10] 10A and 10B are diagrams showing an example of a poster editing screen. [Figure 11] FIG. 10A is a diagram for explaining how to derive a negative prompt character string, and FIG. 10B is a diagram showing an example of a UI screen for setting a negative prompt from the derived character string group. [Figure 12] (a) to (c) are diagrams explaining how to obtain an impression of an image using a trained model. [Figure 13] (a) and (b) are diagrams illustrating the automatic updating of negative prompts. DETAILED DESCRIPTION OF THE INVENTION

[0008] Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the present invention, and not all of the combinations of features described in the embodiments are necessarily essential to the solution of the present invention.

[0009] [Embodiment 1] <System configuration> 1 is a diagram showing an example of the configuration of a client-server system according to this embodiment. This system is composed of a server 101 that provides a content generation service and a client 102 that uses the service. The server 101 and the client 102 communicate with each other from their respective networks 103 via the Internet 104. The server 101 has a back-end application 105. The client 102 has an operation system 106 and a front-end application 107.

[0010] <Hardware configuration> FIG. 2 illustrates an example of a hardware configuration common to the server 101 and client 102, which are information processing devices. The user interface 201 is an operation display unit that inputs and outputs information and signals and is composed of a display, keyboard, mouse, buttons, touch panel, etc. The network interface 202 connects to a network such as a LAN and communicates with other computers and network devices. The communication method may be wired or wireless. The CPU 203 is a central processing unit that executes programs loaded from the ROM 204, RAM 205, secondary storage device 206, etc. The ROM 204 is a non-volatile memory that stores pre-installed programs and data. The RAM 205 is a volatile memory that provides temporary memory space. The secondary storage device 206 is a large-capacity storage device such as a hard disk drive (HDD) or flash memory. Computers that do not have this hardware can be connected to and operated from other computers using remote desktop, remote shell, etc. Each unit is connected via an input / output interface 207.

[0011] <Software configuration> 3 is a functional block diagram showing an example of the software configuration of each of the front-end application 107 and the back-end application 105. The front-end application 107 has a GUI control unit 301, a negative prompt derivation unit 302, a prompt setting unit 303, and a request processing unit 304. The back-end application 105 has a response processing unit 311 and a content generation unit 312. Note that one of the front-end application 107 and the back-end application 105 may have a configuration including all of the above-mentioned units 301 to 304, 311, and 312.

[0012] The GUI control unit 301 controls the GUI (Graphical User Interface) for presenting information to the user or allowing the user to input instructions. Specifically, it controls the display of the UI screen and the acceptance of user input. Specific examples of user input include template selection, instructions on how to insert content, and the acceptance of character strings as prompts (instruction information consisting of words or sentences) to be input to the generation AI. Prompts are divided into positive and negative prompts. Positive prompts specify the type of content the generation AI should generate (the elements it wants to generate). For example, if "reindeer" is input as a positive prompt to an AI that generates images as content (image generation AI), the image generation AI will generate an image of a reindeer. In contrast, negative prompts specify the type of content the generation AI should not generate (the elements it wants to exclude). Figure 4 is a table summarizing examples of common negative prompts input to image generation AI. For example, if "low quality" is input as a negative prompt to the image generation AI, the image generation AI will not generate low-quality images.

[0013] The negative prompt derivation unit 302 derives a character string to be used as the negative prompt from among the prompts to be input to the generation AI based on information obtained about the user. The derivation method will be described later.

[0014] The prompt setting unit 303 sets the character string input by the user via the GUI and the character string derived by the negative prompt derivation unit 302 as a positive prompt and a negative prompt, respectively.

[0015] The request processing unit 304 performs processing to request the server 101 to generate content. When making this request, the request processing unit 304 also sends the positive prompt and negative prompt set by the prompt setting unit 303. Furthermore, the request processing unit 304 also performs processing to receive the content generated by the server 101 in response to the request.

[0016] The response processing unit 311 receives a request from the client 102 and performs response processing such as transmitting content generated in response to the received request to the client 102 that made the request.

[0017] The content generation unit 312 is a generation AI that generates content such as images and text using positive and negative prompts as input. Known image generation AIs that generate images as content include "Stable Diffusion" and "Midjourney." The generation AI is a trained model (content generation model) obtained by performing machine learning on various data using techniques such as deep learning to obtain the desired content.

[0018] Next, the flow of operations of the client 102 and the server 101 will be described using poster creation software as an example of the front-end application 107. Note that the poster creation software is an example of software installed on the client 102, and is not limited to this. For example, the present disclosure can be applied to software for obtaining various deliverables, such as photo album creation software and postcard creation software.

[0019] <Client side operation> 5 is a flowchart showing the flow of operations in the client 102 according to this embodiment. This flowchart begins when the poster creation software is launched based on a user instruction. In the following explanation, the symbol "S" denotes a step.

[0020] In S501, the GUI control unit 301 displays an editing UI screen (hereinafter referred to as a "poster editing screen") compatible with the poster creation software on the display of the user interface 201. FIG. 6(a) shows an example of the poster editing screen. The poster editing screen 600 in FIG. 6(a) is made up of a template list pane 610, a template editing pane 620, and a content editing pane 630. The GUI control unit 301 displays a list of various templates suited to purposes and uses in the template list pane 610. In the example of FIG. 6(a), three pre-prepared templates (611, 612, 613) are displayed in the template list pane 610.

[0021] In S502, the GUI control unit 301 accepts a user selection of a specific template from among the templates displayed in the template list pane 610. FIG. 6(b) shows the change in the screen when the user selects template 612, with a highlight frame 640 added to indicate that the template is selected. Furthermore, the template 612 selected by the user is displayed in the template editing pane 620 as a template 621 to be edited. Now, the template 621 to be edited has an image setting area 622 in which the user can set an arbitrary image, and a text setting area 623 in which the user can set arbitrary text. Hereinafter, the image setting area and text setting area will be collectively referred to as the "content setting area."

[0022] In S503, the GUI control unit 301 determines whether any of the content setting areas in the template to be edited displayed in the template editing pane 620 has been pressed. If a press on a content setting area is detected, the GUI control unit 301 then executes S504, and if a press on a content setting area is not detected, the GUI control unit 301 waits for a predetermined time to elapse before determining whether a press has been detected.

[0023] In S504, the GUI control unit 301 displays a pop-up screen for the user to select a content addition method according to the pressed content setting area, and accepts the user's selection of whether to manually add or automatically generate the target content. Figure 7(a) shows an example in which the user operates the pointer to press the image setting area 622 in the state shown in Figure 6(b), displaying a pop-up screen for selecting whether to automatically generate or manually add the image to be added. Currently, the "Automatic Generation" and "Select from a folder" buttons are displayed as options for the image addition method. Here, the "Automatic Generation" button is used when the generation AI generates an image, and the "Select from a folder" button is used when the user selects and uploads an image from any folder. If the text setting area 623 was pressed in S503, the "Automatic Generation" and "Manual Input" buttons are displayed as options for the text addition method. In this case, the "Automatic Generation" button is used when the generation AI generates text, and the "Manual Input" button is used when the user manually inputs any character string. If the GUI control unit 301 detects that the manual addition button has been pressed, it then executes S505, and if the GUI control unit 301 detects that the automatic generation button has been pressed, it then executes S507.

[0024] In S505, the GUI control unit 301 accepts content designation by the user. For example, if pressing of the "Select from a folder" button for manually adding an image is detected in S504, designation of desired image data from any folder is accepted. The user can specify a desired image by an operation such as drag and drop from among images that the user has taken themselves or obtained via the Internet 104 and that are previously stored in a folder. Also, for example, if pressing of the "Manual Input" button for manually adding text is detected in S504, an input field (not shown) for directly inputting a desired character string is displayed, and designation of a character string via the input field is accepted.

[0025] In S506, the GUI control unit 301 inserts the content specified by the user in S505 into the content setting area of ​​the selected template that was pressed in S503. FIG. 7B shows an example of a state in which an image 710 specified by the user has been inserted into the image setting area 622. Once the content has been inserted, an operation field 720 is displayed in the content editing pane 630, allowing the user to perform editing operations on the inserted content. In FIG. 7B, the inserted image 710 can be edited. In this example, a field 721 for specifying whether to automatically correct the image to be edited, a control bar 722 for adjusting the brightness of the image, and a control bar 723 for adjusting the contrast of the image are provided. If the content to be edited is text, an operation field (not shown) is also displayed in which the font, size, color, and other parameters of the text can be adjusted. The user can modify the content by performing the necessary operations in the operation field appropriate for the content to be edited.

[0026] S507 to S513 are processes for having the generation AI automatically generate and insert content for the content setting area pressed by the user. Figures 8(a) and 8(b) are examples of UI screens when the user selects the Christmas poster template 611 in the state of Figure 6(a), then presses the image setting area 622 of the template 621 to be edited and selects automatic image generation. Below, we will explain the flow from having the generation AI automatically generate content to completing the template, with appropriate reference to the UI screen example.

[0027] First, in S507, the GUI control unit 301 accepts input of a string of characters, such as words, that represent elements of the content that the user wants the generation AI to generate. Note that English is commonly used in prompts, and so the prompts are written in English in this embodiment, but it goes without saying that the language used in the prompts depends on the generation AI and is not limited to English. Here, it is assumed that the words "Reindeer" and "Christmas" are input via an input field (not shown).

[0028] In S508, the prompt setting unit 303 sets the character string of the word or the like input by the user in S507 as a positive prompt. Since the two words "Reindeer" and "Christmas" have now been input by the user, these will be set as positive prompts.

[0029] In S509, the negative prompt derivation unit 302 derives character strings, such as words representing content elements that the user does not want the AI ​​to generate, based on the template selected by the user. A possible template-based derivation method is to assign negative prompt character strings corresponding to the characteristics of each template displayed in the template list as metadata and then refer to the metadata of the template selected by the user. Alternatively, a table correlating each template with negative prompt character strings may be prepared in advance and referenced. Furthermore, the derivation may be performed using a trained model that estimates character strings that are inappropriate for the impression of the template selected by the user. The trained model used for this estimation can be obtained by training a large amount of training data in which templates are paired with words, etc. that are inappropriate for the impression of the template. Note that the estimation may involve estimating the overall impression of the target template, or it may involve estimating the impression of each content, such as images and text, contained in the target template and then estimating the overall impression of the template as a whole. In the example of Figure 8(a), a Christmas-themed illustrated template is selected. In this case, even if a realistic image of a reindeer, like a photograph of a reindeer in the wild, is generated, it is not suitable for an illustration-style template. Therefore, a word such as "Realistic" is pre-associated with an illustration-style template using the method described above. This allows the user to simply select a desired template, and the appropriate negative prompt is automatically acquired and set, eliminating the need for the user to think up and set a negative prompt. When an illustration-style template is selected, the user can manually set a string such as "Illustration style" as the positive prompt to have the AI ​​generate an illustration-style image. However, with the method of this embodiment, the user does not need to think up and set such a positive prompt, further reducing the time and effort required.

[0030] In S510, the prompt setting unit 303 sets the character string of the word or the like derived in S509 as the negative prompt. In the above example, the word "Realistic" is automatically set as the negative prompt. Note that, as shown in FIG. 11(b) described below, the user may be allowed to set the word to be set as the negative prompt from the group of words derived in S509 via a GUI.

[0031] In S511, the request processing unit 304 transmits a content generation request to the server 101 that provides the content generation service. This content generation request includes information on the positive prompt set in S508 and the negative prompt set in S510. In response to this content generation request, the server 101 automatically generates content based on the positive prompt and the negative prompt. Details of the automatic generation of content by the server 101 will be described later.

[0032] In S512, the request processing unit 304 receives from the server 101 the content generated based on the content generation request.

[0033] In S513, the GUI control unit 301 inserts the content received in S512 into the content setting area of ​​the selected template selected in S503. In the example of FIG. 8(a), an illustrative reindeer image 800 is inserted into the image setting area 622, and the content editing pane 630 displays an operation field 810 for the user to edit the image 800. The operation field 810 includes two subfields 820 and 830 that indicate the positive and negative prompts used to generate the inserted image 800, as well as a subfield 840 that displays only the generated image. FIG. 8(b) shows a specific example in which "Realistic" is not selected as the negative prompt. The AI ​​generates a realistic reindeer image 801, resembling a photograph of a natural world. The user may be able to fine-tune the color of the image displayed in the subfield 840 using the control bar described above.

[0034] In S514, it is determined whether all the content to be set has been set for the template selected in S502. If there is any content that has not been set, the process returns to S503 and continues. On the other hand, if all the content has been set, the process ends.

[0035] The above is a description of the operation on the client side. Note that in the flow of Figure 5 described above, a step may be provided in which the user confirms on the UI screen whether or not to apply the negative prompt automatically set in S510 as is before the content generation request is sent. Also, the user may add or change the text in subfields 820 and 830 themselves, causing the generation AI to generate a new image.

[0036] The above-mentioned technique can be applied to a variety of cases. For example, if a certain traditional dish has an implicit rule (such as not using a specific ingredient X), a string such as "X as an ingredient that should not be used" can be linked to the template for that traditional dish. This way, even if a user does not know the rules for that traditional dish, when they select that template, the string representing ingredient X will automatically be set in the negative prompt. This prevents the generation AI from mistakenly creating content that includes ingredient X.

[0037] <Server side operation> 9 is a flowchart showing the flow of operations performed by the back-end application 105 of the server 101 to generate content in response to a content generation request (S511) from the client 102. In the following explanation, the symbol "S" denotes a step.

[0038] In S901, the response processing unit 311 receives a content generation request from the client 102. In S902, the content generation unit 312 acquires a positive prompt and a negative prompt from the content generation request received in S901. In S903, the content generation unit 312 generates content using the positive prompt and negative prompt acquired in S902 as input. In S904, the response processing unit 311 transmits the content data generated in S903 to the client 102 that sent the request.

[0039] The above is a description of the operation on the server side. In this embodiment, the back-end application 105 includes the generation AI, but this is not limited to this. For example, the generation AI may be located outside the back-end application 105, and the back-end application 105 may call the external generation AI to respond to a content generation request.

[0040] <Variations of string derivation methods> In the above example, the character string used as the negative prompt is derived by linking a specific character string to each template in advance, but the method for deriving the character string for the negative prompt is not limited to this. Below, variations of the method for deriving the character string for the negative prompt are described.

[0041] <<Derivation based on user attributes>> Typically, for software that provides functions such as poster creation as a cloud-based service, users register an account in advance and log in to use the software. When registering an account, users also register attribute information such as their name, gender, sex, nationality, region, language, and hobbies. Software that provides services around the world will be used by users of various nationalities. Different nationalities have different cultures and customs, and the same gesture may be perceived differently depending on the country. For example, the peace sign, a type of body language, may convey a positive impression in some countries and a negative impression in others. Therefore, if the nationality of the logged-in user's attribute information indicates that the peace sign conveys a negative impression, the "peace sign" string is derived as the negative prompt string. Specifically, the string representing taboo gestures for each country is registered in a database. When a user logs in, the database is queried for the user's nationality and the registered string associated with the user's country is retrieved. 10(a) and 10(b) are examples of UI screens that appear when a user selects template 611 on the UI screen shown in FIG. 6(a) and then clicks the image setting area 622 of the edit target template 621 to select automatic image generation. FIG. 10(a) shows an example where the logged-in user's nationality is a country where the peace sign does not convey a negative impression, while FIG. 10(b) shows an example where the logged-in user's nationality is a country where the peace sign conveys a negative impression. In both cases, the positive prompt is set to "woman is laughing," while in FIG. 10(b), "Peace sign" is also set as a negative prompt. As a result, in the example shown in FIG. 10(a), the AI ​​generates image 1000 of a woman laughing with a peace sign, while in the example shown in FIG. 10(b), the AI ​​generates image 1001 of a woman laughing without making a peace sign. In this way, by setting negative prompts according to the user's nationality, it is possible to prevent the generation of content that is inappropriate for that user.Here, nationality is used as user attribute information. However, other user attributes, such as region (including multiple countries) and language used, may also be used. In this case, character strings representing taboo gestures and the like for each attribute, such as region and language, may be registered in a database, and a character string may be derived for the logged-in user based on these attributes to set the negative prompt. Furthermore, instead of using these attributes at the time of account registration, the user's attributes may be estimated from information obtained after account registration, such as the user's behavioral history or IP address, and the estimated attribute information may be used to derive the character string for the negative prompt. Note that even if a user's nationality conveys a negative impression when the peace sign is used, if the user enters an intentionally insulting character string in the positive prompt, the positive prompt may be prioritized, and the "Peace sign" may not be set as a negative prompt.

[0042] Derivation based on user selection of content Next, we will explain a method for deriving a negative prompt string based on a user's selection of content within a template. Specifically, the impression of an image selected by a user from among images generated by a generation AI is estimated using a trained model for impression estimation (impression estimation model), and a string corresponding to the antonym of the string representing the estimated impression is obtained from a dictionary database. Figure 11(a) is a diagram illustrating how a negative prompt is set based on an image selected by a user from multiple images generated by an image generation model, which is a generation AI. Currently, "Halloween pumpkin" is set as a positive prompt. Using this as input, an image generation model 1110 generates three images 1120, 1121, and 1122 of pumpkins deformed for Halloween. Among these three images, image 1120 has a "cute" impression, while images 1121 and 1122 have "scary" impressions. Now, let's assume that a user selects pumpkin image 1120 with a "cute" impression as the image to be inserted into a template. In this case, when the image 1120 selected by the user is input into the impression estimation model, the impression "cute" is estimated. Next, the word "pretty," which represents the estimated impression, is input into the dictionary database. As a result, words such as "scary," "eerie," "weird," and "frightening," which represent impressions opposite to or different from "cute," are obtained. FIG. 12(a) shows how the impression "pretty," obtained by inputting the image 1120 selected by the user into the impression estimation model 1200, is further input into the dictionary database, and the opposite impression of "pretty," "scary," is derived. FIG. 11(b) shows an example of a UI screen on which the user selects and sets a negative prompt from the derived word group. The GUI control unit 301 of the front-end application 107 displays a candidate dialog 1130 containing the derived word group. The user selects a word representing an impression that the user does not want to generate from the word group displayed in the candidate dialog 1130. In this example, "eerie" is selected and the negative prompt is set to "eerie."Then, when "Halloween bat" is set as the positive prompt and input to the image generation model 1110, images 1140 and 1141 of bats with a "cute" impression are generated. In this way, a character string for a negative prompt may be derived from an image selected by the user using the impression estimation model and dictionary database. Note that in this example, the user selects one word from the derived word group, but the derived word or words may be set as the negative prompt as is.

[0043] Alternatively, the word "pretty," which expresses the impression estimated from the image selected by the user, may be added to the positive prompt. Furthermore, as shown in FIG. 12(b), a trained model (antonym estimation model) 1210 capable of estimating antonyms of words that express the impression of an image may be used to more directly derive a string of characters for a negative prompt from the image selected by the user. In this case, the antonym estimation model can be obtained by training a large amount of training data in which images are paired with words that do not match the image's impression. Note that the images used as training data during training may be generated by a generation AI or may be photographed by a human.

[0044] In the above example, the negative prompt character string is derived based on the image 1120 selected by the user. However, as shown in FIG. 12(c), it may also be derived based on images 1121 and 1122 not selected by the user. Now, the images 1121 and 1122 not selected by the user are both images that give a "scary" impression. Therefore, by inputting these into the impression estimation model 1200, it is possible to estimate "scary," which represents the impression common to both images. Note that the number of images input to the impression estimation model may be three or more, or may be just one.

[0045] Although the example has been described in which the image is derived based on an image not selected by the user among the images generated by the generation AI, the invention is not limited to this. For example, the image may be derived based on an image not selected by the user among images for inserting into a template prepared in advance by poster creation software.

[0046] Furthermore, images not selected by the user can be used as bad examples, and additional learning can be performed using a method such as LoRA (Low-Rank Adaptation) to link them to the user and update the generation AI. This means that the next time each user uses the system, the generation AI updated for each user will be used, making it less likely that images that do not match the preferences of the logged-in user will be generated.

[0047] As the impression estimation model and the antonym estimation model, it is assumed that they are trained models trained by a deep learning method, for example, like the content generation model, but they are not limited to this.

[0048] <Other variations> The automatically set negative prompts may be automatically updated as the user edits the template. Figures 13(a) and 13(b) are diagrams illustrating the automatic updating of negative prompts. Assume that "realistic" and "pretty" are pre-assigned to template 1300 as negative prompts. The user then instructs the AI ​​to automatically add an image by using "Halloween pumpkin" as a positive prompt. As a result, the AI ​​generates an illustrative and "scary" pumpkin image 1301, which is placed in the upper right corner of template 1300. In Figure 13(a), V1 and V2 are vector representations of "Realistic" and "pretty," respectively, converted using a technique such as Word2Vec. Note that the method for converting words into vector representations is not limited to Word2Vec, and other methods may also be used. Next, assume that the user inserts pumpkin image 1302 into the lower left corner of template 1300 by dragging and dropping, for example. This pumpkin image 1302 is then input into the impression estimation model 1200, and the impression "cute" is estimated. In FIG. 13(a), V represents the estimated impression "cute" converted into a vector representation. In this case, the cosine similarity S1 between V and V1 and the cosine similarity S2 between V and V2 are calculated. Comparing the two results reveals that the value of S2 is larger. This is because "pretty" is closer to the meaning of "cute" than "realistic." If the impression is highly similar to the impression of an image added by the user, it is not suitable as a negative prompt. Therefore, "pretty" is deleted from the negative prompt, as shown in FIG. 12(b). As a result, the negative prompt for the template 1300 is updated to "realistic" only. In this way, negative prompts may be automatically updated during the user's template editing process. For example, if a user specifies "Halloween bat" as a positive prompt and commands automatic image generation, a bat image with a "cute" impression will also be generated.By automatically updating the negative prompts in this way in response to the user's editing operations, the generation AI can generate content that matches the impression that has changed due to editing.

[0049] Furthermore, in the above-described embodiment, the user can check the automatically set negative prompt on the UI screen, but the negative prompt may be applied in a manner that is not displayed on the UI screen and is invisible to the user.

[0050] Furthermore, the expression of the negative prompt automatically derived and set by the above-described embodiment may be changed by the user, for example, from "cute" to "pretty."

[0051] (Other Examples) The present invention can also be realized by supplying a program that realizes one or more functions of the above-described embodiments to a system or device via a network or a storage medium, and having one or more processors in the computer of the system or device read and execute the program.The present invention can also be realized by a circuit (e.g., ASIC) that realizes one or more functions.

[0052] The disclosure of this embodiment also includes the following configurations and methods.

[0053] [Configuration 1] An information processing device for causing a generation AI to generate content, deriving means for deriving a specific character string based on information obtained about the user; A setting means for setting the specific character string derived by the derivation means as a negative prompt that specifies to the generation AI what kind of content it should not generate; An information processing device comprising:

[0054] [Configuration 2] software for obtaining a deliverable incorporating the content generated by the generation AI is installed in the information processing device; the deriving means derives a specific character string that is pre-associated with the selected template based on the user's selection of a template prepared for the deliverable; 2. The information processing device according to configuration 1,

[0055] [Configuration 3] Specific character strings corresponding to the respective characteristics of the plurality of templates are assigned as metadata to the plurality of templates, the deriving means derives a specific character string associated with the template selected by the user by referring to metadata of the template; 3. The information processing device according to configuration 2.

[0056] [Configuration 4] The information processing device according to configuration 2, wherein the derivation means derives a specific character string associated with the template selected by the user by referring to a table that associates each of the plurality of templates with a character string for a negative prompt.

[0057] [Configuration 5] The information processing device according to configuration 1, wherein the derivation means uses a trained model that estimates character strings that do not match the impression of the input template to derive a character string that does not match the impression of the template selected by the user as the specific character string.

[0058] [Configuration 6] a control means for controlling a GUI for the user; The information processing device according to any one of configurations 2 to 5, characterized in that the control means automatically updates the negative prompt set by the setting means based on an editing operation by the user on the GUI for the template selected by the user.

[0059] [Configuration 7] The information processing device described in configuration 6, characterized in that the automatic update is a process of deleting the negative prompt when the impression of a specific string set by the setting means as a negative prompt is similar to the impression of content that the user added to the template selected by the user.

[0060] [Configuration 8] 2. The information processing device according to configuration 1, wherein the deriving means derives the specific character string based on attribute information indicating attributes of the user.

[0061] [Configuration 9] 9. The information processing device according to configuration 8, wherein the derivation means derives a specific character string corresponding to an attribute specified by the attribute information, using a database in which the specific character string is registered for each attribute.

[0062] [Configuration 10] The information processing device according to configuration 9, characterized in that the attributes identified by the attribute information are any of the user's gender, age, hobbies, nationality, region including the country of said nationality, and language used by said user.

[0063] [Configuration 11] 11. The information processing device according to configuration 10, wherein the derivation means estimates the attribute using information on the user's behavior history or the user's IP address to perform the derivation.

[0064] [Configuration 12] The information processing device is an information processing device in which software for obtaining a deliverable incorporating the content generated by the generation AI is installed, a control means for controlling a GUI for the user; the control means accepts, via the GUI, the user's selection of content in a template prepared for the deliverable; the deriving means derives, based on the user's selection of content in a template prepared for the deliverable, a character string that is an antonym of a character string that expresses an impression of the selected content, or a character string that expresses an impression of content that the user did not select, as the specific character string; 2. The information processing device according to configuration 1,

[0065] [Configuration 13] The information processing device described in configuration 12, characterized in that when the derivation means derives a character string equivalent to an antonym of a character string expressing the impression of the content selected by the user as the specific character string, it estimates the impression of the content selected by the user using a trained model that estimates the impression of the input content, and derives a character string expressing an impression opposite to the estimated impression.

[0066] [Configuration 14] 14. The information processing device according to configuration 13, wherein the deriving means derives a character string corresponding to an antonym of a character string expressing the estimated impression using a dictionary database.

[0067] [Configuration 15] The information processing device described in configuration 12, characterized in that when the derivation means derives a character string representing an impression of content not selected by the user as the specific character string, it estimates the impression of the content not selected by the user using a trained model that estimates the impression of input content, and derives a character string representing the estimated impression.

[0068] [Configuration 16] 16. The information processing device according to any one of configurations 12 to 15, wherein the generation AI is updated by performing additional learning using content not selected by the user as a bad example.

[0069] [Configuration 17] a control means for controlling a GUI for the user; The information processing device according to any one of configurations 1, 8, and 12, characterized in that the control means displays the negative prompt set by the setting means on the GUI and accepts the user's selection for the set negative prompt.

[0070] [Configuration 18] An information processing device described in any one of configurations 1 to 17, further comprising a generation means for inputting positive prompts that specify what kind of content the generation AI should generate and negative prompts set by the setting means into the generation AI to generate content.

[0071] [Configuration 19] An information processing device described in any one of configurations 1 to 17, further comprising a processing means for requesting an external information processing device having the generation AI to generate content together with a positive prompt that specifies what kind of content the generation AI should generate and a negative prompt set by the setting means, and receiving the content generated based on the request.

[0072] [Configuration 20] 20. The information processing device according to any one of configurations 1 to 19, wherein the content is an image.

[0073] [Configuration 21] 20. The information processing device according to any one of configurations 1 to 19, wherein the content is text.

[0074] [Method 1] An information processing method for causing a generation AI to generate content, a deriving step of deriving a specific character string based on information obtained about the user; a setting step of setting the specific character string derived in the derivation step as a negative prompt that specifies to the generation AI what content should not be generated; An information processing method comprising:

[0075] [Configuration 22] A program for causing a computer to function as the information processing device according to any one of configurations 1 to 21.

Claims

1. An information processing device for obtaining a deliverable incorporating content generated by a generation AI, a derivation means for deriving a specific character string that is pre-associated with a template selected by the user from among a plurality of templates prepared for the deliverable, based on information obtained about the user; A setting means for setting the specific character string derived by the derivation means as a negative prompt that specifies what content the generation AI should not generate; An information processing device comprising:

2. Specific character strings corresponding to the characteristics of each of the plurality of templates are assigned as metadata to the plurality of templates, the deriving means derives a specific character string associated with the template selected by the user by referring to metadata of the template; 2. The information processing apparatus according to claim 1, wherein:

3. 2. The information processing device according to claim 1, wherein the deriving means derives a specific character string associated with the template selected by the user by referring to a table that associates each of the plurality of templates with a character string for a negative prompt.

4. 2. The information processing device according to claim 1, wherein the deriving means derives, as the specific character string, a character string that does not match the impression of the template selected by the user, using a trained model that estimates a character string that does not match the impression of the input template.

5. a control means for controlling a GUI for the user; 2. The information processing apparatus according to claim 1, wherein the control means automatically updates the negative prompt set by the setting means based on an editing operation by the user on the GUI for the template selected by the user.

6. The automatic update is a process of deleting a negative prompt when an impression of a specific character string set as a negative prompt by the setting means is similar to an impression of content added by the user to a template selected by the user.

6. The information processing device according to claim 5, wherein:

7. 2. The information processing apparatus according to claim 1, wherein the deriving means derives the specific character string based on attribute information indicating attributes of the user.

8. 8. The information processing apparatus according to claim 7, wherein the deriving means derives a specific character string corresponding to an attribute specified by the attribute information, using a database in which the specific character string is registered for each attribute.

9. 9. The information processing device according to claim 8, wherein the attributes specified by the attribute information are any one of the user's gender, age, hobbies, nationality, a region including the country of said nationality, and a language used by said user.

10. 10. The information processing apparatus according to claim 9, wherein the deriving means performs the deriving by estimating the attribute using information on the user's behavior history or the user's IP address.

11. a control means for controlling a GUI for the user; the control means accepts, via the GUI, the user's selection of content within a template prepared for the deliverable; the deriving means derives, based on the user's selection of content in a template prepared for the deliverable, a character string that is an antonym of a character string that expresses an impression of the selected content, or a character string that expresses an impression of content that the user did not select, as the specific character string; 2. The information processing apparatus according to claim 1, wherein:

12. 12. The information processing device according to claim 11, wherein when the derivation means derives a character string corresponding to an antonym of a character string expressing an impression of the content selected by the user as the specific character string, the derivation means estimates the impression of the content selected by the user using a trained model that estimates the impression of input content, and derives a character string expressing an impression opposite to the estimated impression.

13. 13. The information processing apparatus according to claim 12, wherein the deriving means derives a character string corresponding to an antonym of a character string expressing the estimated impression by using a dictionary database.

14. 12. The information processing device according to claim 11, wherein, when deriving a character string representing an impression of content not selected by the user as the specific character string, the derivation means estimates the impression of the content not selected by the user using a trained model that estimates an impression of input content, and derives a character string representing the estimated impression.

15. The information processing device according to claim 11, wherein the generating AI is updated by performing additional learning using content not selected by the user as a bad example.

16. a control means for controlling a GUI for the user; 2. The information processing apparatus according to claim 1, wherein the control means causes the negative prompt set by the setting means to be displayed on the GUI, and receives a selection by the user for the set negative prompt.

17. 2. The information processing device according to claim 1, further comprising a generation means for generating content by inputting positive prompts that specify what content the generation AI should generate and negative prompts set by the setting means to the generation AI.

18. The information processing device described in claim 1 further comprises a processing means for requesting an external information processing device having the generation AI to generate content together with a positive prompt that specifies what content the generation AI should generate and a negative prompt that is set by the setting means, and for receiving the content generated based on the request.

19. 2. The information processing apparatus according to claim 1, wherein the content is an image.

20. 2. The information processing apparatus according to claim 1, wherein the content is text.

21. An information processing method for obtaining a deliverable incorporating content generated by a generation AI, a derivation step in which a deriving means of the information processing device derives a specific character string that is previously associated with a template selected by the user from a plurality of templates prepared for the deliverable based on information obtained about the user; a setting step in which a setting means of the information processing device sets the specific character string derived in the derivation step as a negative prompt that specifies to the generation AI what kind of content should not be generated; An information processing method comprising:

22. A program for causing a computer to function as each of the means of the information processing device according to any one of claims 1 to 20.

Images