JP7329293B1 - Information processing device, method, program, and system - Google Patents

Abstract

An object of the present invention is to support the preparation of visual content suitable as an activity record of a character. A program according to one aspect of the present disclosure causes a computer to perform first model input based on profile information set for a specific character, profile information of a plurality of reference characters and each of the plurality of reference characters from an external source. Means for acquiring feature information representing features of visual content predicted to be suitable for a specific character by giving the relationship between the features of the visual content disclosed to the first model and acquiring feature information, By giving the second model input based on the second model to the second model that has learned to generate an image with the appearance of a specific character as the subject, the appearance of the specific character is used as the subject and the characteristics according to the second model input function as a means for acquiring candidate content, which is visual content with a function, and a means for outputting the candidate content. [Selection drawing] Fig. 5

Description

The present disclosure relates to an information processing device, method, program, and system.

Conventionally, there is known an agent interaction system that interacts with a user through text or voice. Japanese Patent Laid-Open No. 2002-200002 discloses a technical concept in which a character with individuality is designed to automatically reply with a text corresponding to the individuality.

Japanese Patent Application Laid-Open No. 2021-056668

According to the technical idea of Patent Document 1, by automatically replying with a text corresponding to the individuality of the character, there is a possibility that the sense of familiarity with the character and the reality of its existence can be enhanced.

On the other hand, celebrities and influencers in the real world, for example, post photos and videos of their off-time or on-time activities on SNS (Social Networking Service) to attract the interest of recipients. Similarly, with regard to fictitious characters, there is room for utilizing visual content such as photographs and moving images as a tool to increase the sense of familiarity that the recipient has for the character or the reality of the existence of the character. However, it is difficult to say that simply using the character as a subject is sufficient for visual content as a record of the character's activities. Ideally, it should be suitable from the viewpoint of For this reason, it is not easy to frequently and timely prepare visual contents suitable for character activity records. Even if the technical idea of Patent Document 1 is taken into account, it is not possible to automatically generate visual content suitable as a character's activity record.

An object of the present disclosure is to assist in the preparation of visual content suitable as a record of character activities.

A program according to one aspect of the present disclosure causes a computer to input a first model input based on profile information set for a specific character, profile information of a plurality of reference characters, and a visual image of each of the plurality of reference characters exposed to the outside. means for acquiring feature information representing features of visual content predicted to be suitable for a specific character by giving a relationship with features of the target content to the learned first model; a second model based on the feature information; By giving an input to a second model that has learned to generate an image with the appearance of a specific character as the subject, a vision that has the appearance of the specific character as the subject and has features according to the input of the second model. It functions as means for acquiring candidate content, which is target content, and means for outputting candidate content.

1 is a block diagram showing the configuration of an information processing system according to an embodiment; FIG. It is a block diagram showing the configuration of the client device of the present embodiment. It is a block diagram which shows the structure of the server of this embodiment. FIG. 4 is an explanatory diagram of an overview of learning of the first model of the embodiment; 1 is an explanatory diagram of one aspect of the present embodiment; FIG. It is a figure which shows the data structure of the profile table of this embodiment. 4 is a diagram showing the data structure of a posting log table according to the embodiment; FIG. 6 is a flowchart of post log collection processing according to the embodiment; 4 is a flowchart of visual content acquisition processing according to the embodiment;

An embodiment of the present invention will be described in detail below with reference to the drawings. In the drawings for describing the embodiments, in principle, the same constituent elements are denoted by the same reference numerals, and repeated description thereof will be omitted.

(1) Configuration of information processing system The configuration of the information processing system will be described. FIG. 1 is a block diagram showing the configuration of the information processing system of this embodiment.

As shown in FIG. 1 , the information processing system 1 includes a client device 10 and a server 30 .
The client device 10 and server 30 are connected via a network (for example, the Internet or an intranet) NW.
The server 30 and the external system 50 are connected via a network NW.

The client device 10 is an example of an information processing device that transmits requests to the server 30 . The client device 10 is, for example, a smart phone, a tablet terminal, or a personal computer. A user of the client device 10 is, for example, a person who has the authority to decide whether or not to publish visual content related to a specific character (hereinafter referred to as "administrative user"). The administrative user may be, for example, a person who possesses an NFT (Non-Fungible Token) that certifies the management authority of a specific character, and in this case the administrative user changes with the transfer of the NFT.

The server 30 is an example of an information processing device that provides the client device 10 with a response in response to a request transmitted from the client device 10 . Server 30 is, for example, a server computer.

(1-1) Configuration of Client Apparatus The configuration of the client apparatus will be described. FIG. 2 is a block diagram showing the configuration of the client device of this embodiment.

As shown in FIG. 2, the client device 10 includes a storage device 11, a processor 12, an input/output interface 13, and a communication interface . The client device 10 is connected to the display 21 .

Storage device 11 is configured to store programs and data. The storage device 11 is, for example, a combination of ROM (Read Only Memory), RAM (Random Access Memory), and storage (eg, flash memory or hard disk).

Programs include, for example, the following programs.
・OS (Operating System) program ・Application (for example, web browser) program that executes information processing

The data includes, for example, the following data.
・Databases referenced in information processing ・Data obtained by executing information processing (that is, execution results of information processing)

The processor 12 is a computer that implements the functions of the client device 10 by activating programs stored in the storage device 11 . Processor 12 is, for example, at least one of the following:
・CPU (Central Processing Unit)
・GPU (Graphic Processing Unit)
・ASIC (Application Specific Integrated Circuit)
・FPGA (Field Programmable Gate Array)

The input/output interface 13 acquires information (for example, user instructions) from an input device connected to the client device 10 and outputs information (for example, an image) to an output device connected to the client device 10. configured to

Input devices are, for example, keyboards, pointing devices, touch panels, or combinations thereof.
Output devices are, for example, the display 21, speakers, or a combination thereof.

Communication interface 14 is configured to control communications between client device 10 and an external device (eg, server 30).

The display 21 is configured to display images (still images or moving images). The display 21 is, for example, a liquid crystal display or an organic EL display.

(1-2) Server Configuration The server configuration will be described. FIG. 3 is a block diagram showing the configuration of the server of this embodiment.

As shown in FIG. 3, the server 30 includes a storage device 31, a processor 32, an input/output interface 33, and a communication interface .

Storage device 31 is configured to store programs and data. Storage device 31 is, for example, a combination of ROM, RAM, and storage (eg, flash memory or hard disk).

Programs include, for example, the following programs.
・OS program ・Application program that executes information processing

The data includes, for example, the following data.
・Databases referenced in information processing ・Execution results of information processing

The processor 32 is a computer that implements the functions of the server 30 by activating programs stored in the storage device 31 . Processor 32 is, for example, at least one of the following:
・CPU
・GPU
・ASICs
・FPGA

The input/output interface 33 is configured to acquire information (eg, user instructions) from an input device connected to the server 30 and output information (eg, an image) to an output device connected to the server 30. .

Input devices are, for example, keyboards, pointing devices, touch panels, or combinations thereof.
An output device is, for example, a display.

Communication interface 34 is configured to control communications between server 30 and an external device (eg, client device 10 or external system 50).

(2) One aspect of the embodiment One aspect of the present embodiment will be described. FIG. 4 is an explanatory diagram of the outline of the learning of the first model of this embodiment. FIG. 5 is an explanatory diagram of one aspect of the present embodiment.

As shown in FIG. 4, a contributor posts visual content (hereinafter referred to as "visual content of the reference character P1") such as a photograph or video of the appearance of the reference character P1 to the external system 50. do. The reference character P1 may be the same person as the contributor, or may be a fictitious character played by the contributor. Typically, the reference character P1 is a real influencer, celebrity, talent, model, idol, actor, singer, dancer, athlete, etc. It may be a fictitious character. Also, the number of reference characters P1 is an arbitrary number equal to or greater than 2, but it is preferable to secure a sufficient scale for learning the first model LM31, which will be described later.

The external system 50 provides services to users, including sharing visual content. The service provided by the external system is, for example, an SNS such as Instagram (registered trademark), but is not limited to this. Also, visual content sharing may be a side service rather than the primary service of the external system. The external system 50 accumulates the posted visual content of the reference character P1 and makes it available to the user.

The server 30 acquires profile information of the reference character P1. The profile information of the reference character P1 may be obtained from the external system 50, or may be created based on information that can be obtained about the reference character P1. The profile information of the reference character P1 may be created by a human (for example, an operator of the information processing system 1), or may be created by, for example, summarizing and shaping information that can be obtained about the reference character P1 into a trained model. may

Server 30 also obtains the visual content of reference character P1 (which may include metadata for the visual content) from external system 50 .

The server 30 uses the acquired information as learning data to create a first model LM31 that has learned the relationship between the profile information of the reference character P1 and the features of the visual content of the reference character P1. As a result of learning, the first model LM31 can infer, for given profile information, features of visual content that are appropriate for that profile information.

As shown in FIG. 5, the server 30 acquires profile information for a specific character C2. The profile information of the specific character C2 may be created by, for example, a human (for example, the administrative user M3 of the specific character C2), or may be created by a human (for example, the administrative user M3 of the specific character C2, A fan of the character C2, or a combination thereof) may be created by, for example, summarizing/shaping information specified by a learned model.

The server 30 provides the first model input based on the profile information of the specific character C2 to the first model LM31, thereby predicting that the character C2 is suitable for the character C2 (for example, the character C2 is suitable for disclosure). Acquire feature information representing features of the visual content.

Additionally, the server 30 uses the second model LM32 to obtain visual content. Here, the second model LM32 is trained in advance so as to generate an image with the appearance of the specific character C2 as the subject. The server 30 supplies the second model LM32 with a second model input based on the acquired feature information, thereby creating visual content having the appearance of the specific character C2 as a subject and having features corresponding to the feature information. (hereinafter referred to as “candidate content”). Note that the second model LM32 may simultaneously generate a plurality of candidate contents, not limited to one.

The server 30 outputs (transmits) the acquired candidate content to the client device 10 operated by the administrative user M3 of the specific character C2. The administrative user M3 checks whether the candidate content is suitable as visual content for the specific character C2, and instructs the client device 10 to approve or disapprove the disclosure. The client device 10 transmits to the server 30 approval or disapproval of disclosure of the candidate content. The server 30 publishes the candidate content approved for publication to the outside.

Publishing of visual content to the outside means that a third party (a person different from the administrative user M3, typically a fan of a specific character C2) can disclose the whole, part, or all of the visual content. It refers to making a processed version available for viewing, either conditionally or unconditionally. External exposure of visual content can be any of the following:
・Posting visual content to SNS (Social Networking Service) or content sharing services, etc. ・Selling visual content (which may include NFT of visual content) (which may include auction) Listing of visual content to services

(3) Database Tables constituting the database of this embodiment will be described. The following tables are stored in the storage device 31.

(3-1) Profile Table The profile table of this embodiment will be described. FIG. 6 is a diagram showing the data structure of the profile table of this embodiment.

Profile information is stored in the profile table. Profile information is information about the profile of the reference character.

As shown in FIG. 6, the profile table includes a "character ID" field, a "gender" field, an "occupation" field, an "age" field, and a "follower count" field. Each field is associated with each other.

A character ID is stored in the "character ID" field. A character ID is information for identifying a reference character corresponding to a corresponding record. As an example, the character ID may be the account name of the reference character in the service provided by the external system 50. FIG.

The "gender" field stores gender information. The sex information is information on the sex of the reference character corresponding to the record.

The "occupation" field stores occupation information. The occupation information is information on the occupation of the reference character corresponding to the record.

The "age" field stores age information. Age information is information about the age of the reference character corresponding to the record.

The "number of followers" field stores follower number information. The follower count information is, for example, information about the follower count of the account of the reference character corresponding to the record in the service provided by the external system.

Profile information is not limited to the above example. Profile information may not include some or all of the above examples, and may include the following information.
・Local information about the residential area of the reference character corresponding to the corresponding record ・Appearance information about the features of the reference character corresponding to the corresponding record (for example, race, body type, etc.) ・Clothing or clothing of the reference character corresponding to the corresponding record Preference information related to preferences such as makeup ・Personality information related to the personality of the reference character corresponding to the relevant record ・Life rhythm related to the reference character corresponding to the relevant record (for example, time for getting ready in the morning, bathing, eating, or working) information

In addition to reference characters, profile information of characters that can be candidates for creating candidate content (that is, can be subjects of candidate content) can be managed using the same or similar tables.

(3-2) Post Log Table The post log table of this embodiment will be described. FIG. 7 is a diagram showing the data structure of the posting log table of this embodiment.

The posting log table stores posting log information. Posted log information is information relating to the log of the visual content of the reference character posted to the service provided by the external system 50 .

As shown in FIG. 7, the posting log table includes a "posting date and time" field, a "character ID" field, and a "visual feature" field. Each field is associated with each other.

The “posted date and time” field stores posted date and time information. The posted date/time information is information regarding the posted date/time of the visual content corresponding to the record.

A character ID is stored in the "character ID" field. A character ID is information for identifying a reference character that is a subject of visual content corresponding to the record. The character ID is the same as or has a one-to-one relationship with the character ID in the profile table (FIG. 6).

The "visual feature" field stores visual feature information. The visual feature information is information representing the visual features of the visual content corresponding to the record. Visual feature information is typically text information. The server 30 can acquire such visual feature information using a trained model that outputs text representing the visual features of the visual content in response to model input based on the visual content. However, as the visual feature information, it is also possible to use the visual content itself or information indicating its location (for example, URL).

(4) Information processing Information processing according to the present embodiment will be described.

(4-1) Post Log Collection Processing Post log collection processing according to the present embodiment will be described. FIG. 8 is a flowchart of post log collection processing according to the present embodiment.

The post log collection process can be started, for example, when the following start conditions are satisfied.
- An instruction to start the post log collection process is given to the server 30 by, for example, the operator of the information processing system 1 .
- A predetermined period of time has passed since the previous post log collection process was performed.
・The specified date and time has arrived.
• The profile table (Fig. 6) has been updated (which may include inserting or deleting records).
- The table that manages the profile information of characters for which candidate content can be created has been updated (which may include inserting or deleting records).

As shown in FIG. 8, server 30 acquires posted content (S130).
Specifically, the server 30 acquires visual content posted to the external system 50 . Here, the acquired visual content may be limited to those posted within a predetermined period of time in the past. The predetermined period may be, for example, the period from the current time to the time when the previous post log collection process was performed.

As an example, the server 30 may acquire visual content using an API (Application Programming Interface) provided by the external system 50 or by scraping.

Server 30 may also obtain metadata for each visual content. Metadata can include at least one of the following:
- Posting time of visual content (may include posting date and time)
・Account name of the person who posted the visual content ・Location where the visual content was taken
・Characters tagged in visual content

After step S130, the server 30 identifies a reference character (S131).
Specifically, server 30 identifies a reference character corresponding to each visual content acquired in step S130 (that is, a reference character that is the subject of each visual content). As an example, the server 30 identifies the reference character corresponding to the visual content based on the account name of the contributor of the visual content. As a result, each visual content acquired in step S130 is associated with the corresponding profile information of the reference character.

After step S131, server 30 updates the posting log table (S132).
Specifically, server 30 saves the post log information corresponding to each visual content acquired in step S130 in the post log table (FIG. 7). As an example, the server 30 causes an existing image generation AI (Artificial Intelligence) service to read visual content, thereby obtaining text expressing visual features of the visual content. This text corresponds to a prompt (command sentence) for causing the image generation AI to generate visual content that is the same as or similar to the visual content from which the text is based. For example, Midjourney, Stable Diffusion, etc. are available as image generation AI services. Server 30 may also convert the obtained prompts into a format suitable for the second model. For conversion, for example, a large-scale language model such as ChatGPT can be used.

The server 30 associates the posting date and time information acquired in step S130, the visual feature information of the visual content (or the visual content itself) acquired in step S130, and the character ID of the reference character specified in step S131. Save the record in the post log table (Fig. 7).

After step S132, the server 30 performs learning of the first model (S133).
Specifically, server 30 uses learning data based on the updated content of the posted log table in step S132 (that is, the added posted log information) and the profile information of the reference character corresponding to each piece of posted log information. , perform machine learning of the first model. For example, the learning data includes explanatory variables based on the profile information of the reference character, and correct data based on the corresponding visual content features (for example, posting date/time information, visual feature information, or a combination thereof). As an example, the server 30 may create or update the first model by performing additional learning (eg fine tuning or transfer learning) on an existing trained model such as ChatGPT.

Note that the learning of the first model (S133) may be skipped if unnecessary. That is, the learning of the first model (S133) may be repeatedly performed at longer intervals than the post log collection process. Alternatively, learning of the first model (S133) may be performed at a timing independent of the post log collection process (for example, periodically or in response to a request from the operator of the information processing system 1).

(4-2) Visual Content Acquisition Processing Visual content acquisition processing according to this embodiment will be described. FIG. 9 is a flowchart of visual content acquisition processing according to this embodiment.

The visual content acquisition process can be started, for example, in response to the establishment of the following start conditions.
- An instruction to start the visual content acquisition process is given to the server 30 by, for example, the operator of the information processing system 1 or the managing user of a specific character.
- A predetermined period of time has passed since the previous visual content acquisition process was performed.
・The specified date and time has arrived.
• The profile table (Fig. 6) has been updated (which may include inserting or deleting records).
- The table that manages the profile information of characters for which candidate content can be created has been updated (which may include inserting or deleting records).
- New post log information is stored in the post log table (Fig. 7).

As shown in FIG. 9, the server 30 acquires profile information (S230).
Specifically, the server 30 acquires profile information of a character for which visual content is to be acquired (hereinafter referred to as "target character"). The target characters may be all or part of the characters for which candidate content can be created.

After step S230, the server 30 performs acquisition of desired visual content features (S231).
Specifically, the server 30 provides the first model with the first model input based on the profile information of each target character acquired in step S230, thereby providing the visual content predicted to be suitable for the target character ( That is, feature information representing features of the desired visual content is obtained. The feature information may be data in text format. As an example, the server 30 may list the profile information of the target character, and then set the first model input as a prompt requesting output of features of visual content that is suitable (likely to be posted) for the target character. can. Feature information may include, for example, prompts for causing the image generation AI to generate an image, and information about a suitable time or date for publication of the generated image.

After step S231, the server 30 executes model input generation (S232).
Specifically, the server 30 generates the second model input of the target character based on the feature information of each target character acquired in step S231. As an example, the server 30 generates the second model input by removing, from the feature information acquired in step S231, information relating to features that should not be reflected in the candidate content (for example, deleting from the prompt). Features that you do not want reflected in the candidate content can include, for example, at least one of the following.
・Static features of appearance ・Quasi-static features of appearance (when the first condition is satisfied)

Here, static characteristics of appearance refer to long-term Refers to features that do not vary over time (eg, over a month). Also, the quasi-static features of appearance refer to features such as make-up, accessories, or hairstyle, which may change in a short-term (for example, about one day) cycle, but which do not normally change within the same cycle. On the other hand, dynamic features of appearance include, for example, facial expressions, postures, gestures, and the like.

The first condition can include, as a necessary condition, that the scheduled release date and time of the candidate content belongs to a time period from after the target character gets ready in the morning to before taking a bath (hereinafter referred to as "activity time period"). Here, the scheduled release date and time is, for example, the time or date that matches the time or date included in the feature information. Also, the profile information may include the target character's morning preparation time and bathing time. That is, for example, when the scheduled release date and time is before the target character's morning preparation time (and after the wake-up time) or after the bathing time (and before the bedtime), the server 30 selects "nightwear", " Semi-static features such as "no makeup" need not be removed from the feature information. In addition, the first condition may further include, as a necessary condition, that the candidate content is the visual content that will be released second or later in the activity time zone of the day to which the scheduled release date belongs. In other words, if the candidate content is the visual content that is first published during the activity period of the day on which the scheduled publication date and time belongs, the server 30 does not need to remove the semi-static features from the feature information. As a result, it is possible to change the quasi-static features every day while suppressing fluctuations in the quasi-static features such as clothes and make-up within the activity time zone of the same day.

After step S232, the server 30 acquires candidate content (S233).
Specifically, the server 30 provides the second model input of each target character generated in step S232 to the second model, thereby obtaining the candidate content of the target character. The second model is created in advance by subjecting an image generation AI (for example, Stable Diffusion) to additional learning so as to generate an image with the appearance of the target character as a subject. For additional learning, for example, the following machine learning techniques can be used, but are not limited to these.
・Dreambooth
・Textual Inversion
・LoRA
・LoCon
・LoHA
・HyperNetworks
・Dream Artist
・Aesthetic Gradients

By providing the second model input, the second model generates and outputs one or more visual contents having the appearance of the target character as a subject and having characteristics according to the second model input. The server 30 acquires the output visual content as candidate content.

After step S233, the server 30 outputs the candidate content (S234).
Specifically, the server 30 outputs (transmits) the candidate content for each target character generated in step S233 to the client device 10 operated by the administrative user of the target character. The client device 10 displays the received candidate content on the display 21 and requests approval from the administrative user. The administrative user checks whether the quality of the candidate content is sufficient, whether the candidate content is suitable from the viewpoint of the target character's branding and the fan base, etc., and if it determines that it can be published, it instructs the client device 10 to approve it and publish it. If it is determined that it is not possible, it instructs the client device 10 to deny it. The client device 10 transmits the result of approval or disapproval of each candidate content to the server 30 .

If the visual content output in step S234 is approved by the administrative user, the server 30 publishes the candidate content (S235).
Specifically, the server 30 publishes the candidate content to the outside as the visual content of the target character. As an example, server 30 offers candidate content to a service that sells visual content or its NFT. As another example, server 30 shares the candidate content to a subscription-based or fee-based visual content sharing service. As yet another example, the server 30 posts candidate content to SNS.

On the other hand, if the visual content output in step S234 is denied by the administrative user, the server 30 terminates the candidate content acquisition process. Alternatively, the server 30 may re-execute the candidate content acquisition process for a target character for which none of the candidate contents has been denied. As a result, it is possible to repeatedly provide candidate contents until the managing user of the target character is satisfied.

(5) Summary As described above, the server 30 of the present embodiment receives the first model input based on the profile information set for a specific character by combining the profile information of a plurality of reference characters with the profile information of the plurality of reference characters. By providing the first model that has learned the relationship with the features of the visual content each of which has been released to the outside, feature information representing the features of the visual content predicted to be suitable for the specific character is obtained. The server 30 supplies the second model input based on the acquired feature information to the second model that has learned to generate an image with the appearance of the specific character as the subject, thereby taking the appearance of the specific character as the subject, and obtaining candidate content, which is visual content having characteristics according to the second model input. The server 30 outputs candidate content. As a result, it is possible to obtain candidate content that has the appearance of a specific character as a subject and that has features suitable for profile setting of the character, that is, candidate content that has reality as an activity record of the character.

The features of the visual content that each of the plurality of reference characters has exposed to the outside may include the visual content itself or text representing the visual features of the visual content. As a result, learning and inference of the first model can be performed appropriately.

The features of the visual content that each of the plurality of reference characters has released to the outside may further include information regarding at least one of the release time of the visual content and the response to the release of the visual content. As a result, it is possible to perform learning and inference using the release time of visual content as a feature, and to perform learning and inference with emphasis on visual content that has received a large response to release.

Characteristic information may include text representing visual characteristics of visual content predicted to be suitable for a particular character. The server 30 may generate the second model input based on the text representing the visual characteristics excluding the text representing the static characteristics of appearance. As a result, it is possible to prevent static features of the appearance of the character that do not change over a long period of time (for example, one month or longer) from being changed at an unnatural frequency.

Appearance static characteristics may include the size, shape or color of part or all of the body. As a result, it is possible to prevent the size, shape, or color of part or all of the character's body from being changed at an unnatural frequency in the candidate content.

Characteristic information may include text representing visual characteristics of visual content predicted to be suitable for a particular character. When the first condition is satisfied, the server 30 may generate the second model input based on the text representing the visual feature excluding the text representing the quasi-static feature of the appearance. As a result, it is possible to prevent the quasi-static features of the appearance of the character, which are features that can fluctuate in a cycle of a short period (for example, about one day) but do not normally fluctuate within the same cycle, from being changed at an unnatural frequency. can be done.

The first condition may include, as a necessary condition, that the scheduled release date and time of the candidate content belongs to a time period from after the specific character gets ready in the morning to before taking a bath. This makes it possible to suppress variations in semi-static appearance features from when the character gets ready in the morning to when he or she takes a bath.

The first condition is that the scheduled release date and time of the candidate content belongs to the time period from after the specific character gets ready in the morning to before bathing, and the candidate content is within the time zone on the day to which the scheduled release date and time belongs. It may include as a requirement that the visual content be published after the th. As a result, semi-static features of a character's appearance can be determined based on feature information in candidate content that is released first in a day, but the semi-static features are released later in the same day. It can also be maintained in possible candidate content.

Profile information set for a specific character and profile information for a plurality of reference characters may each include information on at least one of gender, occupation, age, residential area, appearance characteristics, or clothing or makeup preferences. . This makes it possible to obtain appropriate feature information from the first model.

The server 30 may output the candidate content to the administrative user associated with the specific character, and disclose the candidate content to the outside in response to approval of the candidate content from the administrative user. As a result, it is possible to prevent disclosure of candidate content judged by the administrative user to be of insufficient quality, or candidate content judged inappropriate from the standpoint of character branding, fan base, or the like.

The externally published visual content may include photos or videos posted on SNS (Social Networking Service) or content sharing services. This makes it possible to obtain appropriate feature information from the first model.

(6) Other Modifications The storage device 11 may be connected to the client device 10 via the network NW. The display 21 may be built into the client device 10 . Storage device 31 may be connected to server 30 via network NW.

Each step of the information processing described above can be executed by either the client device 10 or the server 30 .

In the above description, it is assumed that the visual content of the reference character is posted on an SNS or a content sharing service. However, the visual content of the reference character is not limited to posted content, and may include content that has been released to the outside in various forms (for example, content put up for auction).

The server 30 may reflect the decision-making content of an organization consisting of one or more members associated with the target character in the second model or the second model input. One or more members associated with the subject character may be, for example, holders of governance tokens or NFTs associated with the subject character, or members of the subject character's fan club. The organization may be a DAO (Decentralized Autonomous Organization), and in this case, the decision-making of the organization is determined by voting by governance token holders (however, each person's voting rights depend on the number of governance tokens held). done. The content of decision-making includes, for example, changes in the appearance features of the target character (typically quasi-static features, but may include other features), actions desired for the target character (things to eat, actions to take). location, etc.). For example, the server 30 may embed prompts in the second model input according to the content of the decision, or may perform additional learning of the second model according to the content of the decision. As a result, the appearance of the target character can be brought closer to the taste of the fan of the target character.

The server 30 may reflect, in the second model or the second model input, the member's comment or remark tendency within the community of members associated with the target character. One or more members associated with the subject character may be, for example, holders of governance tokens or NFTs associated with the subject character, or members of the subject character's fan club. A community may be, for example, a chat group. If the member's comments or remarks tend to be positive, the server 30 secondly prompts about features associated with a positive impression (for example, facial expressions, postures, gestures, makeup, accessories, hairstyles, activities, etc.). It may be embedded in the model input, or additional learning of the second model may be performed according to the feature. On the other hand, if the member's comments or remarks tend to be negative, the server 30 prompts about features (for example, facial expressions, postures, gestures, makeup, accessories, hairstyles, activities) that are associated with negative impressions. It may be embedded in the second model input, or additional learning of the second model may be performed according to the feature. As a result, candidate content is generated that expresses how the target character is pleased or depressed by the reaction of the fan, so that the fan can feel the existence of the target character more realistically.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited to the above embodiments. Also, the above embodiments can be modified and modified in various ways without departing from the gist of the present invention. Also, the above embodiments and modifications can be combined.

1 : Information processing system 10 : Client device 11 : Storage device 12 : Processor 13 : Input/output interface 14 : Communication interface 21 : Display 30 : Server 31 : Storage device 32 : Processor 33 : Input/output interface 34 : Communication interface 50 : External system

Claims (16)

the computer,
A first model input based on profile information set for a specific character is a first model that learns the relationship between profile information of a plurality of reference characters and features of visual content externally disclosed by each of the plurality of reference characters. means for obtaining feature information representing features of visual content predicted to be suitable for a particular character by feeding it to the model;
By providing a second model input based on the feature information to a second model that has learned to generate an image with the appearance of the specific character as the subject, the appearance of the specific character is used as the subject, and 2 means for obtaining candidate content, which is visual content with features according to the model input;
means for outputting said candidate content;
A program that functions as The feature of the visual content that each of the plurality of reference characters has disclosed to the outside includes the visual content itself or a text representing the visual feature of the visual content.
A program according to claim 1. The features of the visual content released to the outside by each of the plurality of reference characters further include information on at least one of the release time of the visual content and the reaction to the release of the visual content,
3. A program according to claim 2. The feature information includes text representing visual features of visual content predicted to be suitable for the specific character,
causing the computer to act as means for generating the second model input based on text representing visual characteristics excluding text representing static characteristics of appearance;
A program according to claim 1. said static features of appearance include size, shape or color of part or all of the body;
5. A program according to claim 4. The feature information includes text representing visual features of visual content predicted to be suitable for the specific character,
The computer functions as means for generating the second model input based on the text representing the visual feature excluding the text representing the quasi-static feature of appearance when a first condition is satisfied. ,
A program according to claim 1. The first condition includes, as a necessary condition, that the scheduled release date and time of the candidate content belongs to a time period from after the specific character gets ready in the morning to before taking a bath.
7. A program according to claim 6. The first condition is that the scheduled release date and time of the candidate content belongs to a time zone from after the specific character gets ready in the morning to before bathing, and the candidate content is within the time zone on the day to which the scheduled release date and time belongs. including as a requirement that it be the second or subsequent visual content published in
7. A program according to claim 6. The profile information set for the specific character and the profile information for the plurality of reference characters each include information on at least one of gender, occupation, age, residential area, appearance characteristics, or clothing or makeup preferences. ,
A program according to claim 1. the means for outputting the candidate content outputs the candidate content to an administrative user associated with the specific character;
causing the computer to further function as a means for externally publishing the candidate content in response to approval of the candidate content from the administrative user;
A program according to claim 1. further functioning the computer as a means for reflecting in the second model or the second model input the decision-making content of an organization consisting of one or more members associated with the specific character;
A program according to claim 1. causing the computer to further function as a means for reflecting, in the second model or the second model input, the tendency of comments or remarks of members in a community of members associated with the specific character;
A program according to claim 1. The externally published visual content includes photos or videos posted on SNS (Social Networking Service) or content sharing services,
A program according to claim 1. the computer
A first model input based on profile information set for a specific character is a first model that learns the relationship between profile information of a plurality of reference characters and features of visual content externally disclosed by each of the plurality of reference characters. obtaining feature information representing features of visual content predicted to be suitable for a particular character by feeding it to a model;
By providing a second model input based on the feature information to a second model that has learned to generate an image with the appearance of the specific character as the subject, the appearance of the specific character is used as the subject, and obtaining candidate content, which is visual content with features responsive to the two-model input;
and outputting said candidate content. A first model input based on profile information set for a specific character is a first model that learns the relationship between profile information of a plurality of reference characters and features of visual content externally disclosed by each of the plurality of reference characters. means for obtaining feature information representing features of visual content predicted to be suitable for a particular character by feeding it to a model;
By providing a second model input based on the feature information to a second model that has learned to generate an image with the appearance of the specific character as the subject, the appearance of the specific character is used as the subject, and means for obtaining candidate content, which is visual content with features responsive to two-model inputs;
and means for outputting the candidate content. A system comprising a first information processing device and a second information processing device,
The first information processing device is
A first model input based on profile information set for a specific character is a first model that learns the relationship between profile information of a plurality of reference characters and features of visual content externally disclosed by each of the plurality of reference characters. means for obtaining feature information representing features of visual content predicted to be suitable for a particular character by feeding it to a model;
By providing a second model input based on the feature information to a second model that has learned to generate an image with the appearance of the specific character as the subject, the appearance of the specific character is used as the subject, and means for obtaining candidate content, which is visual content with features responsive to two-model inputs;
means for outputting the candidate content to the second information processing device;
system.