Procedural 3D Asset Generation Support System

Abstract

Provided is a procedural 3D asset generation support system that enables the creation of 3D assets with only text input, without the need for advanced expertise such as that of a 3D engineer. 【Solution means】The procedural 3D asset generation support system 10 selects a 3D asset template from the database 30 based on the text information input into the user prompt 20, generates a 3D asset using AI from the selected template and the generated parameters, and to modify the generated 3D asset, adds text information to the user prompt 20 so that AI modifies the parameters of the selected 3D asset template or AI selects a 3D asset template different from the selected 3D asset template, generates the parameters of that template, and modifies the 3D asset.

Description

Technical Field

[0001] The present invention relates to a procedural 3D asset generation support system that can efficiently produce 3D assets essential for creating computer graphics for users who want to create video games or movies by making full use of computer graphics.

Background Art

[0002] Methods for producing 3D assets, which are essential elements for creating computer graphics, include a method of manually producing 3D assets using 3D modeling software, a method of scanning an object or the like with a 3D scanner and using the data to produce 3D assets, and a method of producing 3D assets (procedural generation) using textures, shapes, etc. generated by an algorithm.

[0003] Among these, the method of producing 3D assets by procedural generation generates 3D assets by a computer program (specific algorithms and parameters), and is automatically generated based on algorithms and rules, rather than being produced by manual work of a 3D modeler. Therefore, it is more efficient than manual work and has the advantage that different variations can be easily generated by repeating the same procedure.

[0004] In this way, procedural generation is an efficient and powerful method in the production of 3D assets. However, appropriate algorithms and parameter settings are important. To achieve a realistic representation, it is necessary to pay attention to details (such as algorithms and parameters). The current templates (materials for generating 3D assets) are difficult to use without prior knowledge of parameters and the like, so high-level expertise is required for 3D engineers. Furthermore, there is a current situation where the initial development of templates responsible for procedural generation takes time. Based on such a situation, the applicants have conducted intensive research and development to overcome these drawbacks and have arrived at the present invention.

[0005] Patent Document 1 has an issue of "providing a computer-implemented method, program, medium, and system for designing a 3D modeling object representing a manufactured product (Patent Document 1: Summary)". The "method includes steps of obtaining a base mesh representing a 3D modeling object, selecting connection edges in the base mesh and obtaining a path of the connection edges composed of two end points, subdividing the base mesh based on the selected edges, and outputting the subdivided base mesh. The step of subdividing includes steps of obtaining a bevel pattern region extending over the selected path, obtaining a transition region by grouping all the faces sharing at least one of the two end points in the path except those of the calculated bevel pattern region, and remeshing the transition region (Patent Document 1: Summary)". The design of a 3D modeling object representing a manufactured product (Patent Document 1: Name of the Invention) is disclosed.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0007] The design of a 3D modeling object representing a manufactured product according to Patent Document 1 (Patent Document 1: Title of the Invention) requires steps such as obtaining a base mesh, selecting connection edges in the base mesh, obtaining a path of connection edges composed of two end points, subdividing the base mesh based on the selected edges, and outputting the subdivided base mesh. For each of these steps, the user has to perform operations (for creating a 3D model), so technical knowledge (experience) for performing operations for each step is required, which is time-consuming and not preferable in 3D model creation.

[0008] An object of the present invention is to provide a procedural 3D asset generation support system that can set appropriate algorithms and parameter settings and can create 3D assets only by text input, even without advanced expertise such as that of a 3D engineer (including preliminary knowledge regarding algorithms, parameters, etc.).

Means for Solving the Problem

[0009] In order to solve the above problems, the invention described in claim 1 is a procedural 3D asset generation support system that can efficiently generate 3D assets for a user who wants to create computer graphics, etc., a user prompt for the user to input text information, a database storing a large number of 3D asset templates, a template selection AI that selects a plurality of 3D asset templates from the database based on the text information input to the user prompt, a parameter generation AI that selects one from the plurality of 3D asset templates selected from the database based on the text information input to the user prompt and generates parameters for the selected one template, The 3D asset template selected by the parameter generation AI, and a generation system that generates a 3D asset from the generated parameters, In order to modify the 3D asset generated by the generation system, by additionally inputting text information into the user prompt, the parameter generation AI modifies the parameters of the 3D asset template selected by the parameter generation AI, and a procedural 3D asset generation support system that modifies the 3D asset with the (re - generated) parameters. Here, in this specification, the parameter is to adjust the setting of the scale (relative size), dimension (absolute size), position, etc. of the generated 3D asset.

[0010] The invention described in claim 2 is, in the invention described in claim 1, in order to regenerate a 3D asset different from the 3D asset generated by the generation system, by additionally inputting text information into the user prompt, a plurality of 3D asset templates in a combination different from the 3D asset template selected by the template selection AI are selected, and (again) the parameter generation AI selects one from the plurality of 3D asset templates, generates the parameters of the selected one 3D asset template, and regenerates a 3D asset with the selected one 3D asset template and the generated parameters. It is characterized by being the procedural 3D asset generation support system described in claim 1.

[0011] The invention described in claim 3 is, in the invention described in claim 1 or claim 2, a procedural 3D asset generation support system in which the parameter generation AI and the generation system generate a scene. Here, in this specification, a scene means an environment where a plurality of 3D assets (for example, a chair and a desk) exist.

[0012] The invention according to claim 4 is a procedural 3D asset generation support system which, in the invention according to claim 1 or claim 2, processes the source code and description text of the 3D asset template stored in the database and the user prompt as learning data, and uses the 3D asset generator LLM trained with the learning data to generate a 3D asset template that does not exist in the database.

[0013] The invention according to claim 5 is a procedural 3D asset generation support system which, in the invention according to claim 3, processes the source code and description text of the 3D asset template stored in the database and the user prompt as learning data, and uses the 3D asset generator LLM trained with the learning data to generate a 3D asset template that does not exist in the database.

Advantages of the Invention

[0014] With the procedural 3D asset generation support system according to the present invention, a user can access a database of 3D asset templates, and by further utilizing AI technology, the AI can select an optimal 3D asset template from a large amount of databases based on text information and set parameters of the template only by inputting text information, so that 3D assets can be generated.

[0015] There is no longer a need for the user to search for a specific (required) 3D asset template from a large number of 3D asset templates or deal with difficult parameters as in the prior art, and 3D assets can be generated only by inputting text information. That is, 3D assets can be generated only from text information. As a result, the 3D asset production cost can be significantly reduced.

Brief Description of the Drawings

[0016]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Modes for Carrying Out the Invention

[0017] <Overview of the Procedural 3D Asset Generation Support System> Hereinafter, an embodiment of a procedural 3D asset generation support system according to the present invention will be described in detail with reference to FIGS. 1 to 13. FIG. 1 is a diagram for explaining the outline of the procedural 3D asset generation support system.

[0018] The procedural 3D asset generation support system 10 according to the present invention can efficiently generate 3D assets for users (such as those without expertise like 3D engineers) who want to create computer graphics and the like. As shown in FIG. 1, the procedural 3D asset generation support system 10 includes a user prompt 20 for a user to input text information (input as "washbasin" in FIG. 1), a database 30 storing a large number of 3D asset templates (in FIG. 1, templates 1, 2, 3, ······, template N), a template selection AI 40 that selects a plurality of 3D asset templates from the database 30 based on the text information input to the user prompt 20 (selects templates 4, 7, 99 in FIG. 1), a parameter generation AI 50 that selects one template from the plurality of 3D asset templates selected by the template selection AI 40 based on the text information input to the user prompt 20 and generates parameters (selects template 4 and generates parameters for template 4 in FIG. 1), and a generation system 60 that generates a "3D asset (washbasin)" from the template selected by the parameter generation AI 50 and the generated parameters.

[0019] After the user inputs "washbasin" as text information to the user prompt 20 and checks the generated 3D asset, if it is necessary to correct it (note: there may also be cases of regenerating; whether to correct or regenerate is determined by the AI), that is, in order to correct the 3D asset generated by the generation system 60, by additionally inputting text information to the user prompt 20, the parameter generation AI 50 corrects the parameters of the (already) selected 3D asset template, and the 3D asset can be corrected.

[0020] On the other hand, when regenerating (judged by AI), from the database 30, the template selection AI 40 selects a plurality of 3D asset templates with a combination different from the already selected 3D asset template (in FIG. 1, select templates with different combinations not limited to templates 4, 7, 99). Then, the parameter generation AI 50 selects one from the plurality of 3D asset templates selected by the template selection AI 40, generates the parameters of that template, and can regenerate a new 3D asset different from the 3D asset generated by the generation system 60 from the template selected by the parameter generation AI 50 and the generated parameters.

[0021] FIG. 2 is a diagram showing an example of 3D asset generation by the procedural 3D asset generation support system 10.

[0022] When the user inputs text information "A torii made of stone" to the user prompt 20, from the text information input from the user prompt 20, AI (artificial intelligence) judges that "the object the user wants to generate is a torii made of stone", and from the 3D asset templates stored in the database 30, the template that AI judges to be optimal for generating the 3D asset related to the object the user wants to generate is selected, and the 3D asset related to the object the user wants to generate is generated (see FIG. 2).

[0023] If the user wants to modify the 3D asset, to modify the 3D asset generated by the generation system 60, the parameter generation AI 50 modifies the parameters of the already selected 3D asset template by adding input of text information such as, for example, "make the pillar a little thicker..." to the user prompt 20, or (depending on the judgment of the AI), the template selection AI 40 selects a plurality of 3D asset templates with a combination different from the 3D asset template already selected from the database 30, and the parameter generation AI 50 selects one from the plurality of 3D asset templates and generates the parameters of that template so that the 3D asset generated by the generation system 60 can be regenerated.

[0024] <Procedural 3D Asset Generation Pipeline> FIG. 3 is a flowchart diagram for explaining the "outline of the procedural 3D asset generation pipeline", which is a series of steps and processes for generating a 3D asset (forming the core) in the procedural 3D asset generation support system 10.

[0025] When the user inputs the "asset description" to be generated into the user prompt (user prompt 20), the procedural 3D asset template selection AI (template selection AI 40) selects a plurality of 3D asset templates based on the text information input to the user prompt.

[0026] If the AI (template selection AI 40) identifies a template related to the user prompt, the identified template is passed to the procedural 3D asset parameter generation AI (parameter generation AI 50).

[0027] On the one hand, if the AI (Template Selection AI 40) cannot identify a template related to the user prompt, the Procedural 3D Asset Generation Support System (Procedural 3D Asset Generation Support System 10) sends a message to the user to add detailed information about the asset it wants to generate. The user needs to provide detailed information about the asset it wants to generate. This flow will be repeated until the AI (Template Selection AI 40) identifies a template.

[0028] The AI (Parameter Generation AI 50) selects one template from the multiple templates identified by the Procedural 3D Asset Template Selection AI (Template Selection AI 40) and generates the parameters of that template.

[0029] If the AI (Parameter Generation AI 50) determines that it can generate one template and the parameters of that template, the Procedural 3D Asset Generation System (Generation System 60) uses the template and the generated parameters selected by the Procedural 3D Asset Parameter Generation AI (Parameter Generation AI 50) to generate a 3D asset. If the user determines that a satisfactory asset has been generated, the generated 3D asset is downloaded.

[0030] On the other hand, if the AI (Parameter Generation AI 50) determines that it cannot generate one template and the parameters of that template, the Procedural 3D Asset Generation Support System (Procedural 3D Asset Generation Support System 10) sends a message to the user to add detailed information about the asset it wants to generate. The user needs to provide detailed information about the asset it wants to generate. This flow will be repeated until the Generation System (Generation System 60) can use the template and the generated parameters selected by the AI (Parameter Generation AI 50) to generate a 3D asset.

[0031] When the AI (parameter generation AI 50) determines that it can generate one template and the parameters of that template, and even when the procedural 3D asset generation system (generation system 60) generates a 3D asset, if the user does not think that a satisfactory asset has been generated, the user can add detailed information about the asset to be generated. This process will also be repeated until the user thinks that a satisfactory asset has been generated.

[0032] Figures 4 to 6 are flowcharts (for explaining the details) of the procedural 3D asset generation pipeline. Figure 4 is for explaining the user prompt 20 and the template selection AI 40 (in detail), Figure 5 is for explaining the parameter generation AI 50 (in detail), and Figure 6 is a flowchart for explaining the details of the generation system 60.

[0033] As described in Figure 4, the user inputs a "description of the asset to be generated" into the user prompt 20. At the same time, the user selects whether to bake the texture and at what resolution.

[0034] The template selection AI 40 creates an embedding (a technique for converting high-dimensional data into low-dimensional real vectors) from the asset description text input by the user into the user prompt 20. It compares the embedding of the created asset description text with the embeddings of the 3D asset templates (stored in the database 30) and measures the similarity. As a result, it selects multiple templates that have a high similarity with the embedding of the created asset description text and fall within the context space of the parameter generation AI 50 (the length of the text considered when the generation model generates, and the size of the past information held by the model). Note that the 3D asset templates stored in the database 30 are equipped with a function to create embeddings of the description text, parameters, and materials for all (3D asset templates) as an automated task.

[0035] As shown in FIG. 5, the parameter generation AI 50 prepares a prompt including a pre-defined generation instruction, an asset description text input by the user, and an explanation of the template, parameter, and material selected by the template selection AI 40. After preparing the prompt, based on the asset description text, it selects one 3D asset template and one or more materials to be applied, and generates a message for the user and parameters to be used in the generation system 60.

[0036] When the AI can generate the parameters of the generation system 60, the parameter generation AI 50 packs (converts the model, algorithm, data, etc. into a form that can be easily used by bundling them together) the selected 3D asset template, the generated parameters, and the material library as a request to the generation system 60. When the AI cannot generate the parameters of the generation system 60, the parameter generation AI 50 prepares a message for the user.

[0037] In the generation system 60, as shown in FIG. 6, it reads the 3D asset template, sets the generation parameters, and generates the mesh and UV. If the bake material is valid, it bakes the material and packs the mesh into a downloadable file (such as GLB, USD, FBX, OBJ, etc.). If the bake material is not valid, it packs the mesh into a downloadable file (such as GLB, USD, FBX, OBJ, etc.) without baking the material. Then, the generation system 60 prepares a message for the user.

[0038] The user checks the message from the AI and the generated 3D asset. If a satisfactory asset is generated, the user downloads it and the 3D asset generation is completed. If the user is not satisfied, the user adds detailed information about the asset to be generated through the user prompt 20. Then, the template selection AI 40 identifies a plurality of templates related to the user prompt, and the parameter generation AI 50 selects one template and one or more materials to be applied based on the added asset description text, and generates a message to the user and the parameters of the generation system 60. Then, the process by the generation system 60 is performed again. This flow will be repeated until the user is satisfied (see FIGS. 4 to 6).

[0039] <AI-Assisted World Builder> FIGS. 7 to 9 are flowcharts for explaining the details of the AI-Assisted World Builder. FIG. 7 is for explaining the user prompt 20 and the template selection AI 40 in detail, FIG. 8 is for explaining the parameter generation AI 50 in detail, and FIG. 9 is a flowchart for explaining the details of the generation system 60. As described in FIG. 7, the user inputs a "description of the scene to be generated" into the user prompt 20. At the same time, the user selects whether to bake the texture and at what resolution. Note that a scene is an environment where a plurality of 3D assets exist.

[0040] The procedural 3D asset generation support system 10 enables a user to create an embedding (a technique for converting high-dimensional data into a low-dimensional real vector) from the scene description text input by the user into the user prompt 20. The embedding of the created scene description text is compared with the embeddings of the 3D asset templates (stored in the database 30), and the similarity is measured. As a result, one or more templates with high similarity to the embedding of the created scene description text and within the context space of the parameter generation AI 50 (the length of the text considered when the generation model generates, and the size of the past information held by the model) are selected. Note that the 3D asset templates stored in the database 30 have a function to create embeddings of the description text, parameters, and materials for all (3D asset templates) as an automated task.

[0041] As shown in FIG. 8, the parameter generation AI 50 prepares a prompt including a pre-defined generation instruction, the scene description text input by the user, and the description of the template, parameters, and materials selected by the template selection AI 40. After preparing the prompt, based on the asset description text, one or more 3D asset templates and one or more materials to be applied are selected, and a message to the user and parameters to be used in the generation system 60 of FIG. 1 are generated.

[0042] When the AI can generate the parameters of the generation system 60, the parameter generation AI 50 packs (converts the model, algorithm, data, etc. into a form that can be easily used by bundling them together) the selected 3D asset template, the generated parameters, and the material library as a request to the generation system 60. When the AI cannot generate the parameters of the generation system 60, the parameter generation AI 50 prepares a message to the user. (See FIG. 8: Parameter Generation AI 50)

[0043] In addition to the 3D assets generated above, if there are 3D asset templates that require processing (for example, if it is necessary to generate 3D assets of a chair in addition to the 3D assets of a desk that have already been generated), (the procedural 3D asset generation support system 10 is related to the chair) the 3D asset template is loaded, generation parameters are set, a mesh (vertices, edges, polygons) and UV (a coordinate system used when applying a texture, which is an image with color and texture information to a 3D model) are generated. If the bake material is valid, the material is baked (a process of calculating the effects of textures and light arranged on the surface of a 3D model and saving the results as a still image or texture), and the position of the asset is set. If the bake material is not valid, the position of the asset is set as it is. If there are no asset templates that require processing, the asset is packed into a file (such as GLB, USD, FBX, OBJ, etc.) that can be downloaded, and a message for the user is prepared (see Figure 9: generation system 60).

[0044] The user checks the message from the AI and the generated 3D scene. If a satisfactory 3D scene is generated, it is downloaded and completed. If not satisfied, detailed information about the 3D scene to be generated is added. Then, the template selection AI 40 identifies the template related to the user prompt 20, and the parameter generation AI 50 selects one or more templates and one or more materials to be applied based on the added scene description text, and generates a message for the user and the parameters of the generation system 60. Then, again, the process by the generation system 60 is performed. This process will be repeated until the user is satisfied.

[0045] <Method for generating 3D assets by a procedural 3D asset generation support system> The procedural 3D asset generation support system 10 is a method that can efficiently generate 3D models for users (without specialized knowledge such as 3D engineers) who want to create computer graphics and the like.

[0046] An input step in which a user inputs text information into the user prompt 20, a template selection AI 40 identifies a relevant template from the database 30 based on the text information input into the user prompt 20, a parameter generation AI 50 selects one or more materials to be applied to one or more templates based on the text information input into the user prompt 20, generates a message to the user and parameters of the generation system 60, and a generation AI step in which the generation system 60 generates a 3D asset based on the template and material selected by the parameter generation AI 50 in FIG. 1 and the generated parameters.

[0047] Furthermore, by adding text information to the user prompt 20, a step in which the parameter generation AI 50 modifies the parameters of the selected 3D asset template for the 3D asset generated in the generation AI step, or a step in which the template selection AI 40 selects a plurality of 3D asset templates in a combination different from the 3D asset template selected by the template selection AI 40 from the database 30, and the parameter generation AI 50 selects one or more from the plurality of 3D asset templates selected by the template selection AI 40, generates the parameters of the template, and regenerates (the 3D asset generated in the generation AI step) from the 3D asset template and the generated parameters selected by the parameter generation AI 50.

[0048] <Server Structure in the Procedural 3D Asset Generation Support System> Figure 10 is a diagram for explaining the server structure in the Procedural 3D Asset Generation Support System 10. As shown in Figure 10, the server structure in the Procedural 3D Asset Generation Support System 10 includes a distributed DB server 11 (which operates behind applications and web services and houses the backend responsible for data processing, business logic, database management, etc.), a distributed storage server 12 (which houses AI models), a distributed embedding storage server 13 (which houses template embeddings), a distributed DB server 14 (which houses material detailed information), a distributed DB server 15 (which houses template detailed information), a distributed storage server 16 (which houses material generation code), and a distributed storage server 17 (which houses template generation code).

[0049] Note that template detailed information refers to the description and usage method of the template, the description and usage method of the template parameters (such as scale, number of polygons, etc.), and the description and usage method of the available material slots of the template. Material detailed information refers to the description and usage method of the material parameters (such as scale, number of polygons, etc.).

[0050] The relevance between the above server structure and the process flow of the Procedural 3D Asset Generation Support System 10 (with descriptions of each element constituting the server structure in (*)) will be described with reference to Figure 10.

[0051] User (Client PC1, Client PC2, ·····, Client PCN: the Nth PC) → Load Balancer (a device that distributes the traffic of servers to multiple servers to equalize the load and improve the performance and reliability of the system) → Backend Server (Server 1, Server 2, ·····, Server N (the Nth server)): (*Distributed DB Server 11 (houses the backend))

[0052] (Continuing from 0050) → Load Balancer → Template Selection AI40 (GPU Server 1, GPU Server 2, ·····, GPU Server N (the Nth GPU Server)): (*Distributed Storage Server 12 (stores AI models): (*Distributed Embedding Storage Server 13 (stores template embeddings)

[0053] (Continuing from 0051) → Parameter Generation AI50 (GPU Server 1, GPU Server 2, ·····, GPU Server N (the Nth GPU Server)): (*Distributed DB Server 14 (stores material details): (*Distributed DB Server 15 (stores template details)

[0054] (Continuing from 0052) → Generation System 60 (GPU Server 1, GPU Server 2, ·····, GPU Server N (the Nth GPU Server)): (*Distributed Storage Server 16 (stores material generation code): (*Distributed Storage Server 17 (stores template generation code) has such a structure. The server structure in the Procedural 3D Asset Generation Support System 10 is characterized by the fact that the Template Selection AI40 and the Parameter Generation AI50 are stored (see Figure 10).

[0055] Figure 11 is a flowchart diagram of the Template Selection AI40. Multiple template selections by the Template Selection AI40 can be realized by the process of adding detailed descriptions to all templates. The detailed descriptions mentioned here refer to the description and usage method of the template, the description and usage method of the template parameters (for example, scale, number of polygons, etc.), and the description and usage method of the available material slots of the template.

[0056] The template selection AI 40 describes the template itself, the parameters and materials that can be used with the template, calculates the embedding (vector representation) of each template, and saves it in an accessible manner. As a result, it is possible to compare the saved embeddings with the embeddings of the text descriptions of the 3D assets provided by the user, and create a list of templates scored by relevance.

[0057] Figure 12 is a flowchart diagram of the parameter generation AI 50. One or more template selections and parameter generation by the parameter generation AI 50 can be realized by describing the parameters and materials that can be used with the template and providing this information to the parameter generation AI 50. The parameter generation AI 50 is set (programmed) to generate the necessary parameters for each template using this information.

[0058] Figure 13 is a flowchart diagram regarding the learning data processing of the 3D asset generator LLM. The 3D asset generator LLM saves the generation code of the procedural 3D asset template in a distributed storage. The generation code is converted into source code, and this source code and the detailed description of the template are used as learning data to train the model of the 3D asset generator LLM. The model learned by the procedural 3D asset template selection AI (template selection AI 40) is utilized to generate 3D asset templates that do not exist in the database.

[0059] <Effect of the Procedural 3D Asset Generation Support System> The procedural 3D asset generation support system 10 enables the efficient production of 3D assets (3D images / videos) used in digital content production such as game development, animation production, virtual reality (VR) and augmented reality (AR) applications, that is, used within a three-dimensional space.

[0060] Specifically, by allowing the user to access the database 30 and further utilize AI, the AI can select the optimal (at least one or more) templates from a large amount of databases based on text information with only text input, and 3D assets can be generated from the parameters of the selected templates and the generated templates.

[0061] Unlike the conventional method, there is no need for the user to search for specific (required) templates from a large number of templates or deal with complex parameters, and 3D models can be created with only text input. That is, 3D models can be created from only text information. As a result, the 3D asset production cost can be significantly reduced.

[0062] <Examples of Changes to the Procedural 3D Asset Generation Support System> The procedural 3D asset generation system according to the present invention is not limited to the aspects of each of the above-described embodiments, and the configurations such as the database, user prompt, template selection AI, parameter generation AI, generation system, etc. can be appropriately changed as necessary without departing from the spirit of the present invention.

Industrial Applicability

[0063] Since the procedural 3D asset generation support system according to the present invention has excellent effects as described above, it can be suitably used as a system for digital content production such as game development, animation production, virtual reality (VR) and augmented reality (AR) applications.

Explanation of Reference Numerals

[0064] 10 ··· 3D Asset Generation Support System 11 ··· Distributed DB Server (Backend) 12 ··· Distributed Storage Server (AI Model) 13·· Distributed Embedding Storage Server (Template Embedding) 14·· Distributed DB Server (Material Details) 15·· Distributed DB Server (Template Details) 16·· Distributed Storage Server (Material Generation Code) 17·· Distributed Storage Server (Template Generation Code) 20·· User Prompt 30·· Database 40·· Template Selection AI 50·· Parameter Generation AI 60·· Generation System

Claims

1. A procedural 3D asset generation support system capable of efficiently generating 3D assets for users who wish to create computer graphics, comprising: a user prompt where the user enters text information; A database that has accumulated a large number of 3D asset templates, A template selection AI that selects a plurality of 3D asset templates from the database based on text information entered in the user prompt; and A parameter generation AI that selects one of a plurality of 3D asset templates selected from the database based on text information input in the user prompt, and generates parameters for the selected one template; A generation system that generates a 3D asset from the 3D asset template selected by the parameter generation AI and the generated parameters; A procedural 3D asset generation support system characterized in that in order to modify the 3D asset generated by the generation system, additional text information is input into the user prompt, and the parameter generation AI modifies the parameters of the 3D asset template selected by the parameter generation AI, thereby modifying the 3D asset using the re-generated parameters.

2. In order to regenerate a 3D asset different from the 3D asset generated by the generation system, by inputting additional text information to the user prompt, a plurality of 3D asset templates having a combination different from the 3D asset template selected by the template selection AI are selected, and the parameter generation AI again selects one from the plurality of 3D asset templates, generates parameters for the selected one 3D asset template, and generates the selected one 3D asset template; The procedural 3D asset generation support system according to claim 1, further comprising: a step of regenerating a 3D asset based on the generated parameters.

3. A procedural 3D asset generation support system as described in claim 1 or claim 2, characterized in that the parameter generation AI and the generation system generate a scene.

4. The procedural 3D asset generation support system according to claim 1 or 2, characterized in that the source code and description of the 3D asset template stored in the database, and the user prompts are processed as learning data, and a 3D asset generator LLM trained with the learning data is used to generate a 3D asset template that does not exist in the database.

5. The procedural 3D asset generation support system according to claim 3, characterized in that the source code and description of the 3D asset template stored in the database, and the user prompts are processed as learning data, and a 3D asset generator LLM trained with the learning data is used to generate a 3D asset template that does not exist in the database.

Images