One-to-many automatic content generation
Techniques are disclosed for automatically generating new content using a trained 1-to-N generative adversarial network (GAN) model. In disclosed techniques, a computer system receives, from a computing device, a request for newly-generated content, where the request includes current content. The computer system automatically generates, using the trained 1-to-N GAN model, N different versions of new content, where a given version of new content is automatically generated based on the current content and one of N different style codes, where the value of N is at least two. After generating the N different versions of new content, the computer system transmits them to the computing device. The disclosed techniques may advantageously automate a content generation process, thereby saving time and computing resources via execution of the 1-to-N GAN machine learning model.
1. A method, comprising:
receiving, by a computer system from a computing device, a request for newly-generated content, wherein the request includes current content;
performing, by the computer system on different sets of current content associated with a plurality of different entities, a clustering operation;
determining, by the computer system based on a plurality of orientation-style centroids generated during the clustering operation, a plurality of different style codes;
automatically generating, by the computer system using a trained 1-to-N generative adversarial network (GAN) model, N different versions of new content, wherein a given version of new content is automatically generated based on the current content and one of N different style codes selected from the plurality of different style codes, and wherein the value of N is at least two; and
transmitting, by the computer system to the computing device, the N different versions of new content.
2. The method of claim 1 , wherein each of the N different versions of the new content are automatically generated using a different one of the N different style codes.
3. The method of claim 1 , wherein the clustering operation is a k-means clustering operation.
4. The method of claim 3 , wherein the plurality of different style codes are determined based on centers of one or more clusters generated during performance of the k-means clustering operation, and wherein the plurality of different style codes correspond to a plurality of different UI design layouts.
5. The method of claim 3 , wherein the computer system obtains the different sets of current content associated with a plurality of different entities by scraping, from a plurality of websites, information specifying one or more existing UI designs, wherein the information includes program code and image renderings for respective ones of the one or more existing UI designs.
6. The method of claim 1 , wherein the current content includes information specifying a current user interface (UI) design, wherein the 1-to-N GAN model is a layout GAN model, and wherein generating the N different versions of the new content includes automatically generating multiple new layouts for a new UI design.
7. The method of claim 6 , further comprising:
generating, using a color GAN model, a set of colors for UI elements included in the new UI design, wherein determining the set of colors is performs by inputting at least one of the multiple new layouts for the new UI design into the color GAN model.
8. The method of claim 6 , further comprising:
generating, by the computer system using a residual neural network (ResNet), a set of properties for text elements included in the new UI design, wherein the generating includes inputting at least one of the new layouts for the new UI design and one or more text elements included in the new UI design into the ResNet.
9. The method of claim 6 , wherein automatically generating the multiple new layouts for the new UI design includes automatically generating multiple different layout images that include a plurality of portions with different pixel colors representing a plurality of elements included in the multiple new layouts for the new UI design.
10. A non-transitory computer-readable medium having instructions stored thereon that are executable by a computer system to perform operations comprising:
receiving, from a computing device, a request for newly-generated content, wherein the request includes current content;
performing, on different sets of current content associated with a plurality of different entities, a clustering operation;
determining, based on a plurality of orientation-style centroids generated during the clustering operation, a plurality of different style codes;
automatically generating, using a trained 1-to-N generative adversarial network (GAN) model, N different versions of new content, wherein a given version of new content is automatically generated based on the current content and one of N different style codes selected from the plurality of different style codes, and wherein the value of N is at least two; and
transmitting, to the computing device, the N different versions of new content.
11. The non-transitory computer-readable medium of claim 10 , wherein each of the N different versions of the new content are automatically generated using a different one of the N different style codes.
12. The non-transitory computer-readable medium of claim 10 , wherein the clustering operation is a k-means clustering operation.
13. The non-transitory computer-readable medium of claim 12 , wherein the plurality of different style codes are determined based on centers of one or more clusters generated during performance of the k-means clustering operation, and wherein the plurality of different style codes correspond to a plurality of different UI design layouts.
14. The non-transitory computer-readable medium of claim 12 , wherein the computer system obtains the different sets of current content associated with a plurality of different entities by scraping, from a plurality of websites, information specifying one or more existing UI designs, wherein the information includes program code and image renderings for respective ones of the one or more existing UI designs.
15. The non-transitory computer-readable medium of claim 10 , wherein the current content includes information specifying a current user interface (UI) design, wherein the 1-to-N GAN model is a layout GAN model, and wherein generating the N different versions of the new content includes automatically generating multiple new layouts for a new UI design.
16. The non-transitory computer-readable medium of claim 15 , wherein automatically generating the multiple new layouts for the new UI design includes automatically generating multiple different layout images that include a plurality of portions with different pixel colors representing a plurality of elements included in the multiple new layouts for the new UI design.
17. The non-transitory computer-readable medium of claim 16 , further comprising:
transforming, by a trained residual neural network (ResNet) executed by the computer system, respective ones of the multiple new layouts to text-based representations of the new layouts that specify, for respective layouts, coordinates of bounding regions of the plurality of elements included in the multiple different layout images, wherein the text-based representations are usable to generate program code executable to render respective ones of the multiple new layouts.
18. The non-transitory computer-readable medium of claim 17 , further comprising:
transmitting, by the computer system to the computing device, the program code for respective ones of the multiple new layouts that is executable to render the new UI design.
19. A system, comprising:
at least one processor; and
a memory having instructions stored thereon that are executable by the at least one processor to cause the system to:
receive, from a computing device, a request for newly-generated content, wherein the request includes current content;
perform, on different sets of current content associated with a plurality of different entities, a clustering operation;
determine, based on a plurality of orientation-style centroids generated during the clustering operation, a plurality of different style codes;
automatically generate, using a trained 1-to-N generative adversarial network (GAN) model, N different versions of new content, wherein a given version of new content is automatically generated based on the current content and one of N different style codes of the plurality of different style codes, and wherein the value of N is at least two; and
transmit, to the computing device, the N different versions of new content.
20. The system of claim 19 , wherein generating the N different versions of the new content includes automatically generating multiple new layouts for a new user interface (UI) design, and wherein the instructions are further executable to cause the system to:
generate, based on one of the multiple new layouts for the new UI design and a set of colors for the new UI design, information specifying the new UI design; and
cause, using the information specifying the new UI design, display of the new UI design.