Generating digital content
In implementations of systems for generating digital content, a computing device implements a generation system to receive a user input specifying a characteristic for digital content. The generation system generates input text based on the characteristic for processing by a first machine learning model. Output text generated by the first machine learning model based on processing the input text is received. The output text describes a digital content component. The generation system generates the digital content component by processing the output text using a second machine learning model. The generation system generates the digital content including the digital content component for display in a user interface based on the characteristic.
1 . A method comprising:
receiving, by a processing device, a user input specifying a characteristic for digital content; and
causing, by the processing device, generation of digital content automatically by a generative model based on the user input, the digital content including a plurality of digital content components, the causing including:
determining a layout for the digital content by comparing a vector representation of the user input in an embedding space with vector representations of candidate layouts for the digital content, the embeddings generated using a natural language processing model;
causing generation of input text based on the characteristic and types of digital content and a relative order specified by the layout for processing by a language model;
receiving data in a structured object format from the language model based on processing the input text, the data describing the digital plurality of digital content components in the relative order and including alternative text generated by the language model describing an additional one or more objects to be depicted in a digital image as a respective said component; and
causing generation of the digital content based on the characteristic by processing the data in the structured object format by the generative model, the digital content including the one or more objects as depicted in the digital image generated based on the alternative text as part of the plurality of digital content components having the types of digital content and compositing the plurality of digital content components in the relative order.
2 . The method as described in claim 1 , wherein the characteristic is a layout and the digital content component is ordered in the digital content relative to an additional digital content component based on the layout.
3 . The method as described in claim 1 , wherein the characteristic is an objective of the digital content and the digital content component includes natural language text having a sentiment based on the objective.
4 . The method as described in claim 1 , wherein the data includes the alternative text for the digital image that describes the object and the digital content component is the digital image that depicts the object.
5 . The method as described in claim 1 , wherein the language model includes a large language model.
6 . The method as described in claim 1 , wherein the layout for the digital content is identified using locality-sensitive hashing.
7 . The method as described in claim 1 , further comprising identifying a strategy for achieving an objective of the digital content by comparing the vector representation of the user input to vector representations of candidate strategies for achieving the objective of the digital content.
8 . The method as described in claim 1 , wherein the structured object format is JavaScript Object Notation that defines the relative order of the plurality of digital content components.
9 . The method as described in claim 8 , wherein the input text includes a request for the data to be formatted using the JavaScript Object Notation.
10 . A system comprising:
a memory component; and
a processing device coupled to the memory component, the processing device to perform operations including generating digital content including a plurality of digital content components, automatically, based on a user input specifying an objective for digital content, the generating including:
determining a layout for the digital content based on the user input, the layout defines types of digital content components to be included in the digital content and a relative order of the types of digital content components, the determining including comparing a vector representation of the user input in an embedding space with vector representations of candidate layouts for the digital content, the embeddings generated using a natural language processing model;
generating input text for processing by a language model, the input text including indications of the types of digital content components in the relative order based on the layout;
receiving data in a structured object format generated by the language model based on processing the input text, the data including descriptions of the types of digital content components in the relative order and including alternative text generated by the language model describing an additional one or more objects to be depicted in a digital image as a respective said component; and
causing generation of the digital content by processing the data in the structured object format by a generative model, the digital content including the one or more objects as depicted in the digital image generated based on the alternative text as composited as part of the types of digital content components in the relative order.
11 . The system as described in claim 10 , wherein the layout for the digital content is determined by comparing a vector representation of the user input with vector representations of candidate layouts for the digital content.
12 . The system as described in claim 11 , wherein the layout for the digital content is determined using locality-sensitive hashing.
13 . The system as described in claim 10 , wherein the structured object format is JavaScript Object Notation that encodes the relative order.
14 . The system as described in claim 13 , wherein the input text includes a request for the data to be formatted using the JavaScript Object Notation.
15 . A non-transitory computer-readable storage medium storing executable instructions, which when executed by a processing device, cause the processing device to perform operations comprising:
receiving a user input specifying a characteristic for digital content; and
generating digital content including a plurality of digital content components, automatically, based on the user input, the generating including:
causing generation of input text based on the characteristic for processing by a language model, the input text including a request to structure data in a structured object format;
receiving the data in a structured object format from the language model based on processing the input text, the data describing a digital content component and including alternative text generated by the language model describing an additional one or more objects to be depicted in a digital image; and
causing generation of the digital content as having the characteristic including the plurality of digital content components by processing the data using the generative model, the digital content including the one or more objects as depicted in the digital Image generated based on the alternative text by the generative model as composited as part of the plurality of digital content components.
16 . The non-transitory computer-readable storage medium as described in claim 15 , wherein the characteristic is an objective of the digital content and the digital content component includes natural language text having a sentiment based on the objective.
17 . The non-transitory computer-readable storage medium as described in claim 15 , wherein the characteristic is a layout and the digital content component is ordered in the digital content relative to an additional digital content component based on the layout.
18 . The non-transitory computer-readable storage medium as described in claim 15 , wherein the data includes the alternative text for the digital image that describes the object and the digital content component is the digital image that depicts the object.
19 . The non-transitory computer-readable storage medium as described in claim 15 , wherein the structured object format is JavaScript Object Notation.
20 . The non-transitory computer-readable storage medium as described in claim 15 , the operations further comprising identifying a strategy for achieving an objective of the digital content by comparing a vector representation of the user input to vector representations of candidate strategies for achieving the objective.