DETAILED ACTION
Continued Examination under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 03/18/2026 has been entered.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Examiner’s Comment
This Action is in response to the Request for Continued Examination filed on 03/18/2026 with Amended Claims and Applicant's Remarks filed on 03/10/2026.
Applicant has amended claims 1, 6, 8, 13, 15, and 20, canceled claims 3, 10, and 17, and newly added claims 24-26 according to Amendments filed on 03/10/2026. Claims 1, 2, 4-9, 11-16, and 18-26 are pending and currently under consideration for patentability.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1, 2, 4-9, 11-16, and 18-26 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claims are directed to a judicial exception (i.e., a law of nature, natural phenomenon, or abstract idea) without significantly more.
Step 1: In a test for patent subject matter eligibility, claims 1, 2, 4-9, 11-16, and 18-26 are found to be in accordance with Step 1 (see 2019 Revised Patent Subject Matter Eligibility), as they are related to a process, machine, manufacture, or composition of matter. Claims 1, 2, 4-7, 21-26 recite a method, claims 8, 9, 11-14 recite a system, and claims 15, 16, 17-20 recite a non-transitory computer-readable medium. When assessed under Step 2A, Prong I, they are found to be directed towards an abstract idea. The rationale for this finding is explained below:
Step 2A, Prong I: Under Step 2A, Prong I, claims 1, 2, 4-9, 11-16, and 18-26 are directed to an abstract idea without significantly more, as they all recite a judicial exception. Independent Claims 1, 8, and 15 recite limitations directed to the abstract idea including “receiving existing digital content associated with a campaign creator; identifying a context associated with the existing digital content; clustering, based on visual similarities, the existing digital content; receiving a natural language prompt trained to generate content from the natural language prompt; determining style embeddings associated with the clustered existing digital content, wherein the style embeddings quantify a style of the existing digital content via a numeric representation; determining, based on the style embeddings, one or more values of a style vector for the campaign creator, wherein the style vector corresponds to a weight or scale factor; and processing the natural language prompt, the context, and the one or more values of the style vector as inputs to generate digital content, wherein the generated digital content is based on the existing digital content.” These further limitations are not seen as any more than the judicial exception. Independent Claims 1, 8, and 15 recite additional limitations including “by one or more processors; for a generative model trained; and through the generative model.” Serving generated digital content based on natural language processing and style vectors is considered to be an abstract idea, specifically, certain methods of organizing human activity; such as managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions) because the claims are directed to managing a relationship between parties in order to provide information (i.e. generating content to be served). Furthermore, serving generated digital content based on natural language processing and style vectors is also considered to be fall under another grouping of abstract idea, specifically, Mental Processes; such as concepts performed in the human mind (including an observation, evaluation, judgment, opinion) because the claims are directed to receiving information (i.e. existing digital content), identifying information (i.e. a context associated with existing digital content), clustering information (i.e. existing digital content based on visual similarities), receiving information (i.e. natural language prompt), determining information (i.e. style embeddings and style vector), and processing information (i.e. natural language prompt) in order to generate content which can all be performed in the human mind. Therefore, under Step 2A, Prong I, Claims 1, 8, and 15 are directed towards an abstract idea.
Step 2A, Prong II: Step 2A, Prong II is to determine whether any claim recites any additional element that integrate the judicial exception (abstract idea) into a practical application. Independent Claims 1, 8, and 15 recite additional limitations including “by one or more processors; for a generative model trained; and through the generative model.” The additional limitations including “by one or more processors; for a generative model trained; and through the generative model” are not found to integrate the judicial exception into a practical application because they are seen as adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) and generally linking the use of the judicial exception to a particular technological environment or field of use – see MPEP 2106.05(h). Furthermore, Examiner would like to note that according to ¶ [0059] of the Applicant’s specification; “In general, any AI model technique for generating digital content from a text prompt may be used to implement the generative model 110.” Accordingly, alone, and in combination, these additional elements are seen as using a computer or tool to perform an abstract idea, adding insignificant-extra-solution activity to the judicial exception. They do no more than link the judicial exception to a particular technological environment or field of use, i.e. processors and generative model, and therefore do not integrate the abstract idea into a practical application. The courts decided that although the additional elements did limit the use of the abstract idea, the court explained that this type of limitation merely confines the use of the abstract idea to a particular technological environment and this fails to add an inventive concept to the claims (See Affinity Labs of Texas v. DirecTV, LLC,). Under Step 2A, Prong II, these claims remain directed towards an abstract idea.
Step 2B: Independent Claims 1, 8, and 15 recite additional limitations including “by one or more processors; for a generative model trained; and through the generative model.” The additional limitations, reciting – “by one or more processors; for a generative model trained; and through the generative model”, do not integrate the judicial exception (abstract idea) into a practical application because of the analysis provided in Step 2A, Prong II. Independent claims 1, 8, and 15 do not include additional elements or a combination of elements that result in the claims amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements listed amount to no more than mere instructions to apply an exception using a generic computer component. In addition, the applicant’s specifications describe generic computer-based elements, ¶ [0148], for implementing “general purpose microprocessors”, which do not amount to significantly more than the abstract idea of itself, which is not enough to transform an abstract idea into eligible subject matter. Furthermore, there is no improvement in the functioning of the computer or technological field, and there is no transformation of subject matter into a different state. Under Step 2B in a test for patent subject matter eligibility, these claims are not patent eligible.
Dependent claims 2, 4-7, 9, 11-14, and 16, 18-26 further recite independent claims 1, 8, and 15. Dependent claims 2, 4-7, 9, 11-14, and 16, 18-26 when analyzed as a whole are held to be patent ineligible under 35 U.S.C. 101 because the additional recited limitation fail to establish that the claims are not directed to an abstract idea:
Under Step 2A, Prong I, these additional claims only further narrow the abstract idea set forth in Independent Claims 1, 8, and 15. For example, claims 2, 4-7, 9, 11-14, and 16, 18-26 describe the limitations for serving generated digital content based on natural language processing and style vectors – which is only further narrowing the scope of the abstract idea recited in the independent claims.
Under Step 2A, Prong II, for dependent claims 2, 4-7, 9, 11-14, and 16, 18-26, there are no additional elements introduced. For example, dependent claim 21 recites – “wherein the generative model is trained based on at least the style embeddings associated with the existing digital content.” Dependent claim 22 recites – “receiving, by the one or more processors, feedback associated with the generated digital content; and updating, by the one or more processors, based on the feedback, the generative model.” Dependent claim 23 recites – “providing, by the one or more processors, as input into a model trained to generate embeddings, the existing digital content; and determining, by the one or more processors executing the model, the style embeddings.” Dependent claim 25 recites – “the one or more processors are of a digital content generation system comprising a plurality of modules.” These additional limitations are not found to integrate the judicial exception into a practical application because they are seen as adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) and generally linking the use of the judicial exception to a particular technological environment or field of use – see MPEP 2106.05(h). Furthermore, Examiner would like to note that according to ¶ [0059] of the Applicant’s specification; “In general, any AI model technique for generating digital content from a text prompt may be used to implement the generative model 110.” Accordingly, alone, and in combination, these additional elements are seen as using a computer or tool to perform an abstract idea, adding insignificant-extra-solution activity to the judicial exception. They do no more than link the judicial exception to a particular technological environment or field of use, i.e. generative model and/or modules, and therefore do not integrate the abstract idea into a practical application. The courts decided that although the additional elements did limit the use of the abstract idea, the court explained that this type of limitation merely confines the use of the abstract idea to a particular technological environment and this fails to add an inventive concept to the claims (See Affinity Labs of Texas v. DirecTV, LLC,).
Under Step 2B, the dependent claims 2, 4-7, 9, 11-14, and 16, 18-26 do not include any additional elements that are sufficient to amount to significantly more than the judicial exception. Additionally, there is no improvement in the functioning of the computer or technological field, and there is no transformation of subject matter into a different state. As discussed above with respect to integration of the abstract idea into a practical application, the additional claims do not provide any additional elements that would amount to significantly more than the judicial exception. Under Step 2B, these claims are not patent eligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1, 2, 4-9, 11-16, and 18-26 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Publication 2024/0362427 to Gupta in view of U.S. Publication 2022/0377424 to Deng.
Claims 1, 2, 4-7, and 21-26; 8, 9, and 11-14; and 15, 16, and 18-20 are method, system, and computer-readable medium claims, respectively, with substantially indistinguishable features between each group. For purposes of compact prosecution, the Office has grouped the common method, system and non-transitory computer readable storage medium claims in applying applicable prior art.
With respect to Claim 1:
Gupta teaches:
A method for serving digital content, comprising: receiving, by one or more processors, existing digital content […] (i.e. receiving a corpus of digital content) (Gupta: ¶ [0047] “In a first example, the candidate layouts are selected from a corpus of layouts of digital content based on an association with instances of digital content which achieved corresponding objectives for the instances of digital content ( e.g., resulted in highly attended events, received greater than a threshold number of user interactions, etc.). In the first example, the candidate strategies for achieving objectives of digital content are selected from a corpus of strategies based on an association with instances of digital content which achieved corresponding objectives. In a second example, the candidate layouts described by the content data 118 include each layout included in the corpus of layouts of digital content.”);
identifying, by the one or more processors, a context associated with the existing digital content (i.e. identifying an objective, theme, or context with the corpus of digital content) (Gupta: ¶ [0032] “For example, the generation module 110 receives and processes the input data 114 to generate a vector representation of the characteristic 116. In some examples, the generation module 110 includes or has access to a machine learning model trained on training data to generate vector representations of natural language statements, and the generation module 110 implements the machine learning model to generate the vector representation of the characteristic 116. In other examples, the generation module 110 generates the vector representation of the characteristic 116 using a hash function such as a locality-sensitive hash function. For example, the generation module 110 leverages the vector representation of the characteristic 116 to identify candidate layouts and/or candidate strategies for achieving the objective of the digital content that is to be generated.” Furthermore, as cited in ¶ [0047] “In a first example, the candidate layouts are selected from a corpus of layouts of digital content based on an association with instances of digital content which achieved corresponding objectives for the instances of digital content ( e.g., resulted in highly attended events, received greater than a threshold number of user interactions, etc.). In the first example, the candidate strategies for achieving objectives of digital content are selected from a corpus of strategies based on an association with instances of digital content which achieved corresponding objectives. In a second example, the candidate layouts described by the content data 118 include each layout included in the corpus of layouts of digital content.”);
clustering, by the one or more processors based on visual similarities, the existing digital content (i.e. clustering or identifying a set of candidate layouts based on visual similarities such as ‘hot air balloon’ style or similar classifications of ‘hot air balloon images’) (Gupta: ¶ [0034] “Consider an example in which the generation module 110 processes the input data 114 and the content data 118 to compare the vector representation of the characteristic 116 with the vector representations of candidate layouts using locality-sensitive hashing. In this example, the generation module 110 identifies a particular layout or a set of particular candidate layouts based on a similarity between the characteristic 116 and the particular layout or the particular candidate layouts.” Furthermore, as cited in ¶¶ [0039] [0040] “For example, the generated digital content components are visually similar to the types of content components that are digital images (e.g., depicting similar colors or themes), semantically similar to the types of content components that are digital images (e.g., depicting objects with semantically similar classifications), etc. In one example, the generation module 110 composites the generated digital content components (e.g., based on the relative order) as generated digital content 120 which is displayed in a user interface 122 of the display device 106…As shown, the generated digital content 120 is an invitation to a hot air balloon festival in Albuquerque which corresponds to the characteristic 116 described by the input data 114. For instance, the generated digital content 120 includes a generated digital content component that is a digital image such as a hero image depicting a hot air balloon 124. The generated digital content 120 also includes a generated digital content component which is natural language text 126 that states "Soar High Above the Enchantment of Albuquerque!"”);
receiving, by the one or more processors, a natural language prompt for a generative model trained to generate content from the natural language prompt (i.e. receiving natural language input for a machine learning model trained to generate content from input/prompt) (Gupta: ¶ [0042] “Consider an example in which a user interacts with an input device ( e.g., a mouse, a keyboard, a microphone, a stylus, a touchscreen, etc.) to modify the input data 114 by modifying the characteristic 116 and replacing "hot air balloon festival" with "farmer's market." In this example, the generation module 110 receives the modified input data 114 and replaces the generated digital content 120 with an invitation to a farmer's market in Albuquerque which includes a digital image depicting fresh fruit and vegetables (e.g., instead of the hot air balloon 124). By receiving the modified input data 114 with different natural language describing digital content to be generated, the generation module 110 is capable of generating digital content such as the generated digital content 120 that maintains a specified theme and/or brand requirements.”);
determining, by the one or more processors, style embeddings associated with the clustered existing digital content, […] (i.e. determining types of digital components for vector representations associated with candidate layouts based on similarly grouped corpus of digital content) (Gupta: ¶¶ [0039] [0040] “For example, the generated digital content components are visually similar to the types of content components that are digital images (e.g., depicting similar colors or themes), semantically similar to the types of content components that are digital images (e.g., depicting objects with semantically similar classifications), etc. In one example, the generation module 110 composites the generated digital content components (e.g., based on the relative order) as generated digital content 120 which is displayed in a user interface 122 of the display device 106…As shown, the generated digital content 120 is an invitation to a hot air balloon festival in Albuquerque which corresponds to the characteristic 116 described by the input data 114. For instance, the generated digital content 120 includes a generated digital content component that is a digital image such as a hero image depicting a hot air balloon 124. The generated digital content 120 also includes a generated digital content component which is natural language text 126 that states "Soar High Above the Enchantment of Albuquerque!"” Furthermore, as cited in ¶¶ [0046] [0047] “To do so in one example, the hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118. For instance, the hashing module 202 also compares the vector representation of the objective 302 with the vector representations of the candidate strategies for achieving objectives of digital content described by the content data 118. Although the hashing module 202 is illustrated as receiving the input data 114 as describing the objective 302 for digital content to be generated in the representation, it is to be appreciated that the hashing module 202 is also capable of receiving the input data 114 as describing a specific audience to receive the digital content to be generated, a generic audience to receive the digital content to be generated, a natural language prompt describing the digital content to be generated, an offer associated with the digital content to be generated, and so forth…In a first example, the candidate layouts are selected from a corpus of layouts of digital content based on an association with instances of digital content which achieved corresponding objectives for the instances of digital content ( e.g., resulted in highly attended events, received greater than a threshold number of user interactions, etc.). In the first example, the candidate strategies for achieving objectives of digital content are selected from a corpus of strategies based on an association with instances of digital content which achieved corresponding objectives. In a second example, the candidate layouts described by the content data 118 include each layout included in the corpus of layouts of digital content.”);
determining, by the one or more processors based on the style embeddings, one or more values of a style vector for the campaign creator, wherein the style vector corresponds to a weight or scale factor (i.e. determining vector representations of content features for candidate layouts, wherein the content features describe the style of the content and wherein vector representation includes a term frequency-inverse which is a value that corresponds to a weight or scale factor) (Gupta: ¶ [0045] “In some examples, the hashing module 202 generates the vector representation of the objective 302 using a natural language processing model trained on training data to generate vector representations of natural language text. In other examples, the hashing module 202 generates the vector representation of the objective 302 using a locality-sensitive hash function ( or hash functions). For example, the hashing module 202 generates the vector representation of the objective 302 using a data-independent technique for representing natural language text in a latent space or an embedding space. In one example, the hashing module 202 generates the vector representation of the objective 302 using a term frequency-inverse document frequency technique. In another example, the hashing module 202 generates the vector representation of the objective 302 using natural language processing feature extracting techniques or a bag of words representation.” Furthermore, as cited in ¶ [0049] “The hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118 using locality-sensitive hashing to identify a top N candidate layouts. In one example, the hashing module 202 identifies a particular layout for the digital content to be generated based on the locality-sensitive hashing. In this example, the particular layout is defined in a digital template 306. For instance, the hashing module 202 generates the match data 210 as describing "Incentives" from the candidate strategies 304 and the digital template 306. Although the hashing module 202 is illustrated as generating the match data 210 describing the candidate strategies 304 and the digital template 306, it is to be appreciated that the hashing module 202 is also capable of generating the match data 210 as describing other content features for digital content such as tonality, colors (e.g., color pallets), image features, etc.”); and
processing, by the one or more processors, the natural language prompt, the context, and the one or more values of the style vector as inputs through the generative model to generate digital content, wherein the generated digital content is based on the existing digital content (i.e. processing the natural language prompt, candidate layout to generate digital content, and values of style vector as match data which is fed into the machine learning model as input data in order to generate content, wherein the generated content is based on the corpus of digital content and wherein vector representation includes a term frequency-inverse which is a value that corresponds to a weight or scale factor) (Gupta: ¶¶ [0049]-[0051] “The hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118 using locality-sensitive hashing to identify a top N candidate layouts. In one example, the hashing module 202 identifies a particular layout for the digital content to be generated based on the locality-sensitive hashing. In this example, the particular layout is defined in a digital template 306. For instance, the hashing module 202 generates the match data 210 as describing "Incentives" from the candidate strategies 304 and the digital template 306. Although the hashing module 202 is illustrated as generating the match data 210 describing the candidate strategies 304 and the digital template 306, it is to be appreciated that the hashing module 202 is also capable of generating the match data 210 as describing other content features for digital content such as tonality, colors (e.g., color pallets), image features, etc…The prompt module 204 receives and processes the match data 210 in order to generate prompt data 212. FIG. 4 illustrates a representation 400 of generating input text. As shown, the representation 400 includes the digital template 306 described by the match data 210, and the prompt module 204 parameterizes the particular layout with placeholders 402 which will be replaced by digital content components generated using the second machine learning model. The prompt module 204 classifies the placeholders 402 into types of digital content components such as digital images, slogans ( e.g., lines of text with less than 10 words rendered using a font having a weight greater than 700), paragraphs of text, headings/headers, footers, call-to-action buttons, etc. The prompt module 204 also classifies content blocks of the particular layout which include digital images and/or text…After classifying the placeholders 402 as types of digital content components, the prompt module 204 includes the types of digital content components in input text 404 to be processed by the first machine learning model. For types of the digital content components that are digital images, the input text 404 requests alternative text describing the digital images. The input text 404 includes portions of the objective 302 such as natural language text stating "Invite registrations for kite flying festival." As shown, the input text 404 also includes a discount of"15% off" based on the particular strategy for achieving the objective 302 of "Incentives." The prompt module 204 generates the input text 404 as including a request for the first machine learning model to generate output text formatted using JavaScript Object Notation. The JavaScript Object Notation encodes a relative order of the types of digital content components defined by the particular layout of the digital template 306.” Furthermore, as cited in ¶¶ [0055] [0056] “The display module 208 receives the text data 214 describing the output text 502. For example, the display module 208 includes or has access to the second machine learning model which includes a generative machine learning model. Examples of generative machine learning models included in the second machine learning model include a model trained on training data to generate digital images, a diffusion model, a Generative Pre-Trained Transformer 4 model (GPT-4), a Hierarchical Text-Conditional Image Generation with CLIP Latents model (DALLE 2), etc. In some examples, the second machine learning model includes systems of generative machine learning models…In an example, the display module 208 implements the second machine learning module to generate digital content components 602-608 by processing the output text 502. For instance, the display module 208 generates digital content component 602 using the second machine learning model based on the alternative text 504 of "hot air balloon." As shown in the representation 600, the digital content component 602 is a digital image that depicts a hot air balloon.”).
Gupta does not explicitly disclose receiving, by one or more processors, existing digital content associated with a campaign creator; and wherein the style embeddings quantify a style of the existing digital content via a numeric representation.
However, Deng further discloses:
receiving, by one or more processors, existing digital content associated with a campaign creator (i.e. receive advertisements from advertisers) (Deng: ¶ [0048] “For example, advertisers may run online campaigns based on outdated information such as expired trends wasting marketing spending, and being unable to adjust the campaign appropriately. Users could receive inappropriate ads during situations resulting in bad experiences and potential damage to the advertised brand, product, and/or service.” Furthermore, as cited in ¶ [0062] “At 306, the processor 102 may employ AI bots to identify existing creatives or generate new creatives that match the information received from the fashion and geolocation data feeds from an advertisement platform hosting the dynamic digital content data store 120 (which in this example would store a collection of advertisements).”);
wherein the style embeddings quantify a style of the existing digital content via a numeric representation (i.e. style embeddings are quantified via numeric representation or values) (Deng: ¶ [0060] “FIG. 2 shows a flowchart that details a method 250 of providing dynamic digital content according to an example. The method 250 may be executed by the processor 102 by accessing the computer-readable instructions 105 stored on the memory 104. The method begins at 252 wherein the processor 102 receives the plurality of data feeds 152, 154, ... etc. At 254, the processor 102 determines the values of the parameterized variables that are to be employed in generating and transmitting the dynamic digital content. One or more variables may be determined at 254 for each data feed. For example, the weather data feed may be used to determine temperature, pressure, humidity, and other variables. The local information data feed may be used to populate variables such as the traffic at a geolocation, the various stores available, the operating hours of the stores, etc. For example, from posts or update feeds, the latest color, fabric and style trend data may be collected and aggregated for a particular geo area/demographic cohort. In this fashion example, color, fabric, style are parameterized variables with values populated from trending fashion data gleaned from tweeter data feeds.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Deng’s receiving, by one or more processors, existing digital content associated with a campaign creator; and wherein the style embeddings quantify a style of the existing digital content via a numeric representation to Gupta’s processing, by the one or more processors, the natural language prompt, the context, and the one or more values of the style vector as inputs through the generative model to generate digital content, wherein the generated digital content is based on the existing digital content. One of ordinary skill in the art would have been motivated to do so because “With the proliferation of different types of digital content delivery mechanisms (e.g., mobile phone, portable computing devices, tablet devices, etc.), it has become crucial that content providers, such as vendors looking to advertise a good or service, engage users with content of interest. As a result, content providers are continuously looking for ways to deliver more appealing content.” (Deng: ¶ [0137]).
With respect to Claims 8 and 15:
All limitations as recited have been analyzed and rejected to claim 1. Claim 8 recites “A system for serving digital content, comprising: one or more processors, the one or more processors configured to:” (Gupta: ¶ [0072]) perform the steps of method claim 1. Claim 15 recites “One or more non-transitory computer-readable storage media encoding instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:” (Gupta: ¶ [0073]) the steps of method claim 1. Claims 8 and 15 do not teach or define any new limitations beyond claim 1. Therefore they are rejected under the same rationale.
With respect to Claim 2:
Gupta does not explicitly disclose the method of claim 1, wherein when determining the style vector, the method further comprises determining, by the one or more processors based on the style embeddings, an average of a difference between the style embeddings, wherein the average of the difference between the style embeddings corresponds to the style vector.
However, Deng further discloses determining, by the one or more processors based on the style embeddings, an average of a difference between the style embeddings, wherein the average of the difference between the style embeddings corresponds to the style vector (i.e. determining a distance or average of a difference between vector embeddings, wherein the distance between vector embeddings corresponds to parameterized variables or style of content) (Deng: ¶ [0061] “Therefore, minimizing the vector distance between embedding (A) and embedding (B), effectively displays the most likely matched product (B) to a given user (A). While this on-the-fly product and user match is done very fast, the underlying embeddings are learned by the artificial intelligence (AI)/Deep Learning (DL) algorithms based on logged data about this user and product over the time, i.e. accumulated learned knowledge represented as embeddings, used to calculate distances in a vector space composed of these parameterized variables.” Furthermore, as cited in ¶ [0056] “In an example wherein the dynamic digital content includes online advertisements, the parameterized variables may include campaign parameters/assets such as campaign spending, campaign types e.g., auction, brand, etc., campaign reach and frequency, Geo and time distributions, targeting audience and segments, situational creatives, campaign size and structure, and optimization goals such as traffic, leads, conversions, purchase, store visits, etc.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Deng’s determining, by the one or more processors based on the style embeddings, an average of a difference between the style embeddings, wherein the average of the difference between the style embeddings corresponds to the style vector to Gupta’s processing, by the one or more processors, the natural language prompt, the context, and the one or more values of the style vector as inputs through the generative model to generate digital content, wherein the generated digital content is based on the existing digital content. One of ordinary skill in the art would have been motivated to do so because “With the proliferation of different types of digital content delivery mechanisms (e.g., mobile phone, portable computing devices, tablet devices, etc.), it has become crucial that content providers, such as vendors looking to advertise a good or service, engage users with content of interest. As a result, content providers are continuously looking for ways to deliver more appealing content.” (Deng: ¶ [0137]).
With respect to Claims 9 and 16:
All limitations as recited have been analyzed and rejected to claim 2. Claims 9 and 16 do not teach or define any new limitations beyond claim 2. Therefore they are rejected under the same rationale.
With respect to Claim 4:
Gupta teaches:
The method of claim 1, further comprising: generating, by the one or more processors based on the natural language prompt, seed information (i.e. generating output text or seed information based on input or prompt) (Gupta: ¶ [0036] “In an example, the input text includes a request for the first machine learning model to generate output text based on the input text such that the output text is formatted in JavaScript Object Notation. In examples in which the types of digital content components include digital images, the generation module 110 generates the input text as including requests for alternative text for the digital images. In these examples, the alternative text describes objects depicted in the digital images, visual themes of the digital images, objectives/purposes of the digital images, and so forth.”); and
determining, by the one or more processors based on the seed information, a theme (i.e. determining visual themes based on output text) (Gupta: ¶ [0036] “In an example, the input text includes a request for the first machine learning model to generate output text based on the input text such that the output text is formatted in JavaScript Object Notation. In examples in which the types of digital content components include digital images, the generation module 110 generates the input text as including requests for alternative text for the digital images. In these examples, the alternative text describes objects depicted in the digital images, visual themes of the digital images, objectives/purposes of the digital images, and so forth.”).
Gupta does not explicitly disclose identifying, by the one or more processors based on the theme, embeddings within a threshold distance of the theme.
However, Deng further discloses identifying, by the one or more processors based on the theme, embeddings within a threshold distance of the theme (i.e. determining a distance or average of a difference between vector embeddings based on parametrized variables or theme or rules) (Deng: ¶ [0061] “When the parameterized variable values are determined, the processor 102 determines at 256 if one or more of predefined rules and/or triggers to be applied based on the parameterized variable values may be identified. If no rules/triggers could be identified at 256, the method terminates on the end block. If one or more rules are identified at 256, the digital content may be dynamically generated or selected from the digital content data source 120 by the processor 102 at 258 based on the triggers/rules. In an example, Artificial Intelligence (AI) bots may be used to select the rules to be applied which may affect one or more of the dynamic digital content generation and the audience group selection. The dynamic digital content is transmitted at 260 by the processor 102 to selected ones of the user communication devices 192, 194, ... , etc. For example, artificial intelligence (AI) algorithms generates embeddings, or condensed knowledge, about an entity, such as a product, content, user, etc. When a user A comes to visit a store, the vast number of products available in the store are ranked, and only those products, say product B with their embeddings closest to the visiting user's embedding are filtered. Therefore, minimizing the vector distance between embedding (A) and embedding (B), effectively displays the most likely matched product (B) to a given user (A). While this on-the-fly product and user match is done very fast, the underlying embeddings are learned by the artificial intelligence (AI)/Deep Learning (DL) algorithms based on logged data about this user and product over the time, i.e. accumulated learned knowledge represented as embeddings, used to calculate distances in a vector space composed of these parameterized variables.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Deng’s identifying, by the one or more processors based on the theme, embeddings within a threshold distance of the theme to Gupta’s generating, by the one or more processors based on the natural language prompt, seed information; and determining, by the one or more processors based on the seed information, a theme. One of ordinary skill in the art would have been motivated to do so because “With the proliferation of different types of digital content delivery mechanisms (e.g., mobile phone, portable computing devices, tablet devices, etc.), it has become crucial that content providers, such as vendors looking to advertise a good or service, engage users with content of interest. As a result, content providers are continuously looking for ways to deliver more appealing content.” (Deng: ¶ [0137]).
With respect to Claims 11 and 18:
All limitations as recited have been analyzed and rejected to claim 4. Claims 11 and 18 do not teach or define any new limitations beyond claim 4. Therefore they are rejected under the same rationale.
With respect to Claim 5:
Gupta teaches:
The method of claim 4, wherein the embeddings are processed through the generative model to generate the digital content (i.e. vector embeddings of the digital content are processed through machine learning model in order to generate content) (Gupta: ¶¶ [0045] [0046] “For example, the hashing module 202 generates the vector representation of the objective 302 using a data-independent technique for representing natural language text in a latent space or an embedding space. In one example, the hashing module 202 generates the vector representation of the objective 302 using a term frequency-inverse document frequency technique. In another example, the hashing module 202 generates the vector representation of the objective 302 using natural language processing feature extracting techniques or a bag of words representation…Consider an example in which the hashing module 202 includes a machine learning predictor model trained on training data to receive inputs and generate content characteristics and/or content components based on the inputs such as communication strategies for digital content, layouts for digital content, etc. To do so in one example, the hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118. For instance, the hashing module 202 also compares the vector representation of the objective 302 with the vector representations of the candidate strategies for achieving objectives of digital content described by the content data 118.”).
With respect to Claims 12 and 19:
All limitations as recited have been analyzed and rejected to claim 5. Claims 12 and 19 do not teach or define any new limitations beyond claim 5. Therefore they are rejected under the same rationale.
With respect to Claim 6:
Gupta teaches:
The method of claim 1, wherein when processing the natural language prompt, the context, and the one or more values of the style vector through the generative model to generate the digital content, the generative model is configured to: modify the existing digital content based on the natural language prompt and the style vector (i.e. modifying ‘hot air balloon festival’ to ‘farmer’s market’ based on natural language input and vector representations) (Gupta: ¶ [0042] “Consider an example in which a user interacts with an input device ( e.g., a mouse, a keyboard, a microphone, a stylus, a touchscreen, etc.) to modify the input data 114 by modifying the characteristic 116 and replacing "hot air balloon festival" with "farmer's market." In this example, the generation module 110 receives the modified input data 114 and replaces the generated digital content 120 with an invitation to a farmer's market in Albuquerque which includes a digital image depicting fresh fruit and vegetables (e.g., instead of the hot air balloon 124).”); or
generate new digital content based on the natural language prompt and the style vector (i.e. digital content of ‘farmer’s marker’ is generated based on natural language prompt and vector representations) (Gupta: ¶ [0042] “In this example, the generation module 110 receives the modified input data 114 and replaces the generated digital content 120 with an invitation to a farmer's market in Albuquerque which includes a digital image depicting fresh fruit and vegetables (e.g., instead of the hot air balloon 124). By receiving the modified input data 114 with different natural language describing digital content to be generated, the generation module 110 is capable of generating digital content such as the generated digital content 120 that maintains a specified theme and/or brand requirements.”).
With respect to Claims 13 and 20:
All limitations as recited have been analyzed and rejected to claim 6. Claims 13 and 20 do not teach or define any new limitations beyond claim 6. Therefore they are rejected under the same rationale.
With respect to Claim 7:
Gupta teaches:
The method of claim 1, wherein the natural language prompt comprises changes to be made to the existing digital content […] (i.e. prompt or input includes changes/modifications to digital content and (Gupta: ¶ [0042] “Consider an example in which a user interacts with an input device ( e.g., a mouse, a keyboard, a microphone, a stylus, a touchscreen, etc.) to modify the input data 114 by modifying the characteristic 116 and replacing "hot air balloon festival" with "farmer's market." In this example, the generation module 110 receives the modified input data 114 and replaces the generated digital content 120 with an invitation to a farmer's market in Albuquerque which includes a digital image depicting fresh fruit and vegetables (e.g., instead of the hot air balloon 124). By receiving the modified input data 114 with different natural language describing digital content to be generated, the generation module 110 is capable of generating digital content such as the generated digital content 120 that maintains a specified theme and/or brand requirements.”).
Gupta does not explicitly disclose […] group information associated with a campaign.
However, Deng further discloses […] group information associated with a campaign (i.e. changes/updates or trends made to user segments with respect to advertisements) (Deng: ¶¶ [0061] [0062] “ In an example, Artificial Intelligence (AI) bots may be used to select the rules to be applied which may affect one or more of the dynamic digital content generation and the audience group selection. The dynamic digital content is transmitted at 260 by the processor 102 to selected ones of the user communication devices 192, 194, ... , etc. For example, artificial intelligence (AI) algorithms generates embeddings, or condensed knowledge, about an entity, such as a product, content, user, etc. When a user A comes to visit a store, the vast number of products available in the store are ranked, and only those products, say product B with their embeddings closest to the visiting user's embedding are filtered… The processor 102 may receive fashion trend information at 302 from a third-party provider who provides fashion trend data via a data feed that provides constant fashion updates. These updates may include information regarding the trends in various categories such as men/women/kids clothing categories or accessory categories, footwear categories, etc. Also, the processor 102 may receive at 304, the geo-location feed associated with the fashion trend information so that the specific location at which a particular style is currently trending may be determined. At 306, the processor 102 may employ AI bots to identify existing creatives or generate new creatives that match the information received from the fashion and geolocation data feeds from an advertisement platform hosting the dynamic digital content data store 120 (which in this example would store a collection of advertisements). The audience to receive the advertisements are identified at 308 based, for example, on not only the geo-location data but also based on personal data such as user preferences, etc.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Deng’s group information associated with a campaign to Gupta’s natural language prompt comprises changes to be made to the existing digital content. One of ordinary skill in the art would have been motivated to do so because “With the proliferation of different types of digital content delivery mechanisms (e.g., mobile phone, portable computing devices, tablet devices, etc.), it has become crucial that content providers, such as vendors looking to advertise a good or service, engage users with content of interest. As a result, content providers are continuously looking for ways to deliver more appealing content.” (Deng: ¶ [0137]).
With respect to Claim 14:
All limitations as recited have been analyzed and rejected to claim 7. Claim 14 does not teach or define any new limitations beyond claim 17. Therefore it is rejected under the same rationale.
With respect to Claim 21:
Gupta teaches:
The method of claim 1, wherein the generative model is trained based on at least the style embeddings associated with the existing digital content (i.e. machine learning model is trained based on vector representations of candidate layouts) (Gupta: ¶ [0046] “Consider an example in which the hashing module 202 includes a machine learning predictor model trained on training data to receive inputs and generate content characteristics and/or content components based on the inputs such as communication strategies for digital content, layouts for digital content, etc. To do so in one example, the hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118.”).
With respect to Claim 22:
Gupta does not explicitly disclose the method of claim 1, further comprising: receiving, by the one or more processors, feedback associated with the generated digital content; and updating, by the one or more processors, based on the feedback, the generative model.
However, Deng further discloses:
receiving, by the one or more processors, feedback associated with the generated digital content (i.e. receive feedback associated with user’s response to digital content) (Deng: ¶¶ [0105] “The user (who may be an opt-out user) may receive a digital content item displayed on a user device wherein the digital content item enables the user to provide feedback related to the digital content item. The feedback provided by the user may include at least exclusion data and action-taken data. In either case, it may signify that the user does not wish to receive the digital content item again. In addition, the user may be enabled to provide feedback with different options. A feedback option may allow the user to convey that the user would like to block/exclude the specific digital content item received by the user or digital content items that are similar to the specific digital content item. Another feedback may pertain to whether the user would like to exclude digital content items from the same content source. Furthermore, the feedback data may further include reasons for the user's exclusions e.g., action-taken data. In an example, the user may indicate that the user has already taken action in response to the received digital content item. For example, in response to an advertisement for a car, the user may provide information regarding the user's recent car purchase so that the user no longer receives digital content items such as ads related to cars. The user's feedback may be recorded in an exclusions list stored in a corresponding user profile.”); and
updating, by the one or more processors, based on the feedback, the generative model (i.e. updating or training the machine learning model using the feedback information) (Deng: ¶ [0106] “The feedback data from the user may be employed to train machine learning (ML) models for digital content filtering so that the users' exclusions may constitute negative label data while digital content items related to action-taken feedback may be used as positive label data for training the machine learning (ML) models. Therefore, the systems and methods according to examples herein avoid over-targeting users with customized content. Additionally, processes to provide incentives may be instituted to encourage users to provide feedback to the digital content items so that relevant digital content items are served to users while complying with the low entropy constraints imposed by various entities in communication networks.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Deng’s receiving, by the one or more processors, feedback associated with the generated digital content; and updating, by the one or more processors, based on the feedback, the generative model to Gupta’s processing, by the one or more processors, the natural language prompt, the context, and the one or more values of the style vector as inputs through the generative model to generate digital content, wherein the generated digital content is based on the existing digital content. One of ordinary skill in the art would have been motivated to do so because “With the proliferation of different types of digital content delivery mechanisms (e.g., mobile phone, portable computing devices, tablet devices, etc.), it has become crucial that content providers, such as vendors looking to advertise a good or service, engage users with content of interest. As a result, content providers are continuously looking for ways to deliver more appealing content.” (Deng: ¶ [0137]).
With respect to Claim 23:
Gupta teaches:
The method of claim 1, wherein when determining the style embeddings, the method further comprises: providing, by the one or more processors, as input into a model trained to generate embeddings, the existing digital content (i.e. machine learning model is provided with input of content characteristics/layouts to generate vector representations or embeddings) (Gupta: ¶ [0046] “Consider an example in which the hashing module 202 includes a machine learning predictor model trained on training data to receive inputs and generate content characteristics and/or content components based on the inputs such as communication strategies for digital content, layouts for digital content, etc. To do so in one example, the hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118. For instance, the hashing module 202 also compares the vector representation of the objective 302 with the vector representations of the candidate strategies for achieving objectives of digital content described by the content data 118. Although the hashing module 202 is illustrated as receiving the input data 114 as describing the objective 302 for digital content to be generated in the representation, it is to be appreciated that the hashing module 202 is also capable of receiving the input data 114 as describing a specific audience to receive the digital content to be generated, a generic audience to receive the digital content to be generated, a natural language prompt describing the digital content to be generated, an offer associated with the digital content to be generated, and so forth.”).; and
determining, by the one or more processors executing the model, the style embeddings (i.e. determining vector representations or embeddings of candidate layouts) (Gupta: ¶ [0046] “Consider an example in which the hashing module 202 includes a machine learning predictor model trained on training data to receive inputs and generate content characteristics and/or content components based on the inputs such as communication strategies for digital content, layouts for digital content, etc. To do so in one example, the hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118. For instance, the hashing module 202 also compares the vector representation of the objective 302 with the vector representations of the candidate strategies for achieving objectives of digital content described by the content data 118. Although the hashing module 202 is illustrated as receiving the input data 114 as describing the objective 302 for digital content to be generated in the representation, it is to be appreciated that the hashing module 202 is also capable of receiving the input data 114 as describing a specific audience to receive the digital content to be generated, a generic audience to receive the digital content to be generated, a natural language prompt describing the digital content to be generated, an offer associated with the digital content to be generated, and so forth.”).
With respect to Claim 24:
Gupta teaches:
The method of claim 1, wherein the one or more values of the style vector provides control over a style of the generated digital content to drive consistency of the generative model (i.e. values or term frequency-inverse of vector representations provide control of the style of the digital content to be generated) (Gupta: ¶¶ [0045] [0046] “In some examples, the hashing module 202 generates the vector representation of the objective 302 using a natural language processing model trained on training data to generate vector representations of natural language text. In other examples, the hashing module 202 generates the vector representation of the objective 302 using a locality-sensitive hash function ( or hash functions). For example, the hashing module 202 generates the vector representation of the objective 302 using a data-independent technique for representing natural language text in a latent space or an embedding space. In one example, the hashing module 202 generates the vector representation of the objective 302 using a term frequency-inverse document frequency technique. In another example, the hashing module 202 generates the vector representation of the objective 302 using natural language processing feature extracting techniques or a bag of words representation…Consider an example in which the hashing module 202 includes a machine learning predictor model trained on training data to receive inputs and generate content characteristics and/or content components based on the inputs such as communication strategies for digital content, layouts for digital content, etc. To do so in one example, the hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118. For instance, the hashing module 202 also compares the vector representation of the objective 302 with the vector representations of the candidate strategies for achieving objectives of digital content described by the content data 118. Although the hashing module 202 is illustrated as receiving the input data 114 as describing the objective 302 for digital content to be generated in the representation, it is to be appreciated that the hashing module 202 is also capable of receiving the input data 114 as describing a specific audience to receive the digital content to be generated, a generic audience to receive the digital content to be generated, a natural language prompt describing the digital content to be generated, an offer associated with the digital content to be generated, and so forth.”).
With respect to Claim 25:
Gupta teaches:
The method of claim 1, wherein: the one or more processors are of a digital content generation system comprising a plurality of modules (Gupta: ¶¶ [0075] [0076] “Various techniques are described herein in the general context of software, hardware elements, or program modules. Generally, such modules include routines, programs, objects, elements, components, data structures, and so forth that perform particular tasks or implement particular abstract data types. The terms "module," "functionality," and "component" as used herein generally represent software, firmware, hardware, or a combination thereof. The features of the techniques described herein are platform-independent, meaning that the techniques are implementable on a variety of commercial computing platforms having a variety of processors… Implementations of the described modules and techniques are storable on or transmitted across some form of computer-readable media. For example, the computer-readable media includes a variety of media that is accessible to the computing device 1002. By way of example, and not limitation, computer-readable media includes "computer-readable storage media" and "computer-readable signal media."”).
Gupta does not explicitly disclose the method further comprises implementing, by one or more processors, between each module, a filter configured to remove at least one of data or inputs that have been identified as causing hallucinations to occur.
However, Deng further discloses the method further comprises implementing, by one or more processors, between each module, a filter configured to remove at least one of data or inputs that have been identified as causing hallucinations to occur (i.e. filtering out inputs that have been identified as similar to excluded items via collaborative filtering) (Deng: ¶ [0106] “The feedback data from the user may be employed to train machine learning (ML) models for digital content filtering so that the users' exclusions may constitute negative label data while digital content items related to action-taken feedback may be used as positive label data for training the machine learning (ML) models. Therefore, the systems and methods according to examples herein avoid over-targeting users with customized content. Additionally, processes to provide incentives may be instituted to encourage users to provide feedback to the digital content items so that relevant digital content items are served to users while complying with the low entropy constraints imposed by various entities in communication networks.” Furthermore, as cited in ¶¶ [0117] [0119] “The digital content item may be suppressed if it is determined to be similar to other/prior digital content items which the user has excluded from the presentation and the user preferences indicated that similar items are to be excluded in the feedback data defined above. The digital content item may also be suppressed if it is determined that the digital content item is from a content source that was indicated by the user 796 as being excluded. Digital content items thus filtered as being dissimilar to the digital content item 722 and therefore not included in the exclusions list 774 are presented to the user 796…If the digital content items pertain to advertisements, the supervised or unsupervised learning may be based on advertiser-provided catalogs or hierarchies. At 856, recommendations regarding the digital content items to be added to the exclusions list 774 which are to be suppressed from presentation to the user 796 may be generated using, for example, collaborative filtering techniques. In an example, collaborative filtering may be based on aggregated content item statistics and the WAIST statistics. At 858, the similarity threshold configured for the recommendations may be retrieved. The similarity threshold may be configured from digital content items such as look-alike ads from seeded as of the user 796, or ads from similar users. The similarity threshold may be based on metrics such as but not limited to precision, recall, accuracy, Fl, or other proprietary measures.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Deng’s implementing, by one or more processors, between each module, a filter configured to remove at least one of data or inputs that have been identified as causing hallucinations to occur to Gupta’s processors are of a digital content generation system comprising a plurality of modules. One of ordinary skill in the art would have been motivated to do so because “With the proliferation of different types of digital content delivery mechanisms (e.g., mobile phone, portable computing devices, tablet devices, etc.), it has become crucial that content providers, such as vendors looking to advertise a good or service, engage users with content of interest. As a result, content providers are continuously looking for ways to deliver more appealing content.” (Deng: ¶ [0137]).
With respect to Claim 26:
Gupta does not explicitly disclose the method of claim 1, wherein the context associated with the existing digital content comprises at least one of information associated with when the existing digital content was served, a format of the existing digital content, or whether the existing digital content was provided for output on a website or in a mobile application.
However, Deng further discloses wherein the context associated with the existing digital content comprises at least one of information associated with when the existing digital content was served, a format of the existing digital content, or whether the existing digital content was provided for output on a website or in a mobile application (i.e. data feeds of parametrized variables associated with digital content include time distributions of when digital content was served, image/video format or campaign type in which the digital content was served, and whether the digital content is associated with different surfaces such as a mobile app such as Instagram or website) (Deng: ¶ [0056] “In an example wherein the dynamic digital content includes online advertisements, the parameterized variables may include campaign parameters/assets such as campaign spending, campaign types e.g., auction, brand, etc., campaign reach and frequency, Geo and time distributions, targeting audience and segments, situational creatives, campaign size and structure, and optimization goals such as traffic, leads, conversions, purchase, store visits, etc. For an offline advertisement campaign, the objectives may include, campaign parameters to be optimized to match the dynamic physical store availability, operating hours, inventory during a crisis/emergency, etc. The parameterized variables associated with the advertisement campaign may be part of the campaign setup. Another parameterized variable associated with the dynamic digital content transmission may include a selection of a particular channel based on infrastructure availability, bandwidth, situational coverage as cell signals or internet overage might be interrupted. This may include but is not limited to direct messaging, email, views from different surfaces (e.g., Whatsapp®, FB Messenger, Instagram, Blue App, etc.) based on telecommunication carrier's data feeds.” Furthermore, as cited in ¶ [0115] “When the user 796 selects option (1) to "Take me off from this ad", it may translate to exclude the user 796 from that ad and all ads from the same ad set which may have been created from the permutations/combinations of contents of the digital content item 722 by varying creatives, contents, image/video formats, colors, durations, frame ratio, other degrees of freedom. The selection of option (1) may add the user 796 explicitly to the highest ad structure, say ad set, applicable to all ads and variations of the ads.” Furthermore, as cited in ¶ [0133] “FIG. 10 shows a data flow diagram 1000 for content-providing networks such as advertising networks to enable the user 196 to request incentives according to an example. While a specific use case for advertising networks is illustrated herein, it can be appreciated that similar processes may be implemented in other digital content-providing networks also. The user 1030 may send a request 1032 to the ad network 1020 to redeem an incentive. The ad network 1020 may transmit an incentive redeeming message 1022 to the advertiser regarding the incentive redeeming request 1032 transmitted by the user 1030. The advertiser 1010 may confirm (e.g., with a backend database, etc.) if the incentive can be provided e.g., if an action such as a purchase, a recommendation, etc., was indeed executed by the user 1020. Upon receiving the confirmation/non-confirmation, an appropriate response 1012 may be provided by the advertiser 1010 to the ad network 1020. The ad network 1020 may forward the incentive redemption/rejection message 1024 to the user 1030. When a user chooses to send a request or receive confirmation via message, ad networks such as Facebook may use messaging applications such as WhatsApp® or other user-provided communication modes to connect the user and the advertiser, possibly engaging for more direct person-to-advertiser interactions. The confirmation from the ad network 1020 may include confirmation of targeting exclusion preferences and receipt of earned incentives with copies to both the user 1030 and the advertiser 1010. Ad networks 1020 may be able to report to both the user 1030 and the advertiser 1010 aggregated incentives for requested durations with backend integration to the advertiser to keep the balance up-to-date. If applicable, an option may be provided for the user 1030 to cash out on the accumulated credits towards future purchases/conversions in line with specifications as provided by the advertiser 1010 specified in a campaign setup.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Deng’s context associated with the existing digital content comprises at least one of information associated with when the existing digital content was served, a format of the existing digital content, or whether the existing digital content was provided for output on a website or in a mobile application to Gupta’s processing, by the one or more processors, the natural language prompt, the context, and the one or more values of the style vector as inputs through the generative model to generate digital content, wherein the generated digital content is based on the existing digital content. One of ordinary skill in the art would have been motivated to do so because “With the proliferation of different types of digital content delivery mechanisms (e.g., mobile phone, portable computing devices, tablet devices, etc.), it has become crucial that content providers, such as vendors looking to advertise a good or service, engage users with content of interest. As a result, content providers are continuously looking for ways to deliver more appealing content.” (Deng: ¶ [0137]).
Response to Arguments
Applicant’s arguments see pages 8-10 of the Remarks disclosed, filed on 03/10/2026, with respect to the 35 U.S.C. § 101 rejection(s) of claim(s) 1-23 have been considered but are not persuasive. The Applicant asserts “The amended claims solve this technical problem and reflect the technical solution by reciting the specific steps of "determining, by the one or more processors based on the style embeddings, one or more values of a style vector for the campaign creator, wherein the style vector corresponds to a weight or scale factor," and "processing, by the one or more processors, the natural language prompt, the context, and the one or more values of the style vector as inputs through the generative model to generate digital content wherein the one or more values of the style vector provides control over a style of the generated digital content to drive consistency of the generative model." Applicant respectfully submits that these claim features are not a mere instruction to "apply" any alleged abstract idea on a generic computer. Rather, the claims recite adjusting specific inputs (i.e., "the style vector correspond[ing] to a weight or scale factor" alongside a context) to physically govern the processing and output of the generative model. Just as the claims in Ex Parte Desjardins were found to reflect an improvement by adjusting the parameters of a machine learning model to solve a model-specific shortcoming (See, Dec. 2025 Memo), the presently amended claims improve the generative model technology itself by forcing stylistic consistency onto a non-deterministic generative model. (See, Specification, [0021], [0058]) Applicant respectfully submits that this constitutes a patent-eligible improvement to computer functionality. Further, the claims, as amended, recite a specific sequence of operations that improves computational efficiency. The claims now recite "clustering, by the one or more processors based on visual similarities, the existing digital content," and subsequently "determining.. style embeddings associated with the clustered existing digital content." By clustering the data prior to generating the embeddings, the claimed systems and methods avoid the need to process and calculate individual embeddings for massive amounts of raw digital content as input. (See, Specification, [0053]). Instead, processing the clustered data translates to processing smaller amounts of data, which physically saves computational resources, such as processing power and network overhead. As noted in the Dec. 2025 Memo, claim elements that allow a system to reduce the use of storage capacity or enable reduced complexity in the system are "tantamount to how the machine learning model itself would function in operation" and demonstrate a clear technological improvement.” The Examiner respectfully disagrees. The claims DO NOT recite “adjusting specific inputs” at most the claims further clarify the inputs being "the style vector correspond[ing] to a weight or scale factor" and processing the inputs through the generative model but do not recite adjusting any inputs. Furthermore, “the claims in Ex Parte Desjardins were found to reflect an improvement by adjusting the parameters of a machine learning model to solve a model-specific shortcoming (See, Dec. 2025 Memo)” and the claims of the instant application recite no such improvement to the generative model. Examiner notes that the claims reciting “clustering” and “determining” is recited at such a high level (i.e. based on visual similarities and style embeddings, respectively) that a person with the necessary information is able to perform this. Examiner would also like to note that the claims do not recite nor take into effect “computational resources, such as processing power and network overhead”, at most these effects to the computer are secondary or ancillary effects. Furthermore, Independent Claims 1, 8, and 15 recite limitations directed to the abstract idea including “receiving existing digital content associated with a campaign creator; identifying a context associated with the existing digital content; clustering, based on visual similarities, the existing digital content; receiving a natural language prompt trained to generate content from the natural language prompt; determining style embeddings associated with the clustered existing digital content, wherein the style embeddings quantify a style of the existing digital content via a numeric representation; determining, based on the style embeddings, one or more values of a style vector for the campaign creator, wherein the style vector corresponds to a weight or scale factor; and processing the natural language prompt, the context, and the one or more values of the style vector as inputs to generate digital content, wherein the generated digital content is based on the existing digital content.” These further limitations are not seen as any more than the judicial exception. Serving generated digital content based on natural language processing and style vectors is considered to be an abstract idea, specifically, certain methods of organizing human activity; such as managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions) because the claims are directed to managing a relationship between parties in order to provide information (i.e. generating content to be served). Serving generated digital content based on natural language processing and style vectors is also considered to be fall under another grouping of abstract idea, specifically, Mental Processes; such as concepts performed in the human mind (including an observation, evaluation, judgment, opinion) because the claims are directed to receiving information (i.e. existing digital content), identifying information (i.e. a context associated with existing digital content), clustering information (i.e. existing digital content based on visual similarities), receiving information (i.e. natural language prompt), determining information (i.e. style embeddings and style vector), and processing information (i.e. natural language prompt) in order to generate content which can all be performed in the human mind. Independent Claims 1, 8, and 15 recite additional limitations including “by one or more processors; for a generative model trained; and through the generative model.” The additional limitations including “by one or more processors; for a generative model trained; and through the generative model” are not found to integrate the judicial exception into a practical application because they are seen as adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) and generally linking the use of the judicial exception to a particular technological environment or field of use – see MPEP 2106.05(h). Furthermore, Examiner would like to note that according to ¶ [0059] of the Applicant’s specification; “In general, any AI model technique for generating digital content from a text prompt may be used to implement the generative model 110.” Accordingly, alone, and in combination, these additional elements are seen as using a computer or tool to perform an abstract idea, adding insignificant-extra-solution activity to the judicial exception. They do no more than link the judicial exception to a particular technological environment or field of use, i.e. processors and generative model, and therefore do not integrate the abstract idea into a practical application. The courts decided that although the additional elements did limit the use of the abstract idea, the court explained that this type of limitation merely confines the use of the abstract idea to a particular technological environment and this fails to add an inventive concept to the claims (See Affinity Labs of Texas v. DirecTV, LLC,).Therefore, the rejection(s) of claim(s) 1, 2, 4-9, 11-16, and 18-26 under 35 U.S.C. § 101 is maintained above with an updated analysis.
Applicant’s arguments see pages 10-14 of the Remarks disclosed, filed on 03/10/2026, with respect to the 35 U.S.C. § 103 rejection(s) of claim(s) 1-23 over Gupta in view of Deng have been considered but are not persuasive:
The Applicant asserts “The portion of Gupta identified and emphasized by the Action (see, Action, p.15) is directed to the generated digital content, not the candidate layouts (the contended existing digital content). Further, paragraph [0046] of Gupta explains that "the hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118." There is nothing in Gupta that teaches or suggests that the candidate layouts, identified by the Action as corresponding to the claimed existing digital content, are clustered "based on visual similarities."” The Examiner respectfully disagrees. The Examiner would like to refer the Applicant to ¶ [0034] of the Gupta reference; “Consider an example in which the generation module 110 processes the input data 114 and the content data 118 to compare the vector representation of the characteristic 116 with the vector representations of candidate layouts using locality-sensitive hashing. In this example, the generation module 110 identifies a particular layout or a set of particular candidate layouts based on a similarity between the characteristic 116 and the particular layout or the particular candidate layouts.” It is clear from the disclosure above that the Gupta reference teaches clustering or identifying a set of candidate layouts based on visual similarities such as ‘hot air balloon’ style or similar classifications of ‘hot air balloon images’.
The Applicant also asserts “Claim 1, as amended, requires "a context associated with the existing digital content" be identified and processing "the context as input[] through the generative model to generate digital content." Applicant respectfully submits that neither Gupta nor Deng teaches or suggests the amended claim feature. Rather, Gupta determines layouts based on objectives (See, Gupta, [0048]-[0052]), and Deng determines parameterized variables from data feeds such as weather or geolocation. (See, Deng, [0059]-[0061]).” The Examiner respectfully disagrees. The Examiner would like to refer Applicant to ¶ [0032] of the Gupta reference; “For example, the generation module 110 receives and processes the input data 114 to generate a vector representation of the characteristic 116. In some examples, the generation module 110 includes or has access to a machine learning model trained on training data to generate vector representations of natural language statements, and the generation module 110 implements the machine learning model to generate the vector representation of the characteristic 116. In other examples, the generation module 110 generates the vector representation of the characteristic 116 using a hash function such as a locality-sensitive hash function. For example, the generation module 110 leverages the vector representation of the characteristic 116 to identify candidate layouts and/or candidate strategies for achieving the objective of the digital content that is to be generated.” Furthermore, as cited in ¶ [0047] “In a first example, the candidate layouts are selected from a corpus of layouts of digital content based on an association with instances of digital content which achieved corresponding objectives for the instances of digital content ( e.g., resulted in highly attended events, received greater than a threshold number of user interactions, etc.). In the first example, the candidate strategies for achieving objectives of digital content are selected from a corpus of strategies based on an association with instances of digital content which achieved corresponding objectives. In a second example, the candidate layouts described by the content data 118 include each layout included in the corpus of layouts of digital content.” It is clear from the disclosure above that the Gupta reference teaches identifying an objective, theme, or context with the corpus of digital content.
The Applicant finally asserts “Applicant respectfully submits that the cited portions of Gupta do not teach or suggest providing the "vector representation of candidate layouts" (the contended "style vector") as input into the a generative machine learning model. Rather, the only inputs described in the cited portions of Gupta are "text data 214 descripting the output text 502" (Gupta, [0055]) or output text 502 (Gupta, [0056]). Applicant respectfully submits that the text data 214 or output text 502 is not equivalent to the vector representations of candidate layouts of Gupta, let alone the style vector as claimed. The vector representations Gupta uses vector representations strictly as a search and selection mechanism to pick a static digital template. There is nothing in Gupta that teaches or suggests that the vector representation of candidate layouts are provided as input through the generative model… That is, the hashing module 202 of Gupta compares the vector representation of the objective to vector representations of candidate layout to identify a top candidate layout defined in a digital template (e.g., template 306). The prompt module 204 of Gupta "parametrizes the particular layout with placeholders 402 which will be replaced by digital content components generated using the second machine learning model." (Gupta, [0050]). "After classifying the placeholders 402 as types of digital content components, the prompt module 204 includes the types of digital content components in input text 404 to be processed by the first machine learning model." (Gupta, [0051]). The "prompt module 204 generates the prompt data 212 as describing the input text 404. The language module 206 receives and processes the prompt data 212 in order to generate text data 214. FIG. 5 illustrates a representation 500 of generating output text [502] by processing input text." (Gupta, [0052]). The generative model of Gupta "receives the text data 214 describing the output text 502." (Gupta, [0055]). Accordingly, it is text data 214, describing output text 502, and not the vector representation of candidate layouts (the contended style vector) that is provided as input to the generative model of Gupta.” The Examiner respectfully disagrees. The Examiner would like to refer the Applicant to ¶¶ [0049]-[0051] of the Gupta reference; “The hashing module 202 compares the vector representation of the objective 302 with the vector representations of the candidate layouts described by the content data 118 using locality-sensitive hashing to identify a top N candidate layouts. In one example, the hashing module 202 identifies a particular layout for the digital content to be generated based on the locality-sensitive hashing. In this example, the particular layout is defined in a digital template 306. For instance, the hashing module 202 generates the match data 210 as describing "Incentives" from the candidate strategies 304 and the digital template 306. Although the hashing module 202 is illustrated as generating the match data 210 describing the candidate strategies 304 and the digital template 306, it is to be appreciated that the hashing module 202 is also capable of generating the match data 210 as describing other content features for digital content such as tonality, colors (e.g., color pallets), image features, etc…The prompt module 204 receives and processes the match data 210 in order to generate prompt data 212. FIG. 4 illustrates a representation 400 of generating input text. As shown, the representation 400 includes the digital template 306 described by the match data 210, and the prompt module 204 parameterizes the particular layout with placeholders 402 which will be replaced by digital content components generated using the second machine learning model. The prompt module 204 classifies the placeholders 402 into types of digital content components such as digital images, slogans ( e.g., lines of text with less than 10 words rendered using a font having a weight greater than 700), paragraphs of text, headings/headers, footers, call-to-action buttons, etc. The prompt module 204 also classifies content blocks of the particular layout which include digital images and/or text…After classifying the placeholders 402 as types of digital content components, the prompt module 204 includes the types of digital content components in input text 404 to be processed by the first machine learning model. For types of the digital content components that are digital images, the input text 404 requests alternative text describing the digital images. The input text 404 includes portions of the objective 302 such as natural language text stating "Invite registrations for kite flying festival." As shown, the input text 404 also includes a discount of"15% off" based on the particular strategy for achieving the objective 302 of "Incentives." The prompt module 204 generates the input text 404 as including a request for the first machine learning model to generate output text formatted using JavaScript Object Notation. The JavaScript Object Notation encodes a relative order of the types of digital content components defined by the particular layout of the digital template 306.” It is clear from the disclosure above that the Gupta reference teaches processing the natural language prompt, candidate layout to generate digital content, and values of style vector as match data which is fed into the machine learning model as input data in order to generate content, wherein vector representation includes a term frequency-inverse which is a value that corresponds to a weight or scale factor. Therefore, the rejection(s) of claim(s) 1, 2, 4-9, 11-16, and 18-26 under 35 U.S.C. § 103 is provided above with updated citations.
Conclusion
The prior art made of record and not relied upon is considered pertinent to Applicant’s disclosure. The following reference are cited to further show the state of the art:
U.S. Publication 2023/0252224 to Tran for disclosing to generate a document with a transformer by prompt-engineering the transformer with a title and a summary to generate a description of the document; displaying a set of claims and allowing user editing of the set of claims; receiving one or more figures; receiving a part list with a plurality of element names for each figure; generating an expanded description of each element name through prompt engineering based on prior text in the document; selecting one or more boilerplate texts for major sections of the document; and organizing the document with the title, a background, the summary, a brief description of the drawings, and a detailed description.
U.S. Patent 10,026,098 to Meyer for disclosing a computerized system and techniques facilitate the monitoring and management of online behaviorally-targeted advertising. In certain embodiments, electronic notifications related to advertising practices of members of an online advertising ecosystem are presented to users based on the discovery of elements of online content aimed at delivering targeted advertising messages to viewers of the content.
U.S. Patent 11,392,643 to Guillen for disclosing a method receives a file describing characteristics for delivery of a creative on a video delivery system. The file is queried to identify elements in the string that define metadata. The string is written in a structural language and defines characteristics for the delivery of the creative. The method retrieves tag metadata for tags that define structural elements and validates the tag metadata based on a first specification. Media file metadata is obtained for a media file based on a link to the media file and the media file metadata is validated based on a second specification. The method outputs a result based on the validations. The creative is eligible for insertion during a break of streaming a main video on the video delivery system when the tag metadata and the media file metadata are validated.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Azam Ansari, whose telephone number is (571) 272-7047. The examiner can normally be reached from Monday to Friday between 8 AM and 4:30 PM.
If any attempt to reach the examiner by telephone is unsuccessful, the examiner's supervisor, Waseem Ashraf, can be reached at (571) 270-3948.
Another resource that is available to applicants is the Patent Application Information Retrieval (PAIR). Information regarding the status of an application can be obtained from the (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAX. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pairdirect.uspto.gov. Should you have questions on access to the Private PAIR system, please feel free to contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
Applicants are invited to contact the Office to schedule either an in-person or a telephonic interview to discuss and resolve the issues set forth in this Office Action. Although an interview is not required, the Office believes that an interview can be of use to resolve any issues related to a patent application in an efficient and prompt manner.
/AZAM A ANSARI/
Primary Examiner, Art Unit 3621
May 8, 2026
Prosecution Insights