DETAILED ACTION
Claims 1-14 are presented for examination.
Claims 1, 4, 8 and 11 were amended.
This is a Final Action.
Response to Arguments
Applicant's arguments filed 08/17/2026 have been fully considered but they are not persuasive.
With respect to claim 1 applicant argues “the combination of Cha and Jain fails to render obvious the feature "using the style description data to affect output of an artificial intelligence model...wherein the style description data is configured to control the specific style of digital content generated by the artificial intelligence model without affecting a subject of the digital content" of the amended claim 1. Referring to paragraphs [0016] of Cha, Cha extracts a user style embedding from reference content, compares the user style embedding with pre-stored model style embeddings, ranks a plurality of generative AI models according to similarity scores, and then routes a user prompt to a selected highest-ranked generative AI model for content generation. Thus, in Cha, the user's style information functions as a criterion for selecting among different Ai models. The style information in Cha is not incorporated into the prompt or instruction itself, in order to affect output of an artificial intelligence model.”
Examiner respectfully disagrees with the applicant. Claim 1 broadly requires “using the style description data to affect output of an AI model” and that style-description data to configured to control the specific style of the generated digital content without affecting its subject. Cha teaches precisely such functional separation. Cha uses the user style embedding to rank and select generative AI models based on similarity to the user’s style preference, thereby determining the stylistic characteristic of the generated output, while a separately received prompt specifies the subject/content to be generated. Cha expressly teaches that the prompt may include a “text description of the content” and that the selected generative AI model generates content corresponding to the prompt, while the style-based model selection produces content that is mode stylistically similar to the user’s style preferences. Accordingly, Cha’s style information affects and controls the style of the AI generated output through style-based model selection, whereas the separately supplied prompt controls the subject of the generated content.
Applicant further argues In the teaching of Cha, the user style embedding is used for selecting an AI model, rather than affecting the output of the selected AI mode. Therefore, Applicant respectfully submits that Cha fails to teach, suggest or disclose the feature "using the style description data to affect output of an artificial intelligence model...wherein the style description data is configured to control the specific style of digital content generated by the artificial intelligence model without affecting a subject of the digital content" of the amended claim 1, and Jain cannot cure the deficiencies of Cha.
Examiner respectfully disagrees with the applicant. Claim 1 does not require the style-description data to be incorporated into a prompt, supplied as an input to the selected model, or used to modify the internal operation of the model. Cha uses the user style embedding, which represents the user’s style preference, to rank and select generative AI models according to style similarity, thereby determining which model generates the output and consequently controlling the style of the resulting digital content. Cha separately teaches that the user prompt may include a text description of the content to be generated, such that the prompt controls the subject while the user style embedding controls the stylistic model selection. Moreover, Jain expressly teaches applying a selected style profile to a separately generated storyline, stating that “the content is transformed by applying the style profile to the storyline”, and that slides may be selected or created to have style attributes matching the style profile. Thus, the combined teaching of Cha and Jain teach or suggest using user-specific style data to control the style of generated content independently of the subject/storyline of the content. Jaine paragraph 70 expressly teaches selecting a style profile based on audience parameters/user fingerprints and states that the user fingerprints determine, for example, image preferences, color preferences, illustration preferences and presentation preferences. Jain further explains that the style profile includes stylistic attributes of the content. Paragraph 72 in Jain explicitly recites “the content is transformed by applying the style profile to the storyline generated by the outline engine 108”. Jain therefore supplies an even more direct teaching of applying style data to the content while maintaining the separately determined storyline/subject.
For similar reasons to above in view of claim 1, the independent 8 and dependent claims 2-7 and 9-14 rejection is also maintained.
In view of compact prosecution, Examiner is providing a new reference which also teaches applicants concerns. Ravi et al. (US 2025007842) teaches the style information does not merely select which AI model is used. Instead, it is supplied to/condition layers of the same image-generation neural network (Fig 5, - teaches conditioning high-resolution up sampling layers of the neural network with an image vector representation, while low-resolution layers receive the text vector representation without the image-vector representation. Thus, the style related information actually enters and conditions the AI model that produces the output, not merely used to choose among models.).
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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 of this title, 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.
Claims 1-4, 7-11, and 14 rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2024/0370660) in view of Jain et al. (US 2024/0061994)
1. Cha teaches, A digital content generation method (Fig 5, Paragraph 14 – teaches a method for style-specific digital content generation, Cha), comprising:
using the style description data to affect output of an artificial intelligence model and enabling the artificial intelligence model to generate digital content having a specific style (paragraph 16, Fig 2 and Fig 5 – teaches using style data to control/select generative AI output, configured to extract user style embedding from reference digital content items received from users and that the extracted user style embedding represents a style preference of the user. The system may compare the user style embeddings to prestored model style embeddings to generate a ranked list of generative AI models, and access one or more highest ranked generative AI models from the ranked list to generate novel digital content based on a prompt from the user; as the input data to a generative AI model that is ranked highest on the ranked list to cause generation of novel digital content item that is stylistically similar to one or more reference digital content items received from the user, Cha), wherein the style description data is configured to control the specific style of digital content generated by the artificial intelligence model (Fig 2 shows reference digital content items 230 being processed by style detection model 202 to produce user style embedding 232, which is supplied to style similarity ranking unit 204. The ranking unit compares the user style embedding with model style embeddings and selects/ranks generative AI models 222-226; Paragraph 30 – teaches that reference data 164 represents a user’s style preference; Paragraph 31 teaches extracting user style embedding 110 and that the style detection model may output a style feature label indicating a style attribute. Paragraphs 32-33 – teaches comparing the user style embedding with model style embeddings and ranking the generative AI models according to those tyle similarity values; Paragraph 36 – teaches that the selected generative AI models produce content that is stylistically similar to the user style preferences, Cha) without affecting a subject of the digital content (Fig 2 – teaches separate style selection input from the content prompt: user style embedding 232 enters style similarity ranking unit 204, wherein user prompt 236 separately enters generative AI model pool 220. Paragraph 34 – teaches that after the style based ranked list is generated, the system separately requests a prompt. The user input may indicate the prompt, including “a text description of the content” and the selected AI model generate digital content “that corresponds to the prompt”. Thus, the prompt defines the content/subject, while the separate style-embedding path determines which stylistically matching AI model generates that content, Cha).
Cha does not explicitly teach or recite,
establishing a style description database;
obtaining identity identification data corresponding to a target user;
according to the identity identification data, obtaining style description data corresponding to the target user from the style description database; and
Jain teaches,
establishing a style description database (Paragraph 33, Fig 1 – teaches the application server 102 databases may include an application database 112a and user database 112b. The application database 112a includes a library of stylistic resources (e.g., font, color, graphics, charts, tables, etc.) that is accessed by the style engine 109 to generate the style profile; The style engine 109 may be configured to store each style profile in the application database 112a, Jain);
obtaining identity identification data corresponding to a target user (Fig 8 – teaches receiving target audience/user identifying information, Paragraph 67 teaches at step 806 the fingerprint engine 106 receives at least one target audience parameter, the ‘target audience’ corresponds to the intended or expected audience that the content is being tailored to, Jain);
according to the identity identification data, obtaining style description data corresponding to the target user from the style description database (Fig 8, Paragraph 70 – teaches at step 812, the style engine 109 selects (or creates) a style profile for the content, Jain).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which said subject matter pertains to modify Cha’s style specific generative AI content generation system to use Jain’s stored user/audience style profile database and user fingerprint retrieval process. Cha already teaches generating digital content using generative AI models selected based on a user style embedding so that the generated content is stylistically similar to the user’s style preference. Jain teaches storing style profiles in an application database and selecting a style profile based on target audience parameters/user fingerprints. A POSITA would have been motivated to incorporate Jain’s stored style profile database into Cha’s generative AI system to provide persistent user specific style information and to avoid requiring the user to provide reference digital content each time style specific content is generated.
2. The combination of Cha and Jain teach, The digital content generation method according to claim 1, wherein the style description database is configured to store a plurality of style description data corresponding to different users (Paragraphs 28 and 45 – teaches that the fingerprint engine may store the fingerprint of each user in the user database; wherein the fingerprints represent style and preference of the user, Jain).
3. The combination of Cha and Jain teach, The digital content generation method according to claim 1, wherein obtaining the style description data corresponding to the target user from the style description database according to the identity identification data comprises:
according to first identity identification data, obtaining first style description data corresponding to a first user from the style description database (Fig 1 and 8, Paragraphs 28, 67– teaches generating and storing user specific fingerprints and retrieving corresponding fingerprints based on target audience parameters including member names and the fingerprint engine is configured to access, retrieve or download the corresponding user fingerprints form the application database, Jain), wherein the first style description data is configured to affect the output of the artificial intelligence model and enable the artificial intelligence model to generate first digital content having a first style (Fig 2/ Fig 5, Paragraphs 32 and 34- teaches the user style embedding affects generative AI output by ranking/selecting models according to style similarity and compares the user style embeddings to model style embeddings to generate similarity values. Further teaching, after generating the ranked list, the prompt is routed to highest ranked generative AI mode to cause the recipient generative AI model to generate one or more digital content items 168, and selected model IDs indicate the models selected for generation of digital content that corresponds to the prompt, Cha); and
according to second identity identification data, obtaining second style description data corresponding to a second user from the style description database (paragraph 28 - teaches multiple different users and corresponding user specific fingerprints and the fingerprint engine may store the fingerprint of each user in the user database 112. Jain further teaches that target audience parameter may include names of members and the fingerprint engine retrieves corresponding user fingerprints, Jain), wherein the second style description data is configured to affect the output of the artificial intelligence model and enable the artificial intelligence model to generate second digital content having a second style (Abstract, Fig 2/ Fig 5, Paragraphs 32 and 34- - teaches generating style specific digital content based on user style preference; that the user style embedding is extracted from reference data and then used to rank generative AI models based on similarity; further teaching that the prompt is outed to selected highest ranked generative AI mode to cause the recipient generative AI model to generate one or more digital content items 168, Cha), and the first style is different from the second style (paragraph 33 – teaches distinguishing styles by comparing style embeddings and ranking AI models based on style similarity values for example different model style embeddings have different cosine similarity values and are ranked accordingly, Cha).
4. The combination of Cha and Jain teach, The digital content generation method according to claim 1, wherein using the style description data to affect the output of the artificial intelligence model, and enabling the digital content generated by the artificial intelligence model to have the specific style comprises:
obtaining a first operation instruction corresponding to the target user, wherein the first operation instruction is configured to control the subject of the digital content (Paragraph 34, Fig 5:508 – teaches receiving user prompt for content generation; the user input data 170 may indicate a prompt for which digital content is to be generated, such as text description of the content, Cha);
according to the first operation instruction and the style description data, generating a second operation instruction, wherein the style description data is configured to control a style of the digital content (Claim 18, Fig. 1/5 – teaches extracted user style embedding represents a style preference of the user… ranking, a plurality of model style embeddings based on similarity with the user style embeddings to generate a ranked list of generative artificial intelligence (AI) models… receiving a user input indicating a prompt for content generation and providing the prompt as input data to one or more highest ranked generative AI models from the ranked list of generative AI models to generate one or more digital content items , Cha); and
based on the second operation instruction, instructing the artificial intelligence model to generate the digital content having the specific style (Fig 1 and Fig 5 Paragraph 34 – teaches a prompt may be provided by the client device to the generative AI model pool to a particular number of highest ranked generative AI model from the ranked list to cause the recipient generative Ai models to generate one or more digital content items and the system provides the prompt to a highest ranked model to cause generation of a novel digital content item that is stylistically similar to one or more reference digital content items received from the user, Cha).
7. The combination of Cha and Jain teach, The digital content generation method according to claim 1, wherein obtaining the identity identification data corresponding to the target user comprises:
obtaining the identity identification data corresponding to the target user from a block chain network or an online storage space (Fig 1, Paragraph 27, 45 & 48 – teaches database accessible through an internet-based application server is an online storage space. Jain’s user fingerprint/profile data corresponds to identity/profile data for the target user because the fingerprint is generated for each user and represents the user’s preferences when the user is the target audience Jain).
Claim 8 is similar to claim 1 hence rejected similarly.
Claim 9 is similar to claim 2 hence rejected similarly.
Claim 10 is similar to claim 3 hence rejected similarly.
Claim 11 is similar to claim 4 hence rejected similarly.
Claim 14 is similar to claim 7 hence rejected similarly.
Claims 5, 6, 12 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2024/0370660) in view of Jain et al. (US 2024/0061994) further in view of Jootoo Ramesh Bapu et al. (US 2026/0003874 – “Jootoo”)
All the limitations of claim 4 are taught above.
5. The combination of Cha and Jain teach, The digital content generation method according to claim 4, wherein instructing the artificial intelligence model to generate the digital content having the specific style based on the second operation instruction (Fig 5, Cha) comprises:
inputting the processed second operation instruction to the artificial intelligence model, so as to instruct the artificial intelligence model to generate the digital content having the specific style (Fig. 5 Paragraph 52 – teaches providing the prompt as input data to selected generative AI models to generate digital content, wherein the input data is provided to one or more highest ranked generative AI models from the ranked list of generative AI models to generate one or more digital content items, Cha).
The combination of Cha and Jain do not explicitly teach, processing the second operation instruction through a large language model to normalize instruction content of the second operation instruction.
However, Jootoo teaches,
processing the second operation instruction through a large language model to normalize instruction content of the second operation instruction (Fig 5:502, 504 – teaches normalizing instructions/query content by rewriting the query to correct errors, clarify ambiguity, remove irrelevant information… Under BRI the claimed “second operation instruction” maps to Jootoo’s query/prompt/instruction content, and the rewritten query is the normalized instruction content, Jootoo).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which said subject matter pertains to modify Cha’s style specific generative AI content generation system uses Jootoo’s query rewriting and normalization technique before submitting the instruction to Cha’s selected generative AI model. Cha already receives a user prompt and provide the prompt to one or more highest ranked generative AI models to generate style specific digital content. Jootoo teaches improving LLM query processing by rewriting a query to correct grammatical errors, clarify ambiguous terms … A POSITA would have been motivated to apply Jootoo’s query normalization process to Cha’s prompt in order to reduce ambiguity, improve instruction quality and provide a clearer processed instruction to the generative AI model for generating the requested digital content in the selected style.
All the limitations of claim 1 are taught above
6. The combination of Cha and Jain teach, wherein obtaining the style description data corresponding to the target user from the style description database according to the identity identification data (Fig 1 and paragraph 28, Jain)comprises:
based on the processed identity identification data, querying the style description database to obtain the style description data (Paragraph 48 and 57 - teaches access, retrieve or download the corresponding user fingerprints and the user fingerprints can be used to determine the image preferences, color preferences… visualizing data for each audience member Jain).
The combination of Cha and Jain do not teach,
processing the identity identification data through an embedded model to normalize data content of the identity identification data.
However, Jootoo teaches,
processing the identity identification data through an embedded model to normalize data content of the identity identification data (Fig 4 and 5 – teaches using an embedding generator /query rewriting process to normalize input data content before retrieval. Under BRI, the claimed “embedded model” can be mapped to Jootoo’s embedding generator/model and normalized data content is taught by Jootoo’s rewriting/clarifying/removing/simplifying of input content and generation of a semantic embedding, Jootoo; Applied to Jain, the identity identification data/user or target audience information would be processed into a normalized/semantic representation before database querying).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which said subject matter pertains to modify Jain’s identify based retrieval of user fingerprints/style profiles to use Jootoo’s embedding based normalization before database querying. Jain already retrieves user-specific preference/style data based on target audience parameters or user identifying information while Jootoo teaches normalizing user input by rewriting ambiguous or unclear content and generating embeddings that represent semantic understanding for retrieval. A POSITA would have been motivated to apply Jootoo’s embedding based normalization to Jain’s user identification/profile lookup to improve matching accuracy, reduce ambiguity in user identifying input and retrieve the correct user specific style description data from the database.
Claim 12 is similar to claim 5 hence rejected similarly.
Claim 13 is similar to claim 6 hence rejected similarly.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Ravi et al (US 2025/0077842) – explicitly teaches the style information does not merely select which AI model is used. Instead, it is supplied to/condition layers of the same image-generation neural network.
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AMRESH SINGH whose telephone number is (571)270-3560. The examiner can normally be reached Monday-Friday 8am-5pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Ann J. Lo can be reached at (571) 272-9767. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/AMRESH SINGH/Primary Examiner, Art Unit 2159
Prosecution Insights