DETAILED ACTION
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 .
Response to Amendment
The objection to the claims, specifications have been withdrawn in view of applicant’s amendments filed on 05/15/2026.
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 (i.e., changing from AIA to pre-AIA ) 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, 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) 1-7 and 9-10 is/are rejected under 35 U.S.C. 103 as being unpatentable over Sarna et al. (U.S. Pub. No. 20200175740) in view of Wang (CN Pub. No. 117689770).
Regarding claim 1, Sarna discloses a method for rapidly generating multiple customized user avatars includes (para 17, “FIG. 6 is a flowchart illustrating a first example method for creating a cartoon using a classifier.”): an avatar training processor of a model training server extracts a model parameter list from a model database (para 25, “the application server 108 may include a processor 216,”; also, para 27, “The memory 218 is also capable of storing other instructions and data, including, for example, an operating system, hardware drivers, other software applications, databases, etc.”), and according to each group of a model parameters, extracts corresponding a plurality of image files from a multimedia database (para 66, “The method 500 accesses 502 a large image data set”; also, para 66, “method 500 then searches 508 for a smaller set of images that include faces (classifier for human faces) from the large image data set”), avatar models are respectively related to those model parameters and are saved to the model database (para 66, “The method 500 then crowd sources 512 the clean smaller set of images to produce positive images (tuples of image, attribute, label).”; also, para 40, “The cartoon asset module 206 may be steps, processes, functionalities or a device including routines for storing, managing, and providing access to the individual sets of cartoon assets.”); through an Internet, an electronic device transmits a plurality of classifying labels selected by a user from a classifying label selection interface of an application program displayed on a displaying screen of the electronic device to the model training server (para 72, “the network 102 may include a local area network (LAN), a wide area network (WAN) (e.g., the Internet), and/or other interconnected data paths across which multiple devices may communicate”; also, para 7, “editing the cartoon asset for the attribute included in the cartoon avatar based on the user input”), the model training server receiving those classifying labels, organizing those classifying labels into a label parameter group (para 66, “In block 506, the method 500 uses the images to train a deep neural network 226”; also, para 66, “The method 500 then cleans 510 the smaller set of images from block 508 by removing images that do not match the class.”), extracting the model parameter list from a database server, further screening out corresponding a plurality of those model parameters same as or similar to the label parameter group (para 27, “The memory 218 may store and provide access to data to the other components of the application server 108”; also, para 66, “The method 500 then crowd sources 512 the clean smaller set of images to produce positive images (tuples of image, attribute, label). In some implementations, this can be done by a dedicated group of human raters rather than crowdsourcing.”), then, based on those model parameters, extracting corresponding a plurality of those avatar models from the database server (para 67, “Then the method 600 determines 608 the cartoon assets corresponding to the attribute and label. For example, the cartoon asset module 206 may search the cartoon assets for particular assets that match the attribute and label provided by the block 606.”), packing those avatar models and transmitting them to the application program, the application program receiving those avatar models, unpacking them, and showing them on the displaying screen for the user to select (para 76, “The application server 108 may be a computing device that includes a processor, a memory and network communication capabilities.”; also, para 7, “presenting the cartoon avatar on a display of a user device, receiving user input on the display, editing the cartoon asset for the attribute included in the cartoon avatar based on the user input, and re-rendering the cartoon avatar based on the edited cartoon asset”). Sarna does not disclose a deep-learning text-to- image diffusion model is used to generate a plurality of avatar models.
However, in a similar field of endeavor, Wang discloses a deep-learning text-to- image diffusion model is used to generate a plurality of avatar models (para 42, “a Diffusion model to generate a character model, wherein the Diffusion model is a deep learning text-to-image generation model for generating high-quality images.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sarna's invention of generating customized cartoon avatars from stored attribute-and-label assets retrieved through a model training server with the features of Wang's invention of a deep-learning text-to-image diffusion model used to generate character models. One of ordinary skill in the art would have been motivated to make this combination for the following reasons. First, Sarna builds a library of avatar assets in advance by training a deep neural network and producing labeled positive images, teaching that "the method 500 uses the images to train a deep neural network 226" and "crowd sources 512 the clean smaller set of images to produce positive images (tuples of image, attribute, label)," but Sarna populates that library from existing image data sets rather than from a generative text-to-image model. Second, Wang supplies precisely that missing generation mechanism, expressly teaching "a Diffusion model to generate a character model, wherein the Diffusion model is a deep learning text-to-image generation model for generating high-quality images." Third, one of ordinary skill in the art populating Sarna's stored avatar library would have been motivated to generate the stored avatar models with Wang's diffusion model because doing so yields higher-quality character images keyed to text labels while preserving Sarna's downstream architecture of relating the avatar models to their parameters, saving them to the model database, and retrieving them by label matching, a predictable improvement in image quality with no change to Sarna's.
Regarding claim 2, Sarna as modified by Wang discloses the method defined in Claim 1, the application program receives the user's selection of one of those avatar models (Sarna: para 43, “the cartoon asset module 206 receives one or more user selections of a cartoon style and an emotional expression”; also, para 7, “editing the cartoon asset for the attribute included in the cartoon avatar based on the user input”), then transmits selected the avatar model along with a registration notice to the database server The user application is coupled for communication with the respective computing device 106 or the application server 108 to receive, send or present messages, status, commands and other information. (Sarna: para 31, “The user application is coupled for communication with the respective computing device 106 or the application server 108 to receive, send or present messages, status, commands and other information.”), the database server, in turn, updates a status code of corresponding the avatar model to "registered" as per the registration notice, and binds the avatar model to the user (Sarna: para 76, “The application server 108 may also be configured to receive status and other information from the computing devices 106a through 106n via the network 102.”).
Regarding claim 3, Sarna as modified by Wang discloses the method defined in Claim 1, the displaying screen shows those avatar models (Sarna: para 26, “the processor 216 may be capable of generating and providing electronic display signals to a display device, supporting the display of images, capturing and transmitting images, performing complex tasks including various types of feature extraction and sampling, etc.”), the displaying screen also displays a re-generation icon, if the user clicks the re-generation icon, the application program transmits a regeneration notice to the model training server (Sarna: para 35, “The user interface module 202 generates user interfaces for a user to input photos, control the generation of cartoon images or icons, and use the cartoon images or icons in one or more user applications.”; para 31, “The user application is coupled for communication with the respective computing device 106 or the application server 108 to receive, send or present messages, status, commands and other information.”).
Regarding claim 4, Sarna as modified by Wang discloses the method defined in Claim 3, the model training server receives the regeneration notice (Sarna: para 31, “The user application is coupled for communication with the respective computing device 106 or the application server 108 to receive, send or present messages, status, commands and other information.”) application program, the application program unpacks received those avatar models and shows them on the displaying screen for the user to select (Sarna: para 7, “presenting the cartoon avatar on a display of a user device”; also, para 76, “The application server 108 may be a computing device that includes a processor, a memory and network communication capabilities”). Sarna does not disclose and uses a deep-learning text-to-image diffusion model to generate those avatar models in real time.
However, in a similar field of endeavor, Wang discloses and uses a deep-learning text-to-image diffusion model to generate those avatar models in real time (para 42, “a Diffusion model to generate a character model, wherein the Diffusion model is a deep learning text-to-image generation model for generating high-quality images.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sarna's invention of a model training server that receives a regeneration notice and provides regenerated avatar models for display with the features of Wang's invention of a deep-learning text-to-image diffusion model used to generate character models in response. One of ordinary skill in the art would have been motivated to make this combination because Wang expressly teaches "a Diffusion model to generate a character model, wherein the Diffusion model is a deep learning text-to-image generation model for generating high-quality images," and applying Wang's diffusion model as the engine that fulfills Sarna's regeneration request lets the model training server produce a freshly generated, higher-quality avatar in response to the user's regeneration notice rather than returning only previously stored assets, while leaving Sarna's packing, transmitting, and on-screen presentation flow unchanged.
Regarding claim 5, Sarna as modified by Wang discloses the method defined in Claim 3, the model training server receives the regeneration notice (Sarna: para 31, “The user application is coupled for communication with the respective computing device 106 or the application server 108 to receive, send or present messages, status, commands and other information.”), and extracts corresponding a plurality of image files from the database server based on those model parameters corresponding to each label parameter group (Sarna: para 66, “The method 500 accesses 502 a large image data set”; also, para 27, “The memory 218 may store and provide access to data to the other components of the application server 108”), application program, the application program receives those avatar models, unpacks them, and shows them on the displaying screen for the user to select (Sarna: para 7, “presenting the cartoon avatar on a display of a user device”; also, para 76, “The application server 108 may be a computing device that includes a processor, a memory and network communication capabilities”). Sarna does not disclose the model training server uses the deep-learning text-to-image diffusion model to generate those avatar models in real time.
However, in a similar field of endeavor, Wang discloses the model training server uses the deep-learning text-to-image diffusion model to generate those avatar models in real time (para 42, “a Diffusion model to generate a character model, wherein the Diffusion model is a deep learning text-to-image generation model for generating high-quality images.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sarna's invention of a model training server that receives a regeneration notice and retrieves image files keyed to model parameters with the features of Wang's invention of a deep-learning text-to-image diffusion model used to generate character models in real time. One of ordinary skill in the art would have been motivated to make this combination because Wang expressly teaches "a Diffusion model to generate a character model, wherein the Diffusion model is a deep learning text-to-image generation model for generating high-quality images," and using Wang's diffusion model to generate the avatar models from the image files Sarna retrieves for each label parameter group provides a real-time, high-quality regeneration path that directly answers the user's regeneration request while reusing Sarna's parameter-keyed retrieval and on-screen presentation steps.
Regarding claim 6, Sarna as modified by Wang discloses the method defined in Claim 5, the method to generate those image files includes, the model training server extracting a classifying label list from the database server (Sarna: para 66, “The method 500 accesses 502 a large image data set”; also, para 38, “the set of classifiers 204 for classifying photos of users into the above attributes are combined into a pipeline. A photo or image may be provided to the pipeline as input”), those image files to those classifying labels, and saving them to the database server (Sarna: para 26, “the processor 216 may be coupled to the memory 218 via the bus 214 to access data and instructions therefrom and store data therein”; also, para 27, “The memory 218 is also capable of storing other instructions and data, including, for example, an operating system, hardware drivers, other software applications, databases, etc.”). Sarna does not disclose based on the text contents of those classifying labels, using the deep-learning text-to-image diffusion model to generate those image files.
However, in a similar field of endeavor, Wang discloses based on the text contents of those classifying labels, using the deep-learning text-to-image diffusion model to generate those image files (para 42, “a Diffusion model to generate a character model, wherein the Diffusion model is a deep learning text-to-image generation model for generating high-quality images.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sarna's invention of building a labeled image library that is related to classifying labels and saved to a database with the features of Wang's invention of a deep-learning text-to-image diffusion model that generates images from text. One of ordinary skill in the art would have been motivated to make this combination because Wang expressly teaches "a Diffusion model to generate a character model, wherein the Diffusion model is a deep learning text-to-image generation model for generating high-quality images," which is a text-to-image generator, and generating Sarna's image files from the text contents of the classifying labels with Wang's diffusion model produces label-faithful, high-quality image files that Sarna then relates to those same classifying labels and saves to the database server, populating the library directly from the labels rather than from a pre-existing image data set.
Regarding claim 7, Sarna as modified by Wang discloses the method defined in Claim 1, those classifying labels including garment, action, object, person, background, ornament, style, and a classifying label input field (Sarna: para 36, “A label corresponds to a value of one or more attributes of the user including the face. For example, attributes of faces may include: skin tone/color, hair length (short, long, medium, bald), hair color (black, dark brown, light brown, auburn, orange, strawberry blonde, dirty blonde, bleached blonde, grey and white), hair texture (straight, wavy, curly, coily), age, gender, eye shape”).
Regarding claim 9, Sarna as modified by Wang discloses the method defined in Claim 1, the model training server further includes a similarity computing to compute a degree of model similarity between those avatar models in the same group (Sarna: para 66, “the method 500 mixes 514 the clean smaller set of images (positive images) with negative images to produce a base set of image. The method 500 splits 516 this base set of images into a training set and a testing set. Then the method 500 trains 518 the classifier of a shallow neural network using the training data. The method 500 tests 518 the performance of classifier using testing set. If the performance is satisfactory, the classifier is provided for use”).
Regarding claim 10, Sarna as modified by Wang discloses the method defined in Claim 1, those classifying labels are preset by a user behavior of the user, the user behavior collects interaction data on the platform related to the user through the application program (Sarna: para 78, “In situations in which certain implementations discussed herein may collect or use personal information about users (e.g., user data, information about a user's social network, user's location, user's biometric information, user's activities and demographic information)”).
Claim(s) 8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Sarna et al. (U.S. Pub. No. 20200175740) in view of Wang (CN Pub. No. 117689770) and in further view of Donnell et al. (U.S. Pub. No. 20240029330).
Regarding clam 8, Sarna as modified by Wang discloses the method defined in Claim 1, natural language to eliminate the unreasonable group.
However, in a similar field of endeavor, Donnell discloses those model parameters are analyzed by a natural language to eliminate the unreasonable group (para 31, “language processing module and/or diagnostic engine may generate the language processing model by any suitable method, including without limitation a natural language processing classification algorithm; language processing model may include a natural language process classification model that enumerates and/or derives statistical relationships between input terms and output terms”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sarna's invention of organizing and screening model parameters for avatar generation, as modified by Wang, with the features of Donnell's invention of a natural language processing classification model that derives statistical relationships between input terms and output terms. One of ordinary skill in the art would have been motivated to make this combination because Donnell expressly teaches a "natural language process classification model that enumerates and/or derives statistical relationships between input terms and output terms," and applying that natural language processing classification model to analyze Sarna's model parameters provides a principled, language-driven way to evaluate the relationships among the parameters and remove the parameter groups whose term relationships are inconsistent, thereby eliminating the unreasonable group and narrowing the parameters carried forward into avatar generation.
Response to Arguments
Applicant's arguments filed 05/15/2026 have been fully considered but they are not persuasive.
At the outset, Applicant's amendment to independent claim 1 to recite "a deep-learning text-to-image diffusion model is used to generate a plurality of avatar models" has been entered. Because Sarna does not disclose that limitation, the prior anticipation rejection of claims 1-3, 7, and 9-10 under §102 has been withdrawn, and those claims are now rejected under §103 over Sarna in view of Wang as set forth above. Wang is the same reference relied upon for the diffusion-model limitation of claims 4-6 in the prior Office action, and its qualification as prior art is unchanged.
Regarding Applicant's "avatar generation process" arguments at pages 7-8 of the Remarks (different source, different approach, different purpose, and unexpected effect), these arguments are directed to a single reference and to features that are not recited in the claims. Applicant argues that the present application pre-generates avatar models and stores them, whereas Sarna allegedly generates a cartoon on demand. This argument is not commensurate in scope with claim 1 and overlooks Sarna's actual disclosure. Sarna builds its avatar assets in advance: the method "uses the images to train a deep neural network 226," "cleans 510 the smaller set of images" and "crowd sources 512 the clean smaller set of images to produce positive images (tuples of image, attribute, label)," and the "cartoon asset module 206" includes "routines for storing, managing, and providing access to the individual sets of cartoon assets." Sarna thus teaches building and storing the avatar/asset library in advance and retrieving from it, which is the pre-generation and retrieval architecture Applicant relies upon. The only feature Sarna lacks, the deep-learning text-to-image diffusion model, is supplied by Wang.
Regarding Applicant's reliance at page 8 of the Remarks on storing "multiple tag parameter sets as a tag parameter set list", storing "multiple model parameters as a model parameter list", these features are not recited in claim 1. Claim 1 recites "a model parameter list" and "organizing those classifying labels into a label parameter group," which Sarna teaches under the broadest reasonable interpretation through its stored databases of parameters and its label-organized, cleaned image tuples. Arguments that rely on limitations appearing only in the specification, and not in the claims, cannot distinguish over the prior art, and one of ordinary skill in the art is free to read the claim terms on Sarna's corresponding stored data structures under the broadest reasonable interpretation.
Regarding Applicant's "unexpected effect" argument at page 8 of the Remarks (that the present invention achieves "rapid generation" and "real-time display" in under five minutes versus the prior art's five to ten minutes), this argument is unpersuasive for two independent reasons. First, the asserted time savings are not claimed; claim 1 recites no time threshold, no "real-time" limitation, and no comparative speed metric, so the argument is again not commensurate in scope with the claims. Second, an assertion of unexpected results must be supported by objective factual evidence, such as a declaration under 37 CFR 1.132 showing a difference in results and that the difference is unexpected; the present record contains only attorney argument, which is not evidence. Moreover, any speed advantage flows from pre-generating and retrieving stored avatars, which Sarna already performs by building its asset library in advance and retrieving stored assets at runtime.
Regarding Applicant's claim 6 argument at pages 8-9 of the Remarks (that Sarna is "retrieval-based" while Wang is "generative," and that combining them would render Sarna's cartoon asset module and template space mapping meaningless and would undermine Sarna's principle of operation), this argument is not persuasive. The rejection does not bodily incorporate Wang's architecture into Sarna or remove Sarna's retrieval scheme. Wang is relied upon only for its teaching that "a Diffusion model to generate a character model, wherein the Diffusion model is a deep learning text-to-image generation model for generating high-quality images." In the combination, Wang's diffusion model generates the avatar models and image files that populate Sarna's stored library, after which Sarna's existing scheme relates those models to their parameters, saves them to the database, and retrieves them by label matching. Sarna's retrieval architecture is preserved, not destroyed; only the source of the stored images changes from a pre-existing image data set to a generative model. The test for obviousness is not whether the references can be physically combined but what their combined teachings would have suggested to one of ordinary skill in the art, and Sarna does not criticize, discredit, or otherwise discourage generating its stored images with a generative model, so there is no teaching away. Combining Wang's known text-to-image generation technique with Sarna's known retrieval system yields the predictable result of a retrievable library of generated avatars.
Regarding Applicant's claim 8 argument at page 9 of the Remarks (that Donnell applies natural language processing to the "statistical relationship between input/output terms" while the present invention applies natural language processing to "logical conflict filtering of image generation parameters," with the basketball-uniform-and-baseball-bat example of specification paragraph [0031]), this argument is not commensurate in scope with claim 8. Claim 8 recites only that "those model parameters are analyzed by a natural language to eliminate the unreasonable group"; it does not recite "logical conflict filtering" or any specific incompatible-attribute example. Donnell's "natural language process classification model that enumerates and/or derives statistical relationships between input terms and output terms" reads on analyzing model parameters by a natural language to identify and remove an unreasonable group under the broadest reasonable interpretation.
Regarding Applicant's claim 9 argument at page 9 of the Remarks (that the claimed similarity calculation differs from Sarna's "classification accuracy testing" because its purpose is to "narrow the filtering scope" or "remove duplicates" per specification paragraph [0034]), this argument relies on an unclaimed purpose. Claim 9 recites only "a similarity computing to compute a degree of model similarity between those avatar models in the same group." Sarna's method that mixes positive and negative images into a base set, splits the set, and trains and tests a classifier on degrees of image correspondence teaches computing a degree of model similarity within a group under the broadest reasonable interpretation, and the unclaimed "narrow the filtering scope" purpose does not patentably distinguish the claim.
Regarding Applicant's claim 10 argument at page 9 of the Remarks (that Sarna involves broad data collection and does not teach automatically presetting classifying labels based on "interactive data" such as tracking points, click counts, and preference ratios per specification paragraph [0035]), this argument is again not commensurate in scope. Claim 10 recites that the classifying labels are preset by a user behavior that "collects interaction data on the platform related to the user through the application program"; it does not recite tracking points, click counts, or preference ratios. Sarna's disclosure of collecting personal information about users, including "user's activities" and "demographic information," teaches collecting interaction data related to the user under the broadest reasonable interpretation.
Finally, to the extent Applicant attacks Sarna individually for failing to disclose the deep-learning text-to-image diffusion model, that feature is supplied by Wang, and nonobviousness cannot be established by attacking references individually where the rejection is based on a combination of references.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 Jai Li whose telephone number is (571)272-1170. The examiner can normally be reached Mon-Thu between 06:00-16:00 EST.
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, Xiao Wu can be reached at (571)272-7761. 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.
/JAI W LI/Junior Examiner, Art Unit 2613
/XIAO M WU/Supervisory Patent Examiner, Art Unit 2613
Prosecution Insights