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 Arguments
Applicant's arguments filed 10/29/25 have been fully considered but they are not persuasive.
Applicant’s arguments with respect to claim(s) 7 and 17 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 7-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 7 recites "the plurality of candidate images" in 2nd limitation. There is insufficient antecedent basis for this limitation in the claim.
Claim 17 recites "the plurality of candidate images" in 3rd limitation. There is insufficient antecedent basis for this limitation in the claim.
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 7 is rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2025/0005901) in view of Wen et al. in view of Lin et al. (US 2021/0271707).
Regarding claim 7, Cha discloses a method (Cha, [0007], “method can improve the performance of the generative AI by providing a way to automatically determine how closely the outputted synthesized digital images represent real-world content”):
obtaining training data (Cha, [0060], “database 710 may store data that may be retrieved by other components for system 700, such as realism scoring guidance, training data, and other features that can be referenced by the generative AI and/or the generative AI image realism assessment system”) including a text prompt describing an element and an image (Cha, [0027], “For generative text-to-image models such as DALL-E or Stable Diffusion, this means that the model generates images conditional on a text description known as a “prompt””. Stable diffusion is trained on a dataset of image-text pairs, where each text prompt describes the contents of the corresponding image);
training an image generation model to generate synthetic images including the element based on the training data (Cha, [0031], “the user submitted a prompt to an AI-based image generation engine (e.g., Midjourney, etc.) to produce a festive-styled representation of the city of Paris, France. In response to the user's prompt, the engine generates a first picture 120 shown in FIG. 1A”);
Cha does not expressly disclose “wherein the text prompt is selected based on having a length less than a threshold length”;
Wen discloses a text prompt is selected based on having a length less than a threshold length (Wen, 4.5 Prompt Distillation, [0001], “where we can use our prompt optimization method is prompt distillation, reducing the length of prompts while preserving their capability…such as the CLIP model, which has a maximum input length of 77 tokens”. In addition, in the same section, [0002], “we show images generated by the original prompts and the distilled prompts with four different distillation ratios: 0.7, 0.5, 0.3, 0.1. We see here that even with only 3 or 4 tokens, the hard prompts can still generate images very similar in concept to the original”. Stable diffusion is trained to generate images using training data including text prompts below a token length).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to generate Cha’s synthetic image using Wen’s prompt optimization method is prompt distillation. The motivation for doing so would have been reducing prompt length while maintaining descriptive capability.
Cha as modified by Wen does not expressly disclose “generating a semantic similarity score for each of the plurality of candidate images”;
Lin et al. (hereinafter Lin) discloses generating a semantic similarity score for each of a plurality of candidate images (Lin, [0074], “the joint embedding model 112 can determine a similarity score based on visual-semantic embedding generated during training for similar training images…the joint embedding model 112 may retrieve candidate image results based on similarity scores related to a vibrancy of colors within the image, a camera angle associated with the image, a brightness of the red tulips in the foreground, a focal point, an amount of visual distortion, the presence of an undesirable artifact, an amount of sunlight depicted in background blue sky, a number of clouds within the image, an overall percentage of the image that is either red or blue, etc”).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the concept of Lin’s semantic similarity score to condition the training of Cha’s generative AI model. The motivation for doing so would have been improving alignment between text and generated images.
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2025/0005901) in view of Wen et al. in view of Lin et al. (US 2021/0271707), as applied to claim 7, in further view of Brade et al. (Promptify: Text-to-Image Generation through Interactive Prompt Exploration with Large Language Models, Human-Computer Interaction, Apr, 2023).
Regarding claim 8, Cha as modified by Wen and Lin does not expressly disclose “an additional element that is not described by the text prompt”;
Brade et al. (hereinafter Brade) discloses an additional element that is not described by a text prompt (Brade, 5.1.1 Subject ideation, [0001], “"change the setting to Japan.". The system will then generate a new prompt that incorporates the user’s instruction, such as "Lion standing majestically by a cherry blossom tree with Mount Fuji in the background."”);
generate the synthetic images to include the additional element based on text prompts describing the element (Brade, Fig. 3 illustrates generate the synthetic images to include the additional element based on text prompts describing the element).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to use Brade’s suggestions to expand Wen’s text prompt. The motivation for doing so would have been allowing more detailed features to be generated, increase the change of producing better images.
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2025/0005901) in view of Wen et al. in view of Lin et al. (US 2021/0271707), as applied to claim 7, in further view of Mokady et al. (Null-text Inversion for Editing Real Images using Guided Diffusion Models, Computer Vision and Pattern Recognition, Nov 2022).
Regarding claim 9, Cha as modified by Wen and Lin does not expressly disclose “obtaining a caption for the image”;
Mokady et al. (hereinafter Mokady) discloses obtaining a caption for an image (Mokady, Fig. 2 illustrates input caption);
removing one or more words from the caption to obtain a text prompt (Mokady, Fig. 2 illustrates modified caption including removing one or more words from the caption to obtain edited text prompt).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to use the concept of Mokady’s prompt to prompt editing to generate Cha’s training data. The motivation for doing so would have been allowing prompt editing to enable image modifications through text.
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2025/0005901) in view of Wen et al. in view of Lin et al. (US 2021/0271707), as applied to claim 7, in further view of Sivakumar et al. (US 2023/0169147).
Regarding claim 11, Cha discloses performing an image search based on the text prompt to obtain a plurality of candidate images (Cha, [0057], “The realism assessment systems can be used to carefully cull candidate images that stray from the branding themes while selecting those candidate images that stay true to (i.e., “are realistic) the ground truth the company is looking to promote”);
Cha as modified by Wen and Lin does not expressly disclose “the training data is based on the plurality of candidate images”;
Sivakumar et al. (hereinafter Sivakumar) discloses training data is based on candidate data (Sivakumar, [0085], “Training data used for retraining the machine learning model may be referred to herein as retraining data…the training module 408 evaluates candidate retraining data before retraining the machine learning model to determine whether the machine learning model would “learn” anything new from the candidate retraining data”).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to generate Cha’s training data using the concept of Sivakumar’s retraining the machine learning model. The motivation for doing so would have been improving the image generation model’s ability to interpret and respond to text parameters.
Claims 13 are rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2025/0005901) in view of Wen et al. in view of Lin et al. (US 2021/0271707) in view of Sivakumar et al. (US 2023/0169147), as applied to claim 11, in view of Liang et al. (US 2024/0028634).
Regarding claim 13, Cha as modified by Wen, Lin and Sivakumar teaches the training data is based on candidate data; Cha as modified by Wen, Lin and Sivakumar does not expressly disclose “generating an aesthetic score for each of the plurality of candidate images”;
Liang et al. (hereinafter Liang) discloses generating an aesthetic score for each of a plurality of candidate images (Liang, [0057], “generate an aesthetic score for the remaining candidate images. The machine-learned model can be trained such that it gives high aesthetic scores for images that are likely to be of interest to users. For example, images that are in focus and frame interesting action, important people, important information, and so on can receive higher aesthetic scores”).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to use the concept of Liang’s aesthetic scoring of candidate images to the training of Cha’s generative AI model. The motivation for doing so would have been improving the quality and visual appeal of the generated output.
Claim 14 is rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2025/0005901) in view of Wen et al. in view of Lin et al. (US 2021/0271707), as applied to claim 7, in further view of Chen et al. (PIXART-α: FAST TRAINING OF DIFFUSION TRANS FORMER FOR PHOTOREALISTIC TEXT-TO-IMAGE SYNTHESIS, Sep, 2023).
Regarding claim 14, Cha discloses modified by Wen with the same motivation from claim 7 discloses an additional image corresponding to the long text prompt (Wen, Fig. 6) and the threshold length (Wen, 4.5 Prompt Distillation, [0001], “where we can use our prompt optimization method is prompt distillation, reducing the length of prompts while preserving their capability…such as the CLIP model, which has a maximum input length of 77 tokens”);
Cha as modified by Wen and Lin does not expressly disclose “a length greater than the threshold length”;
Chen et al. (hereinafter Chen) discloses a length of text prompt is greater than the threshold length (Chen, 3.1 Implementation details, [0001], “Unlike previous works that extract a standard and fixed 77 text tokens, we adjust the length of extracted text tokens to 120, as the caption curated in PIXART-α is much denser to provide more fine-grained details”).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to use the concept of Chen’s length of text token to expand Wen’s token length. The motivation for doing so would have been allowing the model to preserve a greater portion of the prompt’s structure and semantic content.
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2025/0005901) in view of Wen et al. in view of Lin et al. (US 2021/0271707), as applied to claim 7, in further view of Liu et al. (US 2024/0282016).
Regarding claim 16, Cha discloses generating a predicted image using the image generation model (Cha, [0030], “the generative AI can be used to generate a plurality of synthetic digital images based on the user's prompt”);
Cha as modified by Wen and Lin does not expressly disclose “computing a loss function”;
Liu et al. (hereinafter Liu) discloses computing a loss function based on training image and predicted synthesized image (Liu, [0043], “making iterative adjustments based at least on minimizing a loss function between each training image and each predicted synthesized image to train the diffusion model”).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to train Cha’s diffusion model using the concept of Liu’s making iterative adjustments based at least on minimizing a loss function. The motivation for doing so would have been enabling estimation of the differences between the model’s prediction and the ground truth.
Claims 17-20 are rejected under 35 U.S.C. 103 as being unpatentable over Cha et al. (US 2025/0005901) in view of Wen et al. in view of Liang et al. (US 2024/0028634).
Regarding claim 17, Cha discloses an apparatus (Cha, [0012], “the system comprising one or more computers and one or more storage devices storing instructions that may be operable, when executed by the one or more computers”);
at least one processor (Cha, [0059], “a processor 712”);
at least one memory storing instructions and in electronic communication with the at least one processor (Cha, [0068], “a non-transitory computer-readable medium (CRM) storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to perform the disclosed methods”); and
an image generation model comprising parameters stored in the at least one memory (Cha, [0027], “For generative text-to-image models such as DALL-E or Stable Diffusion, this means that the model generates images conditional on a text description known as a “prompt””. In addition, in paragraph [0060], “database 710 may store data that may be retrieved by other components for system 700, such as realism scoring guidance, training data, and other features that can be referenced by the generative AI and/or the generative AI image realism assessment system”) and trained to generate synthetic images (Cha, [0030], “the generative AI can be used to generate a plurality of synthetic digital images based on the user's prompt”) including an element based on training data (Cha, [0043], “generation of the first image 420 by the AI engine (e.g., “The image showcases a vibrant and bustling cityscape that represents Paris, France...”)”. In addition, in paragraph [0054], “This process can includes labeling an image with English (or other language) keywords with the help of datasets provided during model training”) including a text prompt describing the element (Cha, Fig. 4 illustrates user text prompt 410) and an image including the element described in the text prompt (Cha, Fig. 4 illustrates image 420 including the element described in the text prompt);
Cha does not expressly disclose “the text prompt is selected based on having a length less than a threshold length”
Cha as modified by Wen with the same motivation from claim 7 discloses a text prompt is selected based on having a length less than a threshold length (Wen, 4.5 Prompt Distillation, [0001], “where we can use our prompt optimization method is prompt distillation, reducing the length of prompts while preserving their capability…such as the CLIP model, which has a maximum input length of 77 tokens”. In addition, in the same section, [0002], “we show images generated by the original prompts and the distilled prompts with four different distillation ratios: 0.7, 0.5, 0.3, 0.1. We see here that even with only 3 or 4 tokens, the hard prompts can still generate images very similar in concept to the original”. Stable diffusion is trained to generate images using training data including text prompts below a token length).
Cha as modified by Wen does not expressly disclose “the training data is selected by generating an aesthetic score for each of the plurality of candidate images”;
Liang et al. (hereinafter Liang) discloses generating an aesthetic score for each of a plurality of candidate images (Liang, [0057], “generate an aesthetic score for the remaining candidate images. The machine-learned model can be trained such that it gives high aesthetic scores for images that are likely to be of interest to users. For example, images that are in focus and frame interesting action, important people, important information, and so on can receive higher aesthetic scores”).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to use the concept of Liang’s aesthetic scoring of candidate images to the training of Cha’s generative AI model. The motivation for doing so would have been improving the quality and visual appeal of the generated output.
Regarding claim 18, Cha discloses parameters stored in the at least one memory (Cha, [0027], “For generative text-to-image models such as DALL-E or Stable Diffusion, this means that the model generates images conditional on a text description known as a “prompt””. In addition, in paragraph [0060], “database 710 may store data that may be retrieved by other components for system 700, such as realism scoring guidance, training data, and other features that can be referenced by the generative AI and/or the generative AI image realism assessment system”);
Cha as modified by Wen with the same motivation from claim 17 discloses a text encoder (Wen, 4. Prompt inversion with CLIP, [0003], “Formally, given a text encoder function f and an image encoder function g”);
And Cha as modified by Wen with the same motivation from claim 17 discloses trained to encode the text prompt to obtain a text embedding (Wen, 3. Methodology, Algorithm 1 illustrates text embedding).
Regarding claim 19, Cha discloses the image generation model is a diffusion model (Cha, [0027], “For generative text-to-image models such as DALL-E or Stable Diffusion, this means that the model generates images conditional on a text description known as a “prompt””).
Regarding claim 20, Cha discloses a data preparation component (Cha, [0036], “the AI engine can then create images based on the prompt”) comprising parameters stored in the at least one memory (Cha, [0027], “For generative text-to-image models such as DALL-E or Stable Diffusion, this means that the model generates images conditional on a text description known as a “prompt””. In addition, in paragraph [0060], “database 710 may store data that may be retrieved by other components for system 700, such as realism scoring guidance, training data, and other features that can be referenced by the generative AI and/or the generative AI image realism assessment system”) and trained to obtain training data ([0060], “database 710 may store data that may be retrieved by other components for system 700, such as realism scoring guidance, training data, and other features that can be referenced by the generative AI and/or the generative AI image realism assessment system”) including a plurality of text prompts and a plurality of images (Cha, [0027], “For generative text-to-image models such as DALL-E or Stable Diffusion, this means that the model generates images conditional on a text description known as a “prompt””. Stable diffusion is trained on a dataset of image-text pairs (i.e., a plurality of text prompts and a plurality of images)).
Allowable Subject Matter
Claim 1-6 are allowed.
Claims 10, 12 and 15 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
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 KYLE ZHAI whose telephone number is (571)270-3740. The examiner can normally be reached 9AM-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, Ke Xiao can be reached at (571) 272 - 7776. 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.
/KYLE ZHAI/Primary Examiner, Art Unit 2612