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 .
Priority
Applicant claims the benefit of US Provisional Application No. 63/583,380, filed 09/18/2023. Claims 1-20 have been afforded the benefit of this filing date.
Information Disclosure Statement
The information disclosure statements dated 07/29/2026, 09/09/2026, and 09/30/2026 have been considered and placed in the application file.
Response to Arguments
Applicant’s arguments, see pg. 8, filed 07/21/2026, with respect to the objection to claim 4 have been fully considered and are persuasive. The objection of claim 4 has been withdrawn.
Applicant’s arguments, see pg. 8-16, filed 07/21/2026, with respect to the rejection(s) of claim(s) 1 under 35 U.S.C. 103 have been fully considered and are persuasive; namely, argument I. stating that the cited references do not teach the limitation: “training, using the training data and the predicted image, the diffusion prior model by comparing the ground-truth vectorizable image and the predicted image”. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Meeks et al. (US 20240265204 A1).
Applicant’s remaining arguments regarding other aspects of the rejection of claim 1 under 35 U.S.C. have been fully considered but they are not persuasive.
Applicant’s argument II. states that the cited references do not teach the limitation: “generating, using the image generation model, a synthetic vectorizable image based on the text prompt by performing a reverse diffusion process based on an output of the trained diffusion prior model”. Furthermore, applicant states that Ramesh does not teach reducing “high frequency detail” in the synthetic vectorizable image.
In response to applicant's argument, the test for obviousness is not whether the features of a secondary reference may be bodily incorporated into the structure of the primary reference; nor is it that the claimed invention must be expressly suggested in any one or all of the references. Rather, the test is what the combined teachings of the references would have suggested to those of ordinary skill in the art. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981). In this case, the combined teachings of Ramesh in view of Lu teach the claimed limitations (see “Claim Rejections - 35 USC § 103” section).
Applicant’s argument III. states that the cited references do not teach a “vectorizable image”; and that the cited “flat design” of Lu is not equivalent to being vectorizable because it may not “enable efficient raster-to-vector format translation”. Applicant cites a portion of the specification as support: “Characteristics of vectorizable images may include, but are not limited to, flat or solid color regions, clearly defined shapes or boundaries, and the absence of gradient transitions or fuzzy edges.”
In response to applicant's argument, a recitation of the intended use of the claimed invention must result in a structural difference between the claimed invention and the prior art in order to patentably distinguish the claimed invention from the prior art. If the prior art structure is capable of performing the intended use, then it meets the claim.
In this case, the examples of “flat design” provided by Lu have all of these features, consisting of flat, simple solid colored shapes without gradients (see fig. 4). Regardless of whether they were intended to be convertible to vector images, the fact that they bear the characteristics of vectorizable images means that they may be considered vectorizable images.
Applicant's arguments filed 07/21/2026 with respect to the rejection of claim 8 under 35 U.S.C. 103 have been fully considered but they are not persuasive.
Applicant argues that the cited references do not teach the limitations: “generating, using an image generation model, a synthetic vectorizable image based on the text prompt by performing a reverse diffusion process based on the diffusion prior embedding, wherein the image generation model is trained to reduce high-frequency details”. Applicant cites details from the specification as support, including the structure in which the diffusion model takes in both the diffusion prior embedding and the text embedding as input, and an explanation of the upsampling model.
In response to applicant's argument that the references fail to show certain features of the invention, it is noted that the features upon which applicant relies (i.e., the structure of the diffusion model and the upsampling model) are not recited in the rejected claim(s). Although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993).
Therefore, the rejection of claim 8 under 35 U.S.C. 103 is maintained.
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 and 4-7 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ramesh et al. (US 11922550 B1; hereinafter "Ramesh") in view of Lu et al. ("Specialist Diffusion: Plug-and-Play Sample-Efficient Fine-Tuning of Text-to-Image Diffusion Models To Learn Any Unseen Style". CVPR 2023 Day 2 (21 Jun 2023); hereinafter "Lu") and Meeks et al. (US 20240265204 A1; hereinafter "Meeks").
Regarding claim 1, Ramesh teaches: A method of training a machine learning model, the method comprising:
obtaining training data including a ground-truth image and a text prompt describing the ground-truth image (col. 2 lines 37-43 “…collecting a first data set and a second data set from a database, wherein the first data set comprises a set of images, wherein the second data set comprises a set of text descriptions corresponding to the set of images, jointly training an image encoder on the first data set and the text encoder on the second data set…”);
encoding the text prompt to obtain a text embedding (col. 6 lines 42-47 “For example, image-caption pairs may be inputted into a jointly trained neural network. Images may be inputted into an image encoder, and the captions or text descriptions may be inputted into the text encoder. The image encoder may output an image embedding, and the text encoder may output a text embedding.”);
converting, using a diffusion prior model, the text embedding to obtain a diffusion prior embedding (col. 10 lines 32-57 “In some embodiments, the first sub-model is configured to encode, prior to generating the corresponding image embedding, at least one of the text description or the text embedding, via a transformer, as a sequence of tokens predicted autoregressively… In some embodiments, the image embedding caption and the text embedding may be converted into a sequence of discretized codes.”);
generating, using an image generation model, a predicted image with a first level of high frequency detail by performing a reverse diffusion process based on the diffusion prior embedding (col. 11 line 54 to col. 12 line 8 “For example, as referenced in FIG. 1B, a first sub-model 138 may be configured to generate a corresponding image embedding 140… In some embodiments, first sub-model 138, prior to generating corresponding image embedding 140, first sub-model 138 may include a diffusion prior model, which may include a transformer. The output of first sub-model may be the corresponding image embedding 140. For example, the diffusion prior or the autoregressive prior may generate the corresponding image embedding 140, as discussed herein. In some embodiments, corresponding image embedding 140 may be different from image embedding 135 resultant from training of image encoder 134. Corresponding image embedding 140 may be accessible to other models or to be used for image generation.”;
col. 11 lines 6-23 describes the reverse diffusion process); and
training, using the training data and the predicted image, the diffusion prior model (col. 11 lines 14-16 “The diffusion model may be trained to reverse the noising process and attempt to reconstruct the noisy image to its original form.”;
col. 11 lines 25-48 “In some embodiments, prior to generating the corresponding image embedding, the first sub-model may include training a diffusion model including a transformer. In some embodiments, the first sub-model diffusion model, which may be referred to as a diffusion prior, may involve a decoder-only transformer… For example, the diffusion prior may be trained on a sequence of tokens of the text description, text embeddings, an encoding for the diffusion timestep, the noise image inputted to the image encoder, and a final encoding. The output of the final encoding may be an output of the transformer, which may predict the image encoding during the reverse noising process. In some embodiments, the diffusion prior may be trained on the text description and the text embedding.”); and
generating, using the image generation model, a synthetic image based on the text prompt by performing a reverse diffusion process based on an output of the trained diffusion prior model (col. 13 lines 35-42 “As discussed herein, the diffusion model may use (e.g., may include) neural networks, such as convolutional neural networks, in the reverse process of generating images from noise. As such, the token embeddings may be included in the reverse process to provide guidance on generating the output image, thereby enabling improved text-conditional image generation.”).
Ramesh does not explicitly teach that the ground-truth image and the generated synthetic image are vectorizable, or wherein the synthetic vectorizable image has a second level of high frequency detail that is lower than the first level of high frequency detail of the predicted image.
Lu teaches using a training dataset of images that may be considered vectorizable in order to generate images in a similar style (Lu fig. 4 shows “flat design” samples; pg. 14271 section 4.1 “Data Preparation”: “We collect three few-shot style datasets, that are motivated by real-world applications. Specifically, each dataset is created or collected by our in-house graph designers and represents a unique artwork style that is popular for visual content creation and re-mixing among social media users, hence “personalizing” those styles to users’ own artwork makes appealing demands.”; in particular, the “flat design” dataset is described as “2D flat icon style that generally omits shadows and textures.”). If the image generation model of Ramesh were to be trained on these images, the final output would be more similar in style to the ground-truth images than the diffusion prior; therefore teaching the limitation: wherein the synthetic vectorizable image has a second level of high frequency detail that is lower than the first level of high frequency detail of the predicted image.
Ramesh and Lu are both analogous to the claimed invention because they are in the same field of image generation using a diffusion model. 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 the invention of Ramesh with the teachings of Lu to apply it to generating images in the flat design style taught by Lu in order to achieve a desired art style.
The combination of Ramesh in view of Lu is not relied upon to teach the limitation: training, using the training data and the predicted image, the diffusion prior model by comparing the ground-truth image and the predicted image.
Meeks teaches: training, using the training data and the predicted image, the diffusion prior model by comparing the ground-truth image and the predicted image ([0060] “For example, the custom content generation system 106 can cause the content generation model to generate an output image in response to a text description. Then, the custom content generation system 106 can compare the output image to a ground truth image corresponding to the text description (e.g., from the image and text paired training data) to determine a loss between the output image and the ground truth image (e.g., a cross entropy loss, mean square error loss, mean absolute error loss).”).
Meeks is analogous to the claimed invention because it is in the same field of image generation using a diffusion model. 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 the invention of Ramesh in view of Lu with the teachings of Meeks to train the diffusion prior model by comparing an input image from a training dataset with a denoised output image. This is a known concept in the art; the motivation would have been to produce an output image which is more similar to the training data.
Regarding claim 2, the combination of Ramesh in view of Lu and Meeks teaches: The method of claim 1, wherein obtaining the training data comprises:
obtaining a set of vectorizable images; and filtering the set of vectorizable images based on an aesthetic parameter to obtain the training data (Lu fig. 4 shows “flat design” samples; pg. 14271 section 4.1 “Data Preparation”: “We collect three few-shot style datasets, that are motivated by real-world applications. Specifically, each dataset is created or collected by our in-house graph designers and represents a unique artwork style that is popular for visual content creation and re-mixing among social media users, hence “personalizing” those styles to users’ own artwork makes appealing demands.”; in particular, the “flat design” dataset is described as “2D flat icon style that generally omits shadows and textures.”).
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 the invention of Ramesh in view of Lu and Meeks with the additional teachings of Lu to filter the input dataset in order to better achieve the desired art style.
Regarding claim 4, the combination of Ramesh in view of Lu and Meeks teaches: The method of claim 1, further comprising:
obtaining a pre-trained image generation model; and
training the pre-trained image generation model to generate vectorizable images using the training data (Ramesh col. 23 lines 44-52 “A machine learning model may be or include, without limitation, one or more of… a GPT (Generative Pre-trained Transformer) model (or other autoregressive model)”;
Lu pg. 14268 col. 1 “In this paper, we focus on fine-tuning a pre-trained text-to-image diffusion model, to learn a highly specialized unseen style, using a handful of images. Our goal is to have the fine-tuned model generate high-quality images of arbitrary (known) objects in this (previously unknown) style.”; see the “Flat design” style examples in fig. 1 and 5 for examples of generated vectorizable 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 the invention of Ramesh in view of Lu and Meeks with the additional teachings of Lu to use the pre-trained model mentioned by Ramesh to generate images in the “flat design” style of Lu.
Regarding claim 5, the combination of Ramesh in view of Lu and Meeks teaches: The method of claim 1, further comprising:
computing a diffusion loss based on the ground-truth vectorizable image (Ramesh col. 6 lines 32-37 “Embeddings may be compared to evaluate how well each embedding represents the input to the encoder. Evaluations or comparison may include implementations of optimization including but not limited to, one or more of loss functions, least squares, gradient descent, dot products, and cosine similarity.”); and
updating parameters of the image generation model based on the diffusion loss (Ramesh col. 6 lines 37-41 “A neural network model may iteratively update the encoders and embeddings to obtain representations wherein the embeddings more accurately represent their inputs or wherein the embeddings more accurately represent corresponding features between inputs.”).
Regarding claim 6, the combination of Ramesh in view of Lu and Meeks teaches: The method of claim 1, further comprising:
tuning an upsampling model based on the training data (Lu pg. 14268 col. 1 “In this paper, we focus on fine-tuning a pre-trained text-to-image diffusion model, to learn a highly specialized unseen style, using a handful of images.”; Ramesh col. 14 lines 5-7: “In some embodiments, the second sub-model may comprise up-sampling prior to generating the output image.”).
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 the invention of Ramesh in view of Lu and Meeks with the additional teachings of both references to perform upsampling prior to outputting the generated image. The motivation would have been that “higher resolution images may be desired, as they may provide increased quality, clarity, and aesthetic pleasure.” (Ramesh col. 14 lines 24-27).
Regarding claim 7, the combination of Ramesh in view of Lu and Meeks teaches: The method of claim 1, wherein the second level of high frequency detail corresponds to a level of high frequency detail in the ground-truth vectorizable image (Lu, the level of detail in the generated “flat design” images in fig. 1 and 5 is comparable to the level of detail in the “flat detail” example training images in fig. 4).
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 the invention of Ramesh in view of Lu and Meeks with the additional teachings of Lu to achieve the desired art style.
Claim(s) 3 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ramesh (US 11922550 B1) in view of Lu ("Specialist Diffusion: Plug-and-Play Sample-Efficient Fine-Tuning of Text-to-Image Diffusion Models To Learn Any Unseen Style") and Meeks (US 20240265204 A1) as applied to claim 1 above, and further in view of Bean (US 20240320867 A1).
Regarding claim 3, the combination of Ramesh in view of Lu and Meeks teaches: The method of claim 1, wherein obtaining the training data comprises:
obtaining a preliminary caption including a description of the ground-truth vectorizable image (Lu section 4.1 “Data Preparation”: “For all images, we use BLIP [12] to generate text captions and manually inspect their quality with artists. For example, the first example from the dataset “Fantasy” in Fig. 4 is captioned “boats floating in front of a cityscape”.”).
The combination of Ramesh in view of Lu and Meeks does not explicitly teach: appending a description of a vectorizable image category to the preliminary caption to obtain the text prompt.
Bean teaches: appending a description of a vectorizable image category to the preliminary caption to obtain the text prompt ([0044] “In some embodiments, the prompt classifier 148 categorizes and/or classifies the words in the textual description generated by the image-to-text generator 144 for selection by the prompt editor 146. For example, after generating a textual description for a generated image, the prompt classifier 148 may analyze the textual description to identify words, or combinations of words, that are associated with each category or classification described above. Subsequently, the prompt editor 146 may request words from one or more categories from the prompt classifier 148 for inclusion in the subsequent text prompt.”; see fig. 6 for one embodiment in which multiple categories are determined and a user may select which ones to include).
Bean is analogous to the claimed invention because it is in the same field of image generation using a diffusion model. 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 the invention of Lu in view of Ramesh and Meeks with the teachings of Bean to determine a category associated with a generated text prompt and append it to the prompt. The motivation would have been to help a user narrow and improve a text prompt.
Claim(s) 8, 10, 15-17, and 19-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ramesh et al. (US 11922550 B1; hereinafter "Ramesh") in view of Lu et al. ("Specialist Diffusion: Plug-and-Play Sample-Efficient Fine-Tuning of Text-to-Image Diffusion Models To Learn Any Unseen Style". CVPR 2023 Day 2 (21 Jun 2023); hereinafter "Lu") and Ma et al. ("Towards Layer-wise Image Vectorization". arXiv preprint (09 Jun 2022). https://arxiv.org/abs/2206.04655; hereinafter "Ma").
Regarding claim 8, Ramesh teaches: A non-transitory computer readable medium storing code for image processing, the code comprising instructions that, when executed by at least one processor, causes the at least one processor to perform operations (col. 2 lines 12-16 “For example, in an embodiment, a non-transitory computer-readable medium may include instructions that are executable by one or more processors to perform operations for generating an image from a text description.”) comprising:
obtaining a text prompt describing an image element (col. 9 lines 33-34 references “generating an image corresponding to a text input”);
encoding the text prompt to obtain a text embedding (col. 6 lines 42-47 “For example, image-caption pairs may be inputted into a jointly trained neural network. Images may be inputted into an image encoder, and the captions or text descriptions may be inputted into the text encoder. The image encoder may output an image embedding, and the text encoder may output a text embedding.”);
converting, using a diffusion prior model, the text embedding to obtain a diffusion prior embedding (col. 10 lines 32-57 “In some embodiments, the first sub-model is configured to encode, prior to generating the corresponding image embedding, at least one of the text description or the text embedding, via a transformer, as a sequence of tokens predicted autoregressively… In some embodiments, the image embedding caption and the text embedding may be converted into a sequence of discretized codes.”);
generating, using an image generation model, a synthetic image based on the text prompt by performing a reverse diffusion process based on the diffusion prior embedding (col. 11 line 54 to col. 12 line 8 “For example, as referenced in FIG. 1B, a first sub-model 138 may be configured to generate a corresponding image embedding 140… In some embodiments, first sub-model 138, prior to generating corresponding image embedding 140, first sub-model 138 may include a diffusion prior model, which may include a transformer. The output of first sub-model may be the corresponding image embedding 140. For example, the diffusion prior or the autoregressive prior may generate the corresponding image embedding 140, as discussed herein. In some embodiments, corresponding image embedding 140 may be different from image embedding 135 resultant from training of image encoder 134. Corresponding image embedding 140 may be accessible to other models or to be used for image generation.”;
col. 11 lines 6-23 describes the reverse diffusion process).
Ramesh does not explicitly teach: generating, using an image generation model, a synthetic vectorizable image based on the text prompt, wherein the image generation model is trained to reduce high-frequency details.
Lu teaches: generating, using an image generation model, a synthetic vectorizable image based on the text prompt, wherein the image generation model is trained to reduce high-frequency details (fig. 1 and 5 examples with “Flat design” style selected; this style is described in section 4.1 “Data preparation”: “2D flat icon style that generally omits shadows and textures.”).
Ramesh and Lu are both analogous to the claimed invention because they are in the same field of image generation using a diffusion model. 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 the invention of Ramesh with the teachings of Lu to apply it to generating images in the flat design style taught by Lu in order to achieve a desired art style.
The combination of Ramesh in view of Lu does not explicitly teach: generating a vector image based on the synthetic vectorizable image, wherein the vector image includes the image element described by the text prompt.
Ma teaches: generating a vector image based on the synthetic vectorizable image, wherein the vector image includes the image element described by the text prompt (fig. 1 shows conversion of raster input image to vector image, where the content of the image is preserved – if the raster input contains the image element described the text prompt, then so will the vector output).
Ma is analogous to the claimed invention because it is in the same field of image vectorization. Furthermore, one of ordinary skill in the art would have recognized that the “flat design” style of Lu is well suited for vectorization. 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 the invention of Ramesh in view of Lu with the teachings of Ma to vectorize the flat-styled, vectorizable output images of Lu. The motivation would have been to allow users to edit the generated images, as suggested by Ma (abstract and section 2.2).
Regarding claim 10, the combination of Ramesh in view of Lu and Ma teaches: The non-transitory computer readable medium of claim 8, wherein generating the synthetic vectorizable image comprises:
upsampling the synthetic vectorizable image to obtain an upsampled image, wherein the vector image is based on the upsampled image (Ramesh col. 14 lines 5-7: “In some embodiments, the second sub-model may comprise up-sampling prior to generating the output image.”).
Regarding claim 15, it is rejected using the same references, rationale, and motivation to combine as claim 8, with the additional limitation of: A system comprising:
a memory component; and a processing device coupled to the memory component (Ramesh col. 5 lines 15-19 “In some embodiments, a system may include at least one memory storing instructions and at least one processor configured to execute the instructions to perform operations for generating an image corresponding to a text input.”).
Regarding claim 16, the combination of Ramesh in view of Lu and Ma teaches: The system of claim 15, further comprising: a text encoder comprising a transformer architecture (Ramesh fig. 1B text encoder 132; col. 11 lines 59-64 “The prior model may be configured to encode text description 130 or text embedding 136 via a transformer. In some embodiments, first sub-model 138 may comprise an autoregressive prior. For example, the autoregressive prior may encode text description 130 or text embedding 136 via a transformer.”).
Regarding claim 17, the combination of Ramesh in view of Lu and Ma teaches: The system of claim 15, wherein: the image generation model comprises a diffusion model (Ramesh col. 12 lines 31-48 “In some embodiments, the second sub-model may be configured to generate, based on at least one of the text description or the corresponding image embedding, an output image… In some embodiments, the second sub-model may comprise generating a diffusion model.”).
Regarding claim 19, the combination of Ramesh in view of Lu and Ma teaches: The system of claim 15, wherein: the image generation model comprises an upsampling model (Ramesh col. 14 lines 5-7: “In some embodiments, the second sub-model may comprise up-sampling prior to generating the output image.”).
Regarding claim 20, the combination of Ramesh in view of Lu and Ma teaches: The system of claim 15, further comprising: a vectorization component configured to transform pixel data to vector data (Ma fig. 1 shows conversion of raster input image to vector image).
The motivation to combine the invention of Ramesh in view of Lu with the teachings of Ma would have been the same as discussed for claim 8.
Claim(s) 11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ramesh (US 11922550 B1) in view of Lu ("Specialist Diffusion: Plug-and-Play Sample-Efficient Fine-Tuning of Text-to-Image Diffusion Models To Learn Any Unseen Style") and Ma ("Towards Layer-wise Image Vectorization") as applied to claim 10 above, and further in view of Schlemper et al. (US 20200034998 A1; hereinafter "Schlemper").
Regarding claim 11, the combination of Ramesh in view of Lu and Ma teaches: The non-transitory computer readable medium of claim 10, but does not explicitly teach: wherein: the upsampling removes an artifact that interferes with vectorization.
Schlemper teaches: wherein: the upsampling removes an artifact ([0107] “This structure of down-sampling followed by up-sampling allows operations to be performed at different resolutions, which helps the neural network model to capture both local and global features. In turn, this helps to eliminate image artifacts that may result from under-sampling in the spatial frequency domain.”). One of ordinary skill in the art would recognize that vector images are benefited by a certain degree of simplicity in order to be represented mathematically, as suggested by Ma; therefore, artifacts may interfere with vectorization.
Schlemper is analogous to the claimed invention because it is in the same field of image generation using a neural network. 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 the invention of Ramesh in view of Lu and Ma to use the upsampling of Ramesh in order to remove artifacts. The motivation would have been to improve the appearance of the image.
Claim(s) 12-14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ramesh (US 11922550 B1) in view of Lu ("Specialist Diffusion: Plug-and-Play Sample-Efficient Fine-Tuning of Text-to-Image Diffusion Models To Learn Any Unseen Style") and Ma ("Towards Layer-wise Image Vectorization") as applied to claim 8 above, and further in view of Bean (US 20240320867 A1).
Regarding claim 12, the combination of Ramesh in view of Lu and Ma teaches: The non-transitory computer readable medium of claim 8, but does not explicitly teach: wherein: the text prompt includes a vector image category.
Bean teaches: wherein: the text prompt includes a vector image category (figs. 6 and 7 show examples of image categories which can be inserted into a prompt).
Bean is analogous to the claimed invention because it is in the same field of image generation using a diffusion model. 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 the invention of Ramesh in view of Lu and Ma with the teachings of Bean to determine a category associated with a user’s text prompt and append it to the prompt. The motivation would have been to help a user narrow and improve a text prompt.
Regarding claim 13, the combination of Ramesh in view of Lu and Ma and further in view of Bean teaches: The non-transitory computer readable medium of claim 8, wherein obtaining the text prompt comprises:
obtaining an initial prompt from a user; and modifying the initial prompt based on a vector image category to obtain the text prompt (fig. 6; user’s initial text prompt can be modified via recommended, user-selectable categories).
The motivation to combine the invention of Ramesh in view of Lu and Ma with the teachings of Bean would have been the same as discussed for claim 12.
Regarding claim 14, the combination of Ramesh in view of Lu and Ma and further in view of Bean teaches: The non-transitory computer readable medium of claim 13, wherein obtaining the text prompt further comprises:
obtaining a category selection input via a user interface, wherein the vector image category is based on the category selection input (fig. 6; user’s initial text prompt can be modified by selecting recommended categories via a displayed menu).
The motivation to combine the invention of Ramesh in view of Lu and Ma with the teachings of Bean would have been the same as discussed for claim 12.
References Cited
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Shu et al. ("GAN-Based Multi-Style Photo Cartoonization". IEEE Transactions on Visualization and Computer Graphics, vol. 28 no. 10 (01 Oct 2022), pp. 3376-3390. https://ieeexplore.ieee.org/abstract/document/9382902) teaches a method of using a neural network to “cartoonize” an image, which results in an output that may be considered vectorizable.
Schaldenbrand et al. ("StyleCLIPDraw: Coupling Content and Style in Text-to-Drawing Translation". arXiv preprint (24 Feb 2022). https://arxiv.org/abs/2202.12362v1) teaches a variation on the invention of Ramesh in which the image generation model may also be conditioned on a style image input.
Song et al. (US 20240046529 A1) teaches a method of using a neural network to generate an image based on a user’s text input and a selected category.
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 BENJAMIN STATZ whose telephone number is (571)272-6654. The examiner can normally be reached Mon-Fri 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, Tammy Goddard can be reached at (571)272-7773. 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.
/BENJAMIN TOM STATZ/Examiner, Art Unit 2611
/TAMMY GODDARD/Supervisory Patent Examiner, Art Unit 2611
