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 .
Withdrawn of Last Office action
1. Applicant’s arguments, filed 12/02/2025 with respect to the rejection of claim under 35 U.S.C. § 102(a)(2) have been fully considered and are persuasive. The rejection of claims 1, 9, and 16 under 35 U.S.C. § 102(a)(2) has been withdrawn.
Claim Rejections - 35 USC § 103
2. In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
3. The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
4. Claim 1-10, and 14-20 are rejected under 35 U.S.C. 103 as being unpatentable over Karpman et al., US 11995803 B1, and further in view of Wang US 20240242402 A1.
5. As per claim 1, Karpman discloses: A method comprising:
obtaining an input prompt describing an image element; (Karpman, Column 5, lines 65-66, “Receive a text prompt (e.g., image description) from the communication interface 122;”) and
identifying the training image as a source of the output image based on the watermark. (Karpman, column 7, lines 19-23, By executing the watermark detection model on this training set through suitable training techniques, the watermark detection model is configured to identify (e.g., via a confidence score or similar output) images as containing or not containing a digital watermark.”)
6. Karpman doesn’t expressly discloses:
generating, using an image generation model, an output image depicting the image element and including a watermark, wherein the image generation model is trained using a training image including the image element and the watermark;
7. Wang discloses: generating, using an image generation model, an output image depicting the image element and including a watermark, wherein the image generation model is trained using a training image including the image element and the watermark; (Wang, [0026], “ the present disclosure, as an example, audio-based image control module 103 may be a deep learning model based on CLIP (“Contrastive Language-Image Pre-training”)”, and [0047] In FIG. 3A, which illustrates process portion 300a, text encoder 302 may firstly encode text 301 such as “the band is playing,” so as to obtain text features 303 in the text. Meanwhile, image encoder 305 may extract image features 306 of image 304. Then, the CLIP model can project text features 303 and image features 306 into the same shared embedded space.”
8. Karpman is analogous art with respect to Wang because they are from the same field of endeavor, namely image processing. Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to include the process of generating, using an image generation model, an output image depicting the image element and including a watermark, wherein the image generation model is trained using a training image including the image element and the watermark; as taught by Wang into the teaching of Karpman. The suggestion for doing so would enhances the modality consistency. Therefore, it would have been obvious to combine Wang with Karpman.
9. As per claim 2, Karpman in view of Wang discloses: The method of claim 1, wherein generating the output image comprises: generating, using a generator of the image generation model, a latent code representing the input prompt and the watermark; and decoding, using a decoder of the image generation model, the latent code to obtain the output image. (Karpman, Column 3, lines 5-14, “Base image diffusion model 120 may operate in the latent space to interpret the text embedding. High-resolution diffusion models 116 may operate in the pixel space, as opposed to the latent space, which avoids the issues associated with image encoding and decoding, and improves the generation of images, such as faces, text, and other details. Although multiple models are described, models may be used to generate the full-size image from the embedding without upscaling.”)
10. As per claim 3, Karpman in view of Wang discloses: The method of claim 2, wherein generating the latent code comprises: performing a latent diffusion process. (Karpman, Column 3, lines 5-10,” Base image diffusion model 120 may operate in the latent space to interpret the text embedding. High-resolution diffusion models 116 may operate in the pixel space, as opposed to the latent space, which avoids the issues associated with image encoding and decoding.”)
11. As per claim 4, Karpman in view of Wang discloses: The method of claim 2, wherein: the decoder is fixed during a training stage in which the generator is trained using the training image. (Karpman, Column 11, lines 62-67, ”At Block M130, the method generates a final training corpus by executing a multimodal encoder-decoder 126 on the initial training corpus to process text captions, such as to (i) generate text captions describing each image in the set of training images and (ii) identify and remove misaligned text captions associated with images in initial training corpus.”)
12. As per claim 5, Karpman in view of Wang discloses: The method of claim 1, further comprising: determining that the output image is attributable to the training image from a plurality of images in a training set. ( Karpman, column 7, lines 7-23, “More specifically, the watermark detection model includes a pre-trained deep learning network (e.g., a deep vision classifier, a convolutional neural network, or a residual learning network such as ResNet) or other suitable machine learning model trained on aggregated (human) judgments of image contents. For example, the watermark detection model can be a visual classifier trained on a large set of images (e.g., thousands of images, millions of images) that include human-generated labels specifying whether the image contains a visible watermark (e.g., a logo or graphic identifying ownership of the image) or similar digital overlay. By executing the watermark detection model on this training set through suitable training techniques, the watermark detection model is configured to identify (e.g., via a confidence score or similar output) images as containing or not containing a digital watermark.”)
13. As per claim 6, Karpman in view of Wang discloses: The method of claim 1, wherein: the watermark is located in a pre-determined region of the output image, wherein each of a plurality of watermarks corresponds to a plurality of pre-determined regions, respectively. ( (Karpman, column 7, lines 7-23, “ (30) In some implementations, a watermark detection model is configured to classify images as containing overlay content, such as logos, watermarks, and similar digital overlays. Overlay content may be content that is added to main content, such as a video.”)
14. As per claim 7, Karpman in view of Wang discloses: The method of claim 6, wherein: the plurality of pre-determined regions are non-overlapping. (Karpman, column 7, lines 23-32, “As described in more detail below, the system can therefore execute and/or access outputs of the watermark detection model on images in order to identify, remove, and/or exclude images containing digital watermarks or similar overlays from the set of images used in training the text-to-image diffusion model 112, thereby substantially reducing (or eliminating) the likelihood that the text-to-image diffusion model 112 will inadvertently reproduce similar digital watermarks within images it generates during operation.”)
15. As per claim 8, Karpman in view of Wang discloses: The method of claim 1, further comprising: obtaining a noise input, wherein the output image is generated based on the noise input. (Karpman, column 5, lines 25-36, “ The base image diffusion model 120 can therefore: receive one or more text embeddings from the set of pre-trained text encoders 118; receive and/or initialize a (randomly sampled) noise distribution at a preset resolution (e.g., 64 pixels by 64 pixels); and transform the noise distribution into a base image at the preset resolution based on the one or more text embeddings and parameters, weights, and/or paths corresponding to an iterative denoising process learned by the base image diffusion model 120 during training. The system can then pass the base image to the set of high-resolution diffusion models 116 for upsampling and output.)
16. Claims 9, and 16 which are similar in scope to claim 1, thus rejected under the same rationale.
17. Claim 10, which is similar in scope to claim 6, thus rejected under the same rationale.
18. Claims 14, and 18 which are similar in scope to claim 4, thus rejected under the same rationale.
19. As per claim 15, Karpman in view of Wang discloses: The method of claim 9, wherein: a generator of the image generation model is pre-trained prior to training. (Karpman, column 2, lines 41-50, “The information may include images and/or text accessed from by public or private networks (e.g., the internet) by the web intelligence engine 108 and/or pre-defined training or evaluation datasets. Server system 102 generally includes and/or interfaces with a model training management system 128 configured to access and transmit content data from the set of storage devices no to the text-to-image diffusion model 112 during pre-training, training, and fine-tuning stages as described below.”)
20. Claim 17, which is similar in scope to claim 3, thus rejected under the same rationale.
21. Claim 19, which is similar in scope to claim 5, thus rejected under the same rationale.
22. As per claim 20, Karpman in view of Wang discloses: The apparatus of claim 16, further comprising: a training component configured to perform the training. (Karpman, column 6, lines 42-45, “The images retrieved by web intelligence engine 108 may be processed by models 106 to improve the training process. The composition of the training set may be important to train text-to-image diffusion model 112.”)
23. Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Karpman et al., US 11995803 B1, and in view of Wang US 20240242402 A1, and further in view of Seroussi US 2006/0023880 A1.
24. As per claim 11, Karpman in view of Wang discloses: The method of claim 9, (See rejection of claim 9 above).
25. Karpman in view of Wang doesn’t expressly disclose: creating the training set comprises: selecting a plurality of secrets; and generating a plurality of watermarks based on the plurality of secrets, respectively.
26. Seroussi discloses: selecting a plurality of secrets; and generating a plurality of watermarks based on the plurality of secrets, respectively. (Seroussi, [0033], “Secret data key 208 is used to keep the watermark values less readily detected by unauthorized users. In one implementation, secret data key 208 is a complete keying sequence corresponding to the end or run locations in an image suitable for watermarking. Alternatively, secret data key 208 can instead be used as a seed for a pseudo-random generator that then creates the keying sequence needed for watermarking. In either of these or other implementations, the keying sequence associated with the secret data key 208 has a keying bit in the sequence set to a `1` value to indicate the corresponding end or run includes a watermark bit value and has a keying bit in the sequence set to a `0` value to indicate the corresponding end or run in the image does not includes a watermark value.”)
27. Karpman in view of Wang is analogous art with respect to Seroussi because they are from the same field of endeavor, namely image processing. Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to include the process of installing the processes of creating the training set comprises: selecting a plurality of secrets; and generating a plurality of watermarks based on the plurality of secrets, respectively: as taught by Seroussi into the teaching of Karpman in view of Wang. The suggestion for doing so would ensure the authenticity of the documents. Therefore, it would have been obvious to combine Seroussi with Karpman in view of Wang.
28. Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Karpman et al., US 11995803 B1, and in view of Wang US 20240242402 A1, and further in view of Gupta et al. US 20240257470 A1.
29. As per claim 12, Karpman in view of Wang discloses: The method of claim 9, (See rejection of claim 9 above).
30. Karpman in view of Wang doesn’t expressly disclose: creating the image generation model comprises: computing a latent diffusion loss; and updating parameters of the image generation model based on the latent diffusion loss.
31. Gupta discloses: creating the image generation model comprises: computing a latent diffusion loss; and updating parameters of the image generation model based on the latent diffusion loss. (Gupta, [052], “According to embodiments, a full set personalization conditioning of the UNet may include conditioning on the text describing the scene (i.e., text input) and avatar-object interaction (y), the rendered (RGBA) image (p) of the avatar in the target pose, the personalization downsample rate (w), and the timestep (t). The full set personalization conditioning of the UNet (of the personalized image generation model) may be learned via the latent diffusion model loss L.sub.LDM.”)
32. Karpman in view of Wang is analogous art with respect to Gupta because they are from the same field of endeavor, namely image processing. Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to include the process of installing the processes computing a latent diffusion loss; and updating parameters of the image generation model based on the latent diffusion loss; as taught by Gupta into the teaching of Karpman in view of Wang. The suggestion for doing so would increase accuracy and faithfulness to the user's identity. Therefore, it would have been obvious to combine Gupta with Karpman in view of Wang.
33. Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Karpman et al., US 11995803 B1, and in view of Wang US 20240242402 A1, and further in view of Gupta et al. US 20240257470 A1.
34. As per claim 13, Karpman d in view of Wang discloses: The method of claim 9, (See rejection of claim 9 above).
35. Karpman in view of Wang doesn’t expressly disclose: creating the image generation model comprises: computing an encryption loss; and updating parameters of the image generation model based on the encryption loss.
36. O’Brien discloses: creating the image generation model comprises: computing an encryption loss; and updating parameters of the image generation model based on the encryption loss. (O’Brien, [0022], “Even more preferred is such a system that reduces the size of the encrypted file to a size that is many times smaller as compared to the original media file and utilizes loss-less encryption, i.e., the decryption reliably restores the original non-encrypted file without any loss of information. Most preferred is an encryption/decryption system wherein the data is encrypted at least three times.”)
37. Karpman in view of Wang is analogous art with respect to O’Brien because they are from the same field of endeavor, namely image processing. Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to include the process of installing the processes of computing an encryption loss; and updating parameters of the image generation model based on the encryption loss; as taught by O’Brien into the teaching of Karpman in view of Wang. The suggestion for doing so would enhance transfer speeds. Therefore, it would have been obvious to combine O’Brien with Karpman in view of Wang.
Conclusion
38. Any inquiry concerning this communication or earlier communications from the examiner should be directed to ABDERRAHIM MEROUAN whose telephone number is (571)270-5254. The examiner can normally be reached 9 AM -- 5 PM.
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.
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.
/ABDERRAHIM MEROUAN/Supervisory Patent Examiner, Art Unit 2683
Prosecution Insights