pith. sign in

arxiv: 2509.22126 · v2 · submitted 2025-09-26 · 💻 cs.CR · cs.CV

Guidance Watermarking for Diffusion Models

Enoal Gesny , Eva Giboulot , Teddy Furon , Vivien Chappelier This is my paper

Pith reviewed 2026-05-18 12:58 UTC · model grok-4.3

classification 💻 cs.CR cs.CV
keywords watermarkingdiffusion modelsgradient guidanceimage generationrobustnesspost-hoc embeddinggenerative security
0
0 comments X

The pith

By using gradients from any off-the-shelf watermark decoder to guide the diffusion process, post-hoc watermarking schemes can be converted into in-generation embeddings.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows how to embed watermarks directly into images as they are generated by diffusion models. It computes gradients from a watermark decoder after applying various image augmentations and uses those gradients to steer the denoising steps. This turns any existing post-hoc detector into an active embedding tool during generation without retraining or fine-tuning. A sympathetic reader would care because the method offers a flexible way to add security to AI-generated images while keeping output quality and variety intact. It also works alongside other approaches such as modifying the variational autoencoder.

Core claim

The central claim is that guiding the diffusion sampling process with the gradient computed from a watermark decoder, incorporating image augmentations, effectively embeds the watermark along the generation trajectory. This converts post-hoc schemes into in-generation embeddings without requiring model-specific tuning or retraining. The method is shown to be complementary to techniques that modify the variational autoencoder at the end of the diffusion process and preserves both diversity and quality of generated images for a given seed and prompt.

What carries the argument

The watermarking guidance mechanism that computes gradients from the decoder after applying various image augmentations to steer the diffusion trajectory.

If this is right

  • The watermark becomes embedded during generation rather than applied afterward.
  • Robustness to attacks increases because augmentations are included in the gradient computation.
  • No retraining or fine-tuning of the diffusion model or decoder is needed.
  • The approach complements VAE modification techniques at the end of diffusion.
  • Generated image quality and diversity remain largely unchanged.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This guidance approach could extend to other generative models by adapting the gradient computation for their sampling processes.
  • Watermarking might become a built-in step in standard generation pipelines for better traceability of AI content.
  • Further tests with different attack types or augmentation combinations could identify additional robustness benefits.

Load-bearing premise

The gradient signal from an unmodified off-the-shelf watermark decoder remains sufficiently strong and stable throughout the diffusion trajectory even after image augmentations.

What would settle it

Showing that watermark detection rates drop sharply or that images deviate noticeably in quality or diversity for fixed seeds and prompts when the guidance is applied.

Figures

Figures reproduced from arXiv: 2509.22126 by Enoal Gesny, Eva Giboulot, Teddy Furon, Vivien Chappelier.

Figure 1
Figure 1. Figure 1: Differences of (log)-spectrum of generated images with and without watermarking. Left: [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: From left to right: ‘A vibrant autumn forest with red, orange, and yellow leaves and a [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Probability of detection PD of post-hoc and corresponding guided methods as a function of the PFA for different models. The curve is shown over all studied augmentations, with 1000 images generated form the Gustavosta/Stable-Diffusion-Prompts prompts for each augmentation. performance of other in-generation schemes. The same holds for zero-bit detection. By design Tree-Rings is robust against rotation but … view at source ↗
Figure 4
Figure 4. Figure 4: Zero-bit detection on Stable-Signature with SD2 and Sana. The guidance patches a [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Probability of detection PD of post-hoc and corresponding guided methods as a function of the PFA for different models, under the attack using the original VAE to remove the watermark.The curves were computed over 200 images from Gustavosta/Stable-Diffusion-Prompts for each attack. Indeed, since the gradient has to be back-propagated through the (unwatermarked) VAE, such attacks are implicitly within the t… view at source ↗
Figure 6
Figure 6. Figure 6: Example of images generated with our G-SSig and G-VS for different values of ω. The other parameters are referenced in table 4. From Top to Bottom: with Stable Diffusion 2 with G-SSig and Sana with G-VS. The values of ω are intentionally exaggerated to highlight visible artifacts. Our choice appears framed in green. LDM % Clip (τ ) Max norm (η) ω Stable-Diffusion 2 10% 0.3 250 Flux 10% 0.5 500 Sana 10% 0.5… view at source ↗
Figure 7
Figure 7. Figure 7: Estimated biases and covariance matrix computed over [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Empirical probability of false alarm of whitened and non-whitened detectors computed over [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Empirical pdf of the p-values of Tree-Ring. Top - with the original code. Bottom - with our patch. Blue - with a seed randomly drawn from a Gaussian distribution, Orange - with a seed reconstructed from an image (1,300 images from MIRFlickR). secret vector U distribution is known and easy to sample from. In the worst case, it is thus possible to estimate the p-value equation 12 through Monte-Carlo methods.… view at source ↗
Figure 12
Figure 12. Figure 12: Once again, we reliably outperform the baseline post-hoc schemes, though by a smaller [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
Figure 10
Figure 10. Figure 10: Mean Bit Accuracy for all combinations of diffusion model and watermark decoder. [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: − log10(PF A) for all combinations of diffusion model and watermark detector. Higher is better. Thresholds are − log10(PF A) @ PD = 0.9 for the baseline method (red) and guidance (brown). 21 [PITH_FULL_IMAGE:figures/full_fig_p021_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Repeated experiment for 200 images generated form the COCO captions of [PITH_FULL_IMAGE:figures/full_fig_p022_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Examples of MSCOCO images generated by Sana with our [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Images generated by Stable-Diffusion without and with our watermark embedding for [PITH_FULL_IMAGE:figures/full_fig_p023_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Images generated by Flux without and with our watermark embedding for [PITH_FULL_IMAGE:figures/full_fig_p024_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Images generated by Sana without and with our watermark embedding for [PITH_FULL_IMAGE:figures/full_fig_p025_16.png] view at source ↗
read the original abstract

This paper introduces a novel watermarking method for diffusion models. It is based on guiding the diffusion process using the gradient computed from any off-the-shelf watermark decoder. The gradient computation encompasses different image augmentations, increasing robustness to attacks against which the decoder was not originally robust, without retraining or fine-tuning. Our method effectively convert any \textit{post-hoc} watermarking scheme into an in-generation embedding along the diffusion process. We show that this approach is complementary to watermarking techniques modifying the variational autoencoder at the end of the diffusion process. We validate the methods on different diffusion models and detectors. The watermarking guidance does not significantly alter the generated image for a given seed and prompt, preserving both the diversity and quality of generation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces Guidance Watermarking for diffusion models, which embeds watermarks by guiding the denoising process with gradients computed from any off-the-shelf watermark decoder. Augmentations are applied during gradient computation to improve robustness without retraining the decoder or the diffusion model. The central claim is that this converts any post-hoc watermarking scheme into an in-generation embedding along the full diffusion trajectory; the method is presented as complementary to VAE-based watermarking and is claimed to preserve generation quality and diversity, with validation across multiple diffusion models and detectors.

Significance. If the central claim holds with quantitative support, the approach would provide a practical, model-agnostic route to integrate watermarking into existing diffusion pipelines, avoiding the need for decoder or model retraining. The use of augmentations to strengthen the guidance signal and the complementarity to VAE modifications are potentially useful contributions to generative-model security. However, the absence of metrics in the high-level description limits the assessed significance at present.

major comments (2)
  1. Abstract: the claim that the method 'effectively convert[s] any post-hoc watermarking scheme into an in-generation embedding along the diffusion process' and works 'without model-specific tuning' rests on the unexamined assumption that the unmodified decoder gradient supplies a usable signal even at high-noise early timesteps; no analysis, scaling schedule, or ablation addressing timestep dependence is referenced, which is load-bearing for the full-trajectory conversion claim.
  2. Abstract: validation is asserted 'across models and detectors' with 'no significant quality loss,' yet the description supplies no quantitative metrics (e.g., FID, CLIP score, watermark detection rates, or attack success rates), ablation results, or experimental details; this prevents evaluation of whether the guidance actually achieves reliable embedding throughout the trajectory.
minor comments (2)
  1. Abstract: grammatical error in 'Our method effectively convert any post-hoc...'; should read 'converts'.
  2. Abstract: the phrase 'the watermarking guidance does not significantly alter the generated image for a given seed and prompt' would benefit from a brief statement of the quantitative threshold used to define 'significant' alteration.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the presentation of our contributions. We respond to each major comment below and indicate the revisions planned for the next version of the manuscript.

read point-by-point responses

  1. Referee: Abstract: the claim that the method 'effectively convert[s] any post-hoc watermarking scheme into an in-generation embedding along the diffusion process' and works 'without model-specific tuning' rests on the unexamined assumption that the unmodified decoder gradient supplies a usable signal even at high-noise early timesteps; no analysis, scaling schedule, or ablation addressing timestep dependence is referenced, which is load-bearing for the full-trajectory conversion claim.

    Authors: We thank the referee for this observation. The guidance signal is applied at every timestep of the reverse process, and our experiments demonstrate that the unmodified decoder produces a sufficiently informative gradient to achieve reliable embedding without per-model tuning. Nevertheless, we agree that an explicit analysis of signal strength and effectiveness at high-noise timesteps would better substantiate the full-trajectory claim. In the revised manuscript we will add an ablation examining gradient magnitude and watermark success rate across timesteps together with the guidance-weight scaling schedule used in all reported experiments. revision: yes

  2. Referee: Abstract: validation is asserted 'across models and detectors' with 'no significant quality loss,' yet the description supplies no quantitative metrics (e.g., FID, CLIP score, watermark detection rates, or attack success rates), ablation results, or experimental details; this prevents evaluation of whether the guidance actually achieves reliable embedding throughout the trajectory.

    Authors: The abstract is written as a high-level summary; the full manuscript reports quantitative results in the Experiments section, including FID and CLIP scores for quality, detection rates under clean and attacked conditions, and comparisons across multiple diffusion models and watermark decoders. To improve immediate readability we will revise the abstract to incorporate a small number of representative metrics (e.g., average detection rate and FID delta) while preserving conciseness. revision: partial

Circularity Check

0 steps flagged

No circularity: guidance defined directly from external decoder gradients

full rationale

The paper's central construction uses the gradient signal from any unmodified off-the-shelf watermark decoder (with augmentations) to steer the diffusion trajectory. This mechanism is defined independently of the target embedding result and does not reduce to a fitted parameter, self-referential loop, or self-citation chain. The conversion of post-hoc schemes into in-generation embedding is presented as an empirical outcome of the guidance process rather than an input by construction. No uniqueness theorems, ansatzes smuggled via prior author work, or renaming of known results appear in the derivation. The method remains self-contained against external benchmarks such as existing decoders and diffusion models.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The central claim rests on the unstated premise that decoder gradients remain informative across diffusion timesteps.

pith-pipeline@v0.9.0 · 5653 in / 1032 out tokens · 26764 ms · 2026-05-18T12:58:30.233198+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages · 2 internal anchors

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

    work page

  2. [2]

    https://media.defense.gov/2025/Jan/29/2003634788/-1/-1/0/CSI-CONTENT-CREDENTIALS.PDF, 2025

    Content credentials: Strengthening multimedia integrity in the generative AI era. https://media.defense.gov/2025/Jan/29/2003634788/-1/-1/0/CSI-CONTENT-CREDENTIALS.PDF, 2025. USA National Security Agency, Australian Signals Directorate’s Australian Cyber Security Centre, Canadian Centre for Cyber Security, and United Kingdom National Cyber Security Centre

    work page 2025

  3. [3]

    The AI Waterfall : A Case Study in Integrating Machine Learning and Security

    Patrick Bas and Jan Butora. The AI Waterfall : A Case Study in Integrating Machine Learning and Security . working paper or preprint, April 2025. URL https://hal.science/hal-05011387

    work page 2025

  4. [4]

    Black Forest Labs . Flux. https://github.com/black-forest-labs/flux, 2024

    work page 2024

  5. [5]

    Nepotistically trained generative image models collapse

    Maty Bohacek and Hany Farid. Nepotistically trained generative image models collapse. In ICLR 2025 Workshop on Navigating and Addressing Data Problems for Foundation Models, 2025. URL https://openreview.net/forum?id=mkZB0fKLX8

    work page 2025

  6. [6]

    Trustmark: Universal watermarking for arbitrary resolution images.arXiv preprint arXiv:2311.18297, 2023

    Tu Bui, Shruti Agarwal, and John Collomosse. TrustMark : Universal Watermarking for Arbitrary Resolution Images , November 2023. URL http://arxiv.org/abs/2311.18297. arXiv:2311.18297 [cs]

    work page arXiv 2023

  7. [7]

    C2PA : The coalition for content provenance and authenticity

    C2PA. C2PA : The coalition for content provenance and authenticity. https://c2pa.org, 2024

    work page 2024

  8. [8]

    Chinese AI governance rules

    China. Chinese AI governance rules. http://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm, 2023

    work page 2023

  9. [9]

    Asymptotically optimum universal watermark embedding and detection in the high-snr regime

    Pedro Comesana, Neri Merhav, and Mauro Barni. Asymptotically optimum universal watermark embedding and detection in the high-snr regime. IEEE Transactions on Information Theory, 56 0 (6): 0 2804--2815, 2010. doi:10.1109/TIT.2010.2046223

    work page doi:10.1109/tit.2010.2046223 2010

  10. [10]

    On the detection of synthetic images generated by diffusion models

    Riccardo Corvi, Davide Cozzolino, Giada Zingarini, Giovanni Poggi, Koki Nagano, and Luisa Verdoliva. On the detection of synthetic images generated by diffusion models. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 1--5. IEEE, 2023

    work page 2023

  11. [11]

    M. Costa. Writing on dirty paper (corresp.). IEEE Transactions on Information Theory, 29 0 (3): 0 439--441, 1983. doi:10.1109/TIT.1983.1056659

    work page doi:10.1109/tit.1983.1056659 1983

  12. [12]

    Ingemar J. Cox. Digital watermarking and steganography. The Morgan Kaufmann series in multimedia information and systems. Morgan Kaufmann Publishers, Amsterdam Boston, 2nd ed edition, 2008. ISBN 978-0-12-372585-1

    work page 2008

  13. [13]

    Flashattention-2: Faster attention with better parallelism and work partitioning

    Tri Dao. Flashattention-2: Faster attention with better parallelism and work partitioning. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=mZn2Xyh9Ec

    work page 2024

  14. [14]

    Diffusion models beat GAN s on image synthesis

    Prafulla Dhariwal and Alexander Quinn Nichol. Diffusion models beat GAN s on image synthesis. In A. Beygelzimer, Y. Dauphin, P. Liang, and J. Wortman Vaughan (eds.), Advances in Neural Information Processing Systems, 2021. URL https://openreview.net/forum?id=AAWuCvzaVt

    work page 2021

  15. [15]

    Digital watermarking applications

    Digital Watermarking Alliance DWA. Digital watermarking applications. https://digitalwatermarkingalliance.org/digital-watermarking-applications/

    work page

  16. [16]

    European AI Act

    Europe. European AI Act . https://artificialintelligenceact.eu/, 2023

    work page 2023

  17. [17]

    Watermarking Images in Self-Supervised Latent Spaces

    Pierre Fernandez, Alexandre Sablayrolles, Teddy Furon, Herv \'e J \'e gou, and Matthijs Douze. Watermarking Images in Self-Supervised Latent Spaces . In IEEE (ed.), ICASSP 2022 - IEEE International Conference on Acoustics, Speech and Signal Processing , pp.\ 1--5, Singapore, Singapore, May 2022. IEEE , IEEE . URL https://inria.hal.science/hal-03591396

    work page 2022

  18. [18]

    The stable signature: Rooting watermarks in latent diffusion models

    Pierre Fernandez, Guillaume Couairon, Hervé Jégou, Matthijs Douze, and Teddy Furon. The stable signature: Rooting watermarks in latent diffusion models. In Proceedings of the IEEE/CVF international conference on computer vision, 2023

    work page 2023

  19. [19]

    Video seal: Open and efficient video watermarking

    Pierre Fernandez, Hady Elsahar, I. Zeki Yalniz, and Alexandre Mourachko. Video Seal : Open and Efficient Video Watermarking , December 2024 a . URL http://arxiv.org/abs/2412.09492. arXiv:2412.09492 [cs]

    work page arXiv 2024

  20. [20]

    What lies ahead for generative ai watermarking

    Pierre Fernandez, Anthony Level, and Teddy Furon. What lies ahead for generative ai watermarking. In 2nd Workshop on Generative AI and Law (GenLaw ’24), ICML, 2024 b

    work page 2024

  21. [21]

    About zero bit watermarking error exponents

    Teddy Furon. About zero bit watermarking error exponents . In ICASSP2017 - IEEE International Conference on Acoustics, Speech and Signal Processing , New Orleans, United States, March 2017. IEEE . URL https://inria.hal.science/hal-01512705

    work page 2017

  22. [22]

    Byzantine machine learning: A primer

    Rachid Guerraoui, Nirupam Gupta, and Rafael Pinot. Byzantine machine learning: A primer. ACM Computing Surveys, 56 0 (7): 0 1--39, 2024

    work page 2024

  23. [23]

    CLIPS core: A reference-free evaluation metric for image captioning

    Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. CLIPS core: A reference-free evaluation metric for image captioning. In Marie-Francine Moens, Xuanjing Huang, Lucia Specia, and Scott Wen-tau Yih (eds.), Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp.\ 7514--7528, Online and Punta Cana, D...

    work page doi:10.18653/v1/2021.emnlp-main.595 2021

  24. [24]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017

    work page 2017

  25. [25]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 0 6840--6851, 2020

    work page 2020

  26. [26]

    Robin: Robust and invisible watermarks for diffusion models with adversarial optimization

    Huayang Huang, Yu Wu, and Qian Wang. Robin: Robust and invisible watermarks for diffusion models with adversarial optimization. Advances in Neural Information Processing Systems, 37: 0 3937--3963, 2025

    work page 2025

  27. [27]

    Diffusion models for counterfactual explanations

    Guillaume Jeanneret, Loic Simon, and Frederic Jurie. Diffusion models for counterfactual explanations. In Proceedings of the Asian Conference on Computer Vision (ACCV), pp.\ 858--876, December 2022

    work page 2022

  28. [28]

    A watermark for large language models

    John Kirchenbauer, Jonas Geiping, Yuxin Wen, Jonathan Katz, Ian Miers, and Tom Goldstein. A watermark for large language models. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett (eds.), Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learnin...

    work page 2023

  29. [29]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll \'a r, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer vision--ECCV 2014: 13th European conference, zurich, Switzerland, September 6-12, 2014, proceedings, part v 13, pp.\ 740--755. Springer, 2014

    work page 2014

  30. [30]

    Conflict-averse gradient descent for multi-task learning

    Bo Liu, Xingchao Liu, Xiaojie Jin, Peter Stone, and Qiang Liu. Conflict-averse gradient descent for multi-task learning. Advances in Neural Information Processing Systems, 34: 0 18878--18890, 2021

    work page 2021

  31. [31]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps. Advances in Neural Information Processing Systems, 35: 0 5775--5787, 2022

    work page 2022

  32. [32]

    Improved denoising diffusion probabilistic models, 2021

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models, 2021. URL https://openreview.net/forum?id=-NEXDKk8gZ

    work page 2021

  33. [33]

    Channel coding rate in the finite blocklength regime

    Yury Polyanskiy, H Vincent Poor, and Sergio Verd \'u . Channel coding rate in the finite blocklength regime. IEEE Transactions on Information Theory, 56 0 (5): 0 2307--2359, 2010

    work page 2010

  34. [34]

    Hierarchical Text-Conditional Image Generation with CLIP Latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv preprint arXiv:2204.06125, 1 0 (2): 0 3, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  35. [35]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj \"o rn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 10684--10695, 2022 a

    work page 2022

  36. [36]

    Masked feature prediction for self-supervised visual pre-training

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bjorn Ommer. High- Resolution Image Synthesis with Latent Diffusion Models . In 2022 IEEE / CVF Conference on Computer Vision and Pattern Recognition ( CVPR ) , pp.\ 10674--10685, New Orleans, LA, USA, June 2022 b . IEEE. ISBN 978-1-6654-6946-3. doi:10.1109/CVPR52688.2022.01042. URL http...

    work page doi:10.1109/cvpr52688.2022.01042 2022

  37. [37]

    Probability content of regions under spherical normal distributions, iv: The distribution of homogeneous and non-homogeneous quadratic functions of normal variables

    Harold Ruben. Probability content of regions under spherical normal distributions, iv: The distribution of homogeneous and non-homogeneous quadratic functions of normal variables. The Annals of Mathematical Statistics, Ann. Math. Statist, 33 0 (2): 0 542--570, 1962

    work page 1962

  38. [38]

    Proactive detection of voice cloning with localized watermarking

    Robin San Roman, Pierre Fernandez, Hady Elsahar, Alexandre D \'e fossez, Teddy Furon, and Tuan Tran. Proactive detection of voice cloning with localized watermarking. In ICML 2024-41st International Conference on Machine Learning, volume 235, pp.\ 1--17, 2024

    work page 2024

  39. [39]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=St1giarCHLP

    work page 2021

  40. [40]

    Ensuring safe, secure, and trustworthy AI

    USA . Ensuring safe, secure, and trustworthy AI . https://www.whitehouse.gov/wp-content/uploads/2023/07/Ensuring-Safe-Secure-and-Trustworthy-AI.pdf, July 2023. Accessed: [july 2023]

    work page 2023

  41. [41]

    FABRIC : Personalizing diffusion models with iterative feedback, 2024

    Dimitri von R \"u tte, Elisabetta Fedele, Jonathan Thomm, and Lukas Wolf. FABRIC : Personalizing diffusion models with iterative feedback, 2024. URL https://openreview.net/forum?id=zsfrzYWoOP

    work page 2024

  42. [42]

    Tree-rings watermarks: Invisible fingerprints for diffusion images

    Yuxin Wen, John Kirchenbauer, Jonas Geiping, and Tom Goldstein. Tree-rings watermarks: Invisible fingerprints for diffusion images. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=Z57JrmubNl

    work page 2023

  43. [43]

    SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers

    Enze Xie, Junsong Chen, Junyu Chen, Han Cai, Haotian Tang, Yujun Lin, Zhekai Zhang, Muyang Li, Ligeng Zhu, Yao Lu, and Song Han. Sana: Efficient high-resolution image synthesis with linear diffusion transformer, 2024. URL https://arxiv.org/abs/2410.10629

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  44. [44]

    Oona Rainio, Jarmo Teuho, and Riku Klén

    Rui Xu, Mengya Hu, Deren Lei, Yaxi Li, David Lowe, Alex Gorevski, Mingyu Wang, Emily Ching, and Alex Deng. Invismark: Invisible and robust watermarking for ai-generated image provenance. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp.\ 909--918, 2025. doi:10.1109/WACV61041.2025.00098

    work page doi:10.1109/wacv61041.2025.00098 2025

  45. [45]

    Gaussian shading: Provable performance-lossless image watermarking for diffusion models

    Zijin Yang, Kai Zeng, Kejiang Chen, Han Fang, Weiming Zhang, and Nenghai Yu. Gaussian shading: Provable performance-lossless image watermarking for diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 12162--12171, 2024

    work page 2024

  46. [46]

    Gradient surgery for multi-task learning

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. Gradient surgery for multi-task learning. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin (eds.), Advances in Neural Information Processing Systems, volume 33, pp.\ 5824--5836. Curran Associates, Inc., 2020. URL https://proceedings.neurips.cc/paper_...

    work page 2020

  47. [47]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 3836--3847, 2023

    work page 2023

  48. [48]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 586--595, 2018

    work page 2018

  49. [49]

    Hidden: Hiding data with deep networks

    Jiren Zhu, Russell Kaplan, Justin Johnson, and Li Fei-Fei. Hidden: Hiding data with deep networks. In Proceedings of the European conference on computer vision (ECCV), pp.\ 657--672, 2018

    work page 2018

  50. [50]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

    work page

  51. [51]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

    work page

  52. [52]

    Amrum Lighthouse

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

    work page arXiv 2000