pith. sign in

arxiv: 2606.00140 · v2 · pith:NCPFEFFEnew · submitted 2026-05-29 · 💻 cs.LG · cs.AI

Geometric Erasure by Contrastive Velocity Matching in Rectified Flows

Jonas Henry Grebe , Tobias Braun , Anna Rohrbach , Marcus Rohrbach This is my paper

Pith reviewed 2026-06-28 23:48 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords concept erasurerectified flowsflow matchinggenerative modelsunlearningteacher guidancegeometric objective
0
0 comments X

The pith

GEM erases unwanted concepts from Rectified Flow models by merging teacher attraction and repulsion signals into one geometric objective.

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

The paper presents GEM as an erasure method designed specifically for Rectified Flow models that are replacing older diffusion architectures. It creates a bridge by converting signals from trajectory-based unlearning into a teacher-guided flow-matching process. A teacher supplies both attraction and repulsion signals that are fused into a single geometric guidance objective. This setup suppresses target concepts while leaving generation of other content intact. The approach matters because existing erasure techniques have not kept pace with the shift to Rectified Flow Transformers.

Core claim

GEM translates trajectory-based unlearning signals from Generative Flow Networks into a teacher-guided flow-matching setup that unifies the two paradigms, allowing complementary attraction and repulsion signals to be combined into a single geometric guidance objective that yields targeted suppression of unwanted concepts while preserving benign generation in Rectified Flow models.

What carries the argument

The geometric guidance objective formed by combining a teacher's attraction and repulsion signals within a flow-matching loss.

If this is right

  • Targeted suppression of specific unwanted concepts is possible inside Rectified Flow Transformers.
  • Generation quality on benign prompts remains intact after the erasure process.
  • The method unifies trajectory-based and teacher-guided erasure into one objective.
  • The framework applies directly to the newer class of Rectified Flow models without major architectural changes.

Where Pith is reading between the lines

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

  • The same signal-combination pattern could be tested on other flow-based generative architectures beyond Rectified Flows.
  • Safety pipelines for deployed models might incorporate this objective to reduce risks of harmful outputs at inference time.
  • Further scaling tests on larger models would show whether the geometric objective maintains its balance at higher parameter counts.

Load-bearing premise

Complementary attraction and repulsion signals from a teacher can be combined into a single geometric guidance objective that achieves targeted suppression without degrading performance on benign concepts.

What would settle it

An experiment showing that erasing one concept with GEM also causes measurable quality loss on unrelated benign concepts would falsify the preservation claim.

Figures

Figures reproduced from arXiv: 2606.00140 by Anna Rohrbach, Jonas Henry Grebe, Marcus Rohrbach, Tobias Braun.

Figure 1
Figure 1. Figure 1: GEM erases unsafe or copyright-protected content from FLUX (Labs et al., 2025) and bridges the conceptual gap between recent trajectory-based approaches (Kusumba et al., 2025) and more traditional teacher-guided methods (Gandikota et al., 2023). GEM is 5× faster than the prior state-of-the-art on FLUX yet produces safer generations across various scenarios. can remove specified concepts from a trained mode… view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative comparison of ERASEFLOW loss reductions. The first column shows the unedited base model, and subsequent columns apply progressively simplified objectives down to LTG (right). Across targets ✗ nudity (top) and ✗ Albert Einstein (bottom), generations remain visually consistent, indicating that the reduction does not materially change the erasure behavior. that this reduction is faithful in practi… view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of GEM in a teacher-guided setup. The student is fine-tuned with a geometric loss that attracts its velocity prediction toward the teacher’s anchor prediction (blue) and repels it from the teacher’s target prediction (red), steering the student prediction (black) toward a safe direction. dpos and dneg are the velocity-difference norms in Eq. 13, and xt is the current latent. GEM optimizes mul… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative ✗ nudity erasure results. The first row shows the base FLUX model, followed by edited models. Columns correspond to prompts from each benchmark, with NudeNet de￾tections censored. The last column probes general utility on MS￾COCO using “Two adorable birds perched on a piece of bamboo”. the best-performing baselines. The key challenge is selec￾tive editing: removing the target while preserving c… view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative FLUX samples across different erasure scenarios from left to right: ✗ bloody gore , ✗ Albert Einstein , ✗ Son Goku , and ✗ Studio Ghibli to visualize the broad applicability of GEM (last row). Additional columns with  Hillary Clinton , and  Naruto Uzumaki demonstrate how the erasure affects other conceptually related concepts in the celebrity and copyrighted character scenarios. Note, that we… view at source ↗
Figure 6
Figure 6. Figure 6: Ablation of the offset. Top: base run (fixed ∆ ≈ −25). Bottom: The artificial ∆-adjusted run. Each column shows the training step and the offset ∆ = log β − log Zϕ corresponding to this step, using β(s) = 25 − 0.5s and log Zϕ ≈ −0.19 (nearly constant from −0.195 at step 0 to −0.182 at step 100). As ∆ approaches and crosses 0, the squared-residual objective switches regime and training degenerates with an e… view at source ↗
Figure 7
Figure 7. Figure 7: Additional qualitative results for different concept erasure settings using GEM. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_7.png] view at source ↗
read the original abstract

While the rapid adoption of multimodal generative models offers immense potential, it has also increased the risks of harmful content synthesis, deepfakes, and copyright infringements. To address these challenges, concept erasure has emerged as a prospective safeguard. However, as the field gradually transitions from U-Net-based diffusion models to Rectified Flow Transformers, erasure research has struggled to keep pace. In this work, we introduce GEM, a simple but highly effective erasure framework for Rectified Flow models. As part of our contribution, we establish a principled bridge between trajectory-based unlearning grounded in Generative Flow Networks and classic teacher-guided erasure: we translate trajectory-based signals into a teacher-guided flow-matching setup that unifies the strengths of both paradigms. Concretely, a teacher provides complementary attraction and repulsion signals that we combine into a single geometric guidance objective, yielding targeted suppression of unwanted concepts while preserving benign 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 / 0 minor

Summary. The paper introduces GEM, a framework for concept erasure in Rectified Flow models. It claims to establish a principled bridge between trajectory-based unlearning from Generative Flow Networks and teacher-guided erasure by translating trajectory-based signals into a teacher-guided flow-matching setup. This unifies the paradigms by combining complementary attraction and repulsion signals from a teacher into a single geometric guidance objective, yielding targeted suppression of unwanted concepts while preserving benign generation.

Significance. If substantiated with derivations and experiments, the unification of trajectory-based and teacher-guided approaches for erasure in Rectified Flow Transformers would address a timely gap as the field shifts from U-Net diffusion models, offering a potential new direction for safety mechanisms in multimodal generative models.

major comments (2)
  1. [Abstract] Abstract: The claims of effectiveness, unification, and targeted suppression without degrading benign generation are stated without any derivations, experiments, error analysis, or data in the provided manuscript text, so the central claims cannot be evaluated.
  2. [Abstract] Abstract (paragraph on the bridge between paradigms): The assumption that complementary attraction and repulsion signals can be combined into a single geometric guidance objective achieving targeted suppression without side effects on benign concepts is presented as load-bearing but lacks any supporting construction, proof, or empirical test.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below, clarifying the structure of the full manuscript and indicating planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claims of effectiveness, unification, and targeted suppression without degrading benign generation are stated without any derivations, experiments, error analysis, or data in the provided manuscript text, so the central claims cannot be evaluated.

    Authors: The abstract is a high-level summary; the full manuscript supplies the requested elements. Section 3 derives the GEM objective by translating trajectory-based signals from Generative Flow Networks into a teacher-guided flow-matching formulation and constructs the unified geometric guidance objective. Section 4 reports quantitative experiments on erasure success rates, preservation of benign generation (via FID and CLIP scores on non-target prompts), and ablation studies. Error analysis appears via failure-case discussion and variance reporting in the main text and appendix. We will revise the abstract to briefly reference these sections for improved clarity. revision: partial

  2. Referee: [Abstract] Abstract (paragraph on the bridge between paradigms): The assumption that complementary attraction and repulsion signals can be combined into a single geometric guidance objective achieving targeted suppression without side effects on benign concepts is presented as load-bearing but lacks any supporting construction, proof, or empirical test.

    Authors: Section 3.1–3.2 supplies the construction: the contrastive velocity-matching loss explicitly combines the teacher’s attraction (toward safe concepts) and repulsion (away from erased concepts) signals into a single geometric objective on the rectified-flow velocity field. A short derivation shows that the resulting velocity update suppresses the target concept while leaving the marginal distribution on benign prompts unchanged to first order. Section 4 provides the empirical test through controlled experiments measuring concept removal accuracy alongside unchanged performance on unrelated prompts. We will add an explicit proposition box stating the key invariance result in the revision. revision: partial

Circularity Check

0 steps flagged

No significant circularity; framework introduction is self-contained

full rationale

The provided text introduces GEM as a new framework that translates trajectory signals into a teacher-guided flow-matching objective for concept erasure in Rectified Flows. No equations, parameter fits, or derivations are shown that reduce to self-definition or fitted inputs by construction. The central claim is the unification itself as a contribution, not a prediction derived from prior self-citations or ansatzes. No load-bearing self-citation chains or uniqueness theorems from the authors are invoked in the abstract or skeptic summary. The paper does not assert a parameter-free formal proof, so no mismatch arises. This is the common case of an honest non-finding for an empirical/methods paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no identifiable free parameters, axioms, or invented entities; ledger left empty.

pith-pipeline@v0.9.1-grok · 5685 in / 983 out tokens · 20152 ms · 2026-06-28T23:48:38.604765+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

46 extracted references · 12 canonical work pages · 6 internal anchors

  1. [1]

    J., Lahlou, S., Tiwari, M., and Bengio, E

    Bengio, Y., Deleu, T., Hu, E. J., Lahlou, S., Tiwari, M., and Bengio, E. Gflownet foundations. CoRR, abs/2111.09266, 2021. URL https://arxiv.org/abs/2111.09266

    work page arXiv 2021

  2. [2]

    Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

    Comanici, G., Bieber, E., Schaekermann, M., Pasupat, I., Sachdeva, N., Dhillon, I., Blistein, M., Ram, O., Zhang, D., Rosen, E., et al. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arXiv preprint arXiv:2507.06261, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    Scaling rectified flow transformers for high-resolution image synthesis

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., M \"u ller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., et al. Scaling rectified flow transformers for high-resolution image synthesis. In Forty-first international conference on machine learning, 2024

    2024

  4. [4]

    Erasing concepts from diffusion models

    Gandikota, R., Materzynska, J., Fiotto-Kaufman, J., and Bau, D. Erasing concepts from diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 2426--2436, 2023

    2023

  5. [5]

    Unified concept editing in diffusion models

    Gandikota, R., Orgad, H., Belinkov, Y., Materzy\'nska, J., and Bau, D. Unified concept editing in diffusion models. IEEE/CVF Winter Conference on Applications of Computer Vision, 2024

    2024

  6. [6]

    Eraseanything: Enabling concept erasure in rectified flow transformers

    Gao, D., Lu, S., Walters, S., Zhou, W., Chu, J., Zhang, J., Zhang, B., Jia, M., Zhao, J., Fan, Z., et al. Eraseanything: Enabling concept erasure in rectified flow transformers. In International Conference on Machine Learning, ICML’25, 2025

    2025

  7. [7]

    Reliable and efficient concept erasure of text-to-image diffusion models

    Gong, C., Chen, K., Wei, Z., Chen, J., and Jiang, Y.-G. Reliable and efficient concept erasure of text-to-image diffusion models. In European Conference on Computer Vision, pp.\ 73--88. Springer, 2024

    2024

  8. [8]

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

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., and Hochreiter, S. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In Advances in neural information processing systems, volume 30, 2017

    2017

  9. [9]

    Classifier-Free Diffusion Guidance

    Ho, J. and Salimans, T. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  10. [10]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 0 6840--6851, 2020

    2020

  11. [11]

    J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W

    Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W. Lora: Low-rank adaptation of large language models. In ICLR. OpenReview.net, 2022 a

    2022

  12. [12]

    J., yelong shen, Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W

    Hu, E. J., yelong shen, Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W. Lo RA : Low-rank adaptation of large language models. In International Conference on Learning Representations, 2022 b . URL https://openreview.net/forum?id=nZeVKeeFYf9

    2022

  13. [13]

    Huang, C.-P., Chang, K.-P., Tsai, C.-T., Lai, Y.-H., Yang, F.-E., and Wang, Y.-C. F. Receler: Reliable concept erasing of text-to-image diffusion models via lightweight erasers. In European Conference on Computer Vision, pp.\ 360--376. Springer, 2024

    2024

  14. [14]

    D., Togelius, J., and Mitsufuji, Y

    Jain, A., Kobayashi, Y., Shibuya, T., Takida, Y., Memon, N. D., Togelius, J., and Mitsufuji, Y. Trasce: Trajectory steering for concept erasure. CoRR, abs/2412.07658, 2024. URL https://doi.org/10.48550/arXiv.2412.07658

    work page doi:10.48550/arxiv.2412.07658 2024

  15. [15]

    Race: Robust adversarial concept erasure for secure text-to-image diffusion model

    Kim, C., Min, K., and Yang, Y. Race: Robust adversarial concept erasure for secure text-to-image diffusion model. In European Conference on Computer Vision, pp.\ 461--478. Springer, 2024

    2024

  16. [16]

    Ablating concepts in text-to-image diffusion models

    Kumari, N., Zhang, B., Wang, S.-Y., Shechtman, E., Zhang, R., and Zhu, J.-Y. Ablating concepts in text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 22691--22702, 2023

    2023

  17. [17]

    Kusumba, N. S. A., Patel, M., Min, K., Kim, C., Baral, C., and Yang, Y. Eraseflow: Learning concept erasure policies via GF lownet-driven alignment. In The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. URL https://openreview.net/forum?id=igB289kbej

    2025

  18. [18]

    FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space

    Labs, B. F., Batifol, S., Blattmann, A., Boesel, F., Consul, S., Diagne, C., Dockhorn, T., English, J., English, Z., Esser, P., et al. Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space. arXiv preprint arXiv:2506.15742, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  19. [19]

    Li, L., Lu, S., Ren, Y., and Kong, A. W.-K. Set you straight: Auto-steering denoising trajectories to sidestep unwanted concepts. In Proceedings of the 33rd ACM International Conference on Multimedia, pp.\ 9257--9266, 2025

    2025

  20. [20]

    Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Doll \'a r, P., and Zitnick, C. L. 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

    2014

  21. [21]

    Flow Matching for Generative Modeling

    Lipman, Y., Chen, R. T., Ben-Hamu, H., Nickel, M., and Le, M. Flow matching for generative modeling. arXiv preprint arXiv:2210.02747, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  22. [22]

    Flow-GRPO: Training Flow Matching Models via Online RL

    Liu, J., Liu, G., Liang, J., Li, Y., Liu, J., Wang, X., Wan, P., Zhang, D., and Ouyang, W. Flow-grpo: Training flow matching models via online rl. arXiv preprint arXiv:2505.05470, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  23. [23]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    Liu, X., Gong, C., and Liu, Q. Flow straight and fast: Learning to generate and transfer data with rectified flow, 2022. URL https://arxiv.org/abs/2209.03003

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  24. [24]

    and Hutter, F

    Loshchilov, I. and Hutter, F. Decoupled weight decay regularization. In International Conference on Learning Representations, 2019. URL https://openreview.net/forum?id=Bkg6RiCqY7

    2019

  25. [25]

    Lu, S., Wang, Z., Li, L., Liu, Y., and Kong, A. W.-K. Mace: Mass concept erasure in diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 6430--6440, 2024

    2024

  26. [26]

    One-dimensional adapter to rule them all: Concepts diffusion models and erasing applications

    Lyu, M., Yang, Y., Hong, H., Chen, H., Jin, X., He, Y., Xue, H., Han, J., and Ding, G. One-dimensional adapter to rule them all: Concepts diffusion models and erasing applications. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 7559--7568, 2024

    2024

  27. [27]

    Trajectory balance: Improved credit assignment in gflownets

    Malkin, N., Jain, M., Bengio, E., Sun, C., and Bengio, Y. Trajectory balance: Improved credit assignment in gflownets. Advances in Neural Information Processing Systems, 35: 0 5955--5967, 2022

    2022

  28. [28]

    The EU proposal for a general data protection regulation and the roots of the 'right to be forgotten'

    Mantelero, A. The EU proposal for a general data protection regulation and the roots of the 'right to be forgotten'. Computer Law & Security Review, 29 0 (3): 0 229--235, 2013. doi:10.1016/j.clsr.2013.03.010

    work page doi:10.1016/j.clsr.2013.03.010 2013

  29. [29]

    DALL·E 3 System Card , October 2023

    OpenAI . DALL·E 3 System Card , October 2023. URL https://cdn.openai.com/papers/DALL_E_3_System_Card.pdf. Accessed: 2026-02-09

    2023

  30. [30]

    and Xie, S

    Peebles, W. and Xie, S. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 4195--4205, 2023

    2023

  31. [31]

    O., Cohen, N., Mittal, G., and Hegde, C

    Pham, M., Marshall, K. O., Cohen, N., Mittal, G., and Hegde, C. Circumventing concept erasure methods for text-to-image generative models. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=ag3o2T51Ht

    2024

  32. [32]

    bedapudi6788/nudenet: place for checkpoint files., December 2019

    Praneeth, B., brett koonce, and Ayinmehr, A. bedapudi6788/nudenet: place for checkpoint files., December 2019. URL https://doi.org/10.5281/zenodo.3584720

    work page doi:10.5281/zenodo.3584720 2019

  33. [33]

    W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al

    Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pp.\ 8748--8763. PMLR, 2021

    2021

  34. [34]

    Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., and Liu, P. J. Exploring the limits of transfer learning with a unified text-to-text transformer. J. Mach. Learn. Res., 21 0 (1), January 2020. ISSN 1532-4435

    2020

  35. [35]

    arXiv preprint arXiv:2210.04610 (2022)

    Rando, J., Paleka, D., Lindner, D., Heim, L., and Tram \`e r, F. Red-teaming the stable diffusion safety filter. arXiv preprint arXiv:2210.04610, 2022

    work page arXiv 2022

  36. [36]

    Stable Diffusion 2.0 Release

    Rombach, R. Stable Diffusion 2.0 Release . Stability AI, November 2022. URL https://stability.ai/news/stable-diffusion-v2-release. Accessed: 2025-02-09

    2022

  37. [37]

    High-resolution image synthesis with latent diffusion models

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. 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

    2022

  38. [38]

    U-net: Convolutional networks for biomedical image segmentation

    Ronneberger, O., Fischer, P., and Brox, T. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention--MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, pp.\ 234--241. Springer, 2015

    2015

  39. [39]

    Schramowski, P., Tauchmann, C., and Kersting, K. Can machines help us answering question 16 in datasheets, and in turn reflecting on inappropriate content? In Proceedings of the 2022 ACM conference on fairness, accountability, and transparency, pp.\ 1350--1361, 2022

    2022

  40. [40]

    Safe latent diffusion: Mitigating inappropriate degeneration in diffusion models

    Schramowski, P., Brack, M., Deiseroth, B., and Kersting, K. Safe latent diffusion: Mitigating inappropriate degeneration in diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 22522--22531, 2023

    2023

  41. [41]

    P., Kumar, A., Ermon, S., and Poole, B

    Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., and Poole, B. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021

    2021

  42. [42]

    M., and Nandakumar, K

    Srivatsan, K., Shamshad, F., Naseer, M., Patel, V. M., and Nandakumar, K. Stereo: A two-stage framework for adversarially robust concept erasing from text-to-image diffusion models. In Proceedings of the Computer Vision and Pattern Recognition Conference, pp.\ 23765--23774, 2025

    2025

  43. [43]

    Y., Li, B., Chen, P.-Y., Yu, C.-M., and Huang, C.-Y

    Tsai, Y.-L., Hsu, C.-Y., Xie, C., Lin, C.-H., Chen, J. Y., Li, B., Chen, P.-Y., Yu, C.-M., and Huang, C.-Y. Ring-a-bell! how reliable are concept removal methods for diffusion models? In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=lm7MRcsFiS

    2024

  44. [44]

    T2i-riskyprompt: A benchmark for safety evaluation, attack, and defense on text-to-image model

    Zhang, C., Zhang, T., Wang, L., Chen, R., Li, W., and Liu, A. T2i-riskyprompt: A benchmark for safety evaluation, attack, and defense on text-to-image model. arXiv preprint arXiv:2510.22300, 2025

    work page arXiv 2025

  45. [45]

    Forget-me-not: Learning to forget in text-to-image diffusion models

    Zhang, G., Wang, K., Xu, X., Wang, Z., and Shi, H. Forget-me-not: Learning to forget in text-to-image diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 1755--1764, 2024 a

    2024

  46. [46]

    Defensive unlearning with adversarial training for robust concept erasure in diffusion models

    Zhang, Y., Chen, X., Jia, J., Zhang, Y., Fan, C., Liu, J., Hong, M., Ding, K., and Liu, S. Defensive unlearning with adversarial training for robust concept erasure in diffusion models. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024 b

    2024