Pith. sign in

REVIEW 5 major objections 7 minor 52 references

Erase but Preserve: Controllable Removal of Copyrighted Animation Characters via Optimized Semantic Anchors

T0 review · 5 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A text-to-image diffusion model can be asked to omit a specific copyrighted cartoon character during generation by swapping in an optimized anchor word embedding, leaving the surrounding scene intact and requiring no retraining.

desk verdict Solid applied paper with a genuinely new anchor-optimization procedure, but the erasure only works for exact-name prompts; the headline claim needs scope reduction. read the letter →

arxiv 2608.12806 v1 pith:Z4A37KC6 submitted 2026-08-13 cs.CV cs.AI

Qiao Li , Xiaomeng Fu , Wangjia Yu , Runze He , Baisen Wang , Jiao Dai , Jizhong Han This is my paper · ORCID
classification cs.CVcs.AI
keywords copyrightprotectionconcepterasuretext-to-imagediffusionmodelsanimationcharacterssemanticanchorembeddingreplacementfine-grainedcontrolmodeltransferability
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Copyrighted animation characters can leak out of text-to-image models even when the model is not meant to reproduce them. This paper argues that the leak can be stopped at inference time, without altering weights, by replacing the character's token embeddings with a specially optimized 'anchor' embedding: a word-vector surrogate that keeps the character's outline and pose but drops its identifying details. The authors build a dataset of 80 characters, optimize one anchor per character, and report that after replacement, vision-language recognizers identify the character in only 6.0% of images (versus 66.9% before erasure), while image quality metrics stay close to the original model. If the method works as described, platforms could offer character-level opt-out filters on top of an existing diffusion model, with a dial for how strongly a character is suppressed and support for removing several characters at once.

What carries the argument

The central object is the optimized anchor embedding, a learnable vector named 'Anchor*' inserted into the text encoder's vocabulary and trained under two opposing losses: a structural loss that maximizes similarity of low-frequency outlines between images reconstructed from the target name and from the anchor, and a detail-differentiation loss that maximizes the anchor's noise-prediction error on a noised clean image of the target, preventing the anchor from reproducing identifying details. At inference, a structure-aware adaptive replacement monitors the L2 change of low-frequency components between consecutive denoising steps and swaps the target-related embeddings, together with the end-of-text and padding token embeddings fused by element-wise addition, for the anchor's embeddings only after the global layout has stabilized. This mechanism is what lets the method erase the character while preserving background and composition.

What would settle it

Generate images from a held-out set of prompts that describe each of the 80 characters without using their listed names, for example a paraphrase of the character's appearance or a translated alias, and run the same vision-language recognition checks. If recognition stays near the 66.9% and 64.7% base rates instead of the reported 6.0% and 4.0%, the method erases known tokens, not the underlying concept.

Watch

Extended reading notes

Core claim

In the paper's own terms, the central discovery is that a distinctive animation character can be erased from generated images by optimizing a single continuous anchor embedding, denoted 'Anchor*', that shares the target's coarse structural outline but is forced to differ in fine details, and then replacing the target-related textual embeddings with that anchor during denoising. The replacement is not applied from the first step: the method tracks low-frequency structural change across denoising timesteps and triggers embedding replacement once the layout has stabilized, which preserves scene coherence. On the paper's 80-character benchmark, recognition by two vision-language models drops from 66.9% to 6.0% and from 64.7% to 4.0%, while the Frechet Inception Distance stays at 33.2 versus 33.7 for the unmodified model. The same anchor embeddings can be reused across model versions or plugged into weight-modification erasure methods to improve their erasure.

Load-bearing premise

The load-bearing premise is that the target character is requested in the prompt by a name or term that has been pre-listed as target-related; if a user asks for the character through a paraphrase, a visual description, attributes, or another language, no embedding is replaced and the erasure does not happen.

Editorial extensions

If this is right

  • A platform can add per-character erasure filters to an already-deployed diffusion model by swapping embeddings at inference time, with no weight updates and nearly no change to normal generation quality.
  • Users can dial the erasure degree continuously: interpolating between anchor and target embeddings with a coefficient in [0,1] moves from full removal to a near-original image.
  • Multiple copyrighted characters can be removed from one scene simultaneously, because each character has its own anchor embedding and the replacements are applied together.
  • Anchors optimized for one model transfer to other models with the same text-encoder embedding dimension, including models with dual text encoders and a diffusion-transformer-based model, so per-character optimization is not repeated for every architecture.
  • The learned anchors can be dropped into existing weight-modification erasure methods as replacement anchor concepts, generally improving erasure accuracy and fidelity compared with generic anchors such as null text or 'toy'.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • One consequence the paper leaves implicit is that the erasure is lexical at its core: it fires only when the prompt contains a term that has been pre-listed as target-related. A natural stress test, not reported in the paper, would paraphrase or describe the character without naming it; if recognition returns to near-baseline, the filter is a vocabulary filter rather than semantic unlearning.
  • Because the replacement happens in embedding space and is reversible, an extension the paper does not pursue is an audit trail: one can inspect exactly which token was swapped. This suggests a design where platform policies expose the anchor and the interpolation dial to creators, turning copyright compliance into a tunable similarity knob rather than a binary block.
  • The same anchor-optimization recipe should transfer to other conditional generators that share a text-embedding interface, such as text-to-video or text-to-3D pipelines, whenever early generation steps are structure-dominated and later steps fill in details.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 7 minor

Summary. The paper proposes an inference-time method for erasing copyrighted animation characters from text-to-image diffusion models. The method first optimizes a new 'Anchor*' text embedding so that, when conditioned on it, the model produces images whose low-frequency structure resembles the target character but whose fine details differ. At generation time, the embedding of any pre-listed target term in the prompt is replaced by the optimized anchor embedding, and special EOT/Padding embeddings are fused with the anchor's special embeddings. A structure-aware adaptive schedule is used to trigger the replacement only after the low-frequency layout has stabilized. Experiments on a self-built 80-character benchmark with Stable Diffusion v1.4 report the lowest character recognition rates among baselines (6.0% with LLaVA-1.5 and 4.0% with BLIP-3 versus 66.9% and 64.7% for the base model), together with the best SSIM and LPIPS among baselines, plus additional results on controllable erasure degree, multi-target removal, transferability to other model families, and integration with model-modification baselines.

Significance. If the central claims are correct, the paper offers a useful and practical capability: a specific, named copyrighted character can be excised from generated images at inference time without retraining, while preserving the remainder of the scene. The paper's main strengths are the clarity of the core idea, the inclusion of ablations for each component, and the breadth of experiments covering different model architectures and integration with existing erasure methods. The reported effect sizes are large, and the qualitative examples are consistent with the erasure claim. However, the benchmark is self-built and unreleased, no code is provided, all numbers are single-run point estimates with no error bars, and the fidelity metric is computed against a character-present original rather than a context-only ground truth. The most consequential limitation is that the erasure mechanism only triggers on exact, pre-listed target terms in the prompt; the paper does not demonstrate concept-level erasure under paraphrases, aliases, attribute descriptions, or translations, which weakens the practical copyright-compliance claim.

major comments (5)
  1. [§3.3, Eq. (13)–(17)] The erasure mechanism is gated on the prompt containing "n target-related terms that are pre-listed in a subspace associated with copyrighted characters," but the paper never defines how this subspace is populated, how terms are detected, or how the method copes with paraphrases, aliases, visual descriptions, or non-English names. All quantitative results in Tables 1 and 2 are obtained from GPT-4o-generated prompts for 80 named characters, so those prompts by construction contain the canonical name. The reported 6.0% and 4.0% recognition rates therefore demonstrate erasure of an exact lexical token, not erasure of the visual concept under arbitrary user prompts. Since the abstract and introduction claim the method can "erase target characters during generation" for copyright compliance, this lexical gating is a load-bearing assumption, and it is currently untested in the paper.
  2. [§4.2, Table 1] Image fidelity preservation is measured by SSIM and LPIPS between the original (character-present) image and the erased image. Because the target character typically occupies a substantial image region, these metrics necessarily penalize the desired removal of the character itself and cannot separate 'preserving the context' from 'retaining the character.' The paper should compare against a context-only ground truth, for example by computing fidelity on the background or masked-out regions, or by evaluating prompts where the character is absent from the reference. As reported, the SSIM/LPIPS values do not by themselves establish the abstract's claim of high fidelity preservation.
  3. [§4.2, 'Impact on Unrelated Image Quality' and Table 1] The FID and CLIP-score check on COCO-30K is uninformative in its current form: those prompts do not contain the pre-listed target terms, so the proposed method should be a no-op, yet the reported FID differs from the base model (33.2 versus 33.7) and CLIP score differs (0.312 versus 0.326). The paper does not explain this discrepancy or report the sampling protocol, random seeds, or variance. Without this, the reader cannot tell whether the difference is sampling noise or whether the method perturbs generation even when no target is present.
  4. [§3.1 and §3.3] Several hyperparameters and components that are essential for reproducing the method are not specified: the structural and detail loss weights α and β in Eq. (5); the special-embedding fusion weights λ1 and λ2 in Eqs. (15)–(16); the timestep bounds (T_M, T_H) and (T_L, T_M) for structural and detail optimization; the low-frequency filter f_L used in Eqs. (8) and (18); and the threshold that triggers adaptive replacement. The adaptive replacement strategy is justified by a single denoising trajectory shown in Figure 3, but no evidence is provided that the transition point is consistent across characters, prompts, or model versions. Without concrete values and a robustness analysis, the central mechanism is not reproducible and the threshold claim is not falsifiable.
  5. [§4.8, Table 4] The claim that the optimized anchors are 'plug-and-play' and 'generally improve' existing model-modification baselines is weakened by the MACE row: erasure accuracy with the proposed anchor is 17.5% LLaVA-1.5, which is substantially worse than the 9.0% obtained with the general anchor. The paper should either report the full per-baseline behavior and explain why MACE degrades, or qualify the claim so that it is not contradicted by one of the four tested baselines.
minor comments (7)
  1. [Abstract] There is a grammatical error in 'We optimizes an anchor embedding'; it should be 'We optimize' or 'The method optimizes.'
  2. [§3.3] The phrase 'an structure-aware adaptive replacement' should be 'a structure-aware adaptive replacement.'
  3. [§1, Contributions] The list says 'achieves state-of-the-arts' which should be 'state-of-the-art results' or similar.
  4. [References] Reference [25] is cited as BLIP-3, but the listed title is an RGB-T tracking paper; this appears to be an incorrect reference and should be corrected.
  5. [Figure 1 caption] The caption runs several character names together without spacing ('Spider ManKung Fu PandaMinions'), which makes the figure difficult to read.
  6. [§4.1] The description of evaluation metrics should state explicitly what reference images are used for SSIM and LPIPS, and whether the reported values are averaged over all 8,000 generated images or per character; the current wording is ambiguous.
  7. [§4.2] The sentence 'as it effectively deceive multi-modal large models' should be 'as it effectively deceives multimodal large models' or similar.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the anchor is an optimized artifact, and the central erasure claim is verified by external vision-language models rather than by construction.

full rationale

The paper's central claim is an empirical method result, not a derived prediction. The anchor embedding is optimized via Eq. 5-11 to make an 'Anchor*' prompt structurally similar but detail-different from the target, and Eq. 13-17 replace target-token embeddings with this anchor at inference. Erasure effectiveness is then measured by LLaVA-1.5 and BLIP-3, external models that play no role in anchor optimization, while fidelity is checked with SSIM, LPIPS, Aesthetic, FID, and CLIP. There is no self-citation chain and no fitted parameter is renamed as a prediction. The exact-token gating in Sec. 3.3 is a coverage limitation (paraphrases, aliases, and non-English descriptions are untested), but it is not a circular reduction because the paper never derives paraphrase robustness from its equations. The anchor is a fitted artifact, but the state-of-the-art claim is checked against external benchmarks, so the derivation is self-contained.

Assumptions & free parameters 7 free parameters · 5 assumptions · 1 invented entities

The central pipeline rests on empirically tuned loss weights, an unspecified replacement threshold, and token-level target detection. No physical entities are introduced, but the Anchor* embedding is an invented, unreleased learned object whose behavior is only evaluated inside the paper's own pipeline.

free parameters (7)
  • alpha (structural loss weight) = not reported
    Weight on structural outline loss L_S in Equations 5 and 11; the paper says it is determined empirically.
  • beta (detail loss weight) = not reported
    Weight on detailed differentiation loss L_D in Equations 5 and 11; determined empirically.
  • lambda_1 (original special embedding weight) = not reported
    Weight for original EOT and Padding embeddings in Equations 15 and 16; determined empirically.
  • lambda_2 (anchor special embedding weight) = not reported
    Weight for optimized EOT and Padding embeddings in Equations 15 and 16; determined empirically.
  • Timestep bounds for anchor optimization = not reported
    Sampling ranges (T_M, T_H) for structural loss and (T_L, T_M) for detail loss are not specified, which changes what content the anchor captures.
  • Adaptive replacement threshold = not reported
    The threshold that triggers embedding replacement is motivated by one observed trajectory around t=720 and applied globally without specification.
  • Low-frequency filter f_L = not reported
    The filter used to extract structural components in Equations 8 and 18 has no implementation or cutoff parameters.
assumptions (5)
  • standard math The DDIM deterministic denoising recurrence and the diffusion noise-prediction objective describe the model's behavior (Equations 1 to 4).
    Background from Ho et al. 2020 and Song et al. 2020; used throughout the method.
  • domain assumption CLIP textual embeddings are sufficiently disentangled that replacing only target-related token embeddings leaves unrelated context semantically unchanged.
    Needed for targeted replacement in Equations 13 to 17; the paper cites additivity results but does not verify locality per character.
  • domain assumption Early denoising timesteps encode global structure while later timesteps add details.
    Basis for choosing large timesteps for the structural loss and small timesteps for the detail loss; supported by cited prior work [18,30,39].
  • ad hoc to paper A single transition point in low-frequency L2 change, observed on one denoising trajectory, is a reliable global trigger for replacement across all characters and prompts.
    Section 3.3 presents one example with a peak around t=720 and generalizes it to all cases without an ablation on the threshold.
  • domain assumption User prompts contain exactly the pre-listed target terms, so paraphrase handling is unnecessary.
    Section 3.3 restricts replacement to pre-listed terms and the evaluation does not test paraphrases or aliases.
invented entities (1)
  • Anchor* token and its optimized embedding vector v*
    purpose: Acts as a benign surrogate for a copyrighted character during embedding replacement (Equations 11 to 14).
    It is an invented, per-character learnable vector optimized against the target diffusion model; no public release, no external falsifiable prediction, and no handle outside the paper's own pipeline.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Erase but Preserve: Controllable Removal of Copyrighted Animation Characters via Optimized Semantic Anchors." pith.science (2026). https://pith.science/paper/Z4A37KC6

@misc{pith2026260812806,
  author       = {Pith},
  title        = {Pith review of: Erase but Preserve: Controllable Removal of Copyrighted Animation Characters via Optimized Semantic Anchors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z4A37KC6}},
  note         = {Machine review of arXiv:2608.12806}
}
read the original abstract

The exceptional generation capabilities of text-to-image diffusion models have raised copyright concerns, particularly the unauthorized reproduction of animation characters. Existing concept erasure methods fall short for animation character erasure: model modification methods struggle to identify suitable anchors for diverse, highly distinctive characters; prompt-based steering methods lack fine-grained control for precise intervention. These approaches often yield incomplete erasure and degraded image fidelity, hindering real-world deployment. In this paper, we propose a controllable method operating on the model's continuous textual representation to erase target characters during generation. We optimizes an anchor embedding via structural and detailed constraints to serve as a character surrogate, then replaces target-related embeddings with the anchor via a structure-aware adaptive strategy. Experiments show that our method achieves state-of-the-art erasure effectiveness and image fidelity preservation, while supporting controllable erasure degree, multi-target removal, and model transferability. Moreover, our optimized anchors are plug-and-play with current model modification baselines to improve their erasure performance.

Figures

Figures reproduced from arXiv: 2608.12806 by the authors.

Figure 1
Figure 1. During image generation, our method effectively erases diverse animation concepts using optimized semantic anchors, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The overall pipeline of our proposed method. We construct an anchor by applying structural constraints (top-left) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the structure-aware adaptive replace [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison between our method and baseline methods. Our method completely erases target animation [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 6
Figure 6. Figure 6: Left: original images (target), erased version (an [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 5
Figure 5. Figure 5: Visualization of ablation results on key modules. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 8
Figure 8. Figure 8: Simultaneous erasure of multiple characters. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 8 canonical work pages

  1. [1]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Floren- cia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report.arXiv preprint arXiv:2303.08774 (2023)

    arXiv 2023

  2. [2]

    Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Aittala, Timo Aila, Samuli Laine, et al. 2022. ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers.arXiv preprint arXiv:2211.01324(2022)

    arXiv 2022

  3. [3]

    BBC News. 2025. Disney and Universal sue AI firm Midjourney over images. https://www.bbc.com/news/articles/cg5vjqdm1ypo. Accessed: 2025-07-1

    work page 2025

  4. [4]

    Zachary Bozard. 2023. What does it mean to create art? Intellectual Property rights for Artificial Intelligence generated artworks.SCJ Int’l L. & Bus.20 (2023), 83

    work page 2023

  5. [5]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners.Advances in neural information processing systems33 (2020), 1877–1901

    2020

  6. [6]

    Anh Bui, Long Vuong, Khanh Doan, Trung Le, Paul Montague, Tamas Abraham, and Dinh Phung. 2024. Erasing undesirable concepts in diffusion models with adversarial preservation.arXiv preprint arXiv:2410.15618(2024)

    arXiv 2024

  7. [7]

    Prafulla Dhariwal and Alexander Nichol. 2021. Diffusion models beat gans on image synthesis.Advances in neural information processing systems34 (2021), 8780–8794

    2021

  8. [8]

    Rohit Gandikota, Joanna Materzynska, Jaden Fiotto-Kaufman, and David Bau

Show all 52 references
  1. [9]

    Rohit Gandikota, Hadas Orgad, Yonatan Belinkov, Joanna Materzyńska, and David Bau. 2024. Unified concept editing in diffusion models. InProceedings of the IEEE/CVF winter conference on applications of computer vision. 5111–5120

    2024

  2. [10]

    Chao Gong, Kai Chen, Zhipeng Wei, Jingjing Chen, and Yu-Gang Jiang. 2024. Re- liable and efficient concept erasure of text-to-image diffusion models. InEuropean Conference on Computer Vision. Springer, 73–88

    2024

  3. [11]

    Matt Growcoot. 2022. Midjourney founder admits to using a ‘hundred mil- lion’images without consent.PetaPixel, Dec(2022)

    2022

  4. [12]

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. 2017. Gans trained by a two time-scale update rule converge to a local nash equilibrium.Advances in neural information processing systems30 (2017)

    2017

  5. [13]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models.Advances in neural information processing systems33 (2020), 6840–6851

    2020

  6. [14]

    Jonathan Ho and Tim Salimans. 2022. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598(2022)

    2022 arXiv

  7. [15]

    Taihang Hu, Linxuan Li, Joost Van de Weijer, Hongcheng Gao, Fahad S Khan, Jian Yang, Ming-Ming Cheng, Kai Wang, and Yaxing Wang. 2024. Token merging for training-free semantic binding in text-to-image synthesis.Advances in Neural Information Processing Systems37 (2024), 137646–137672

    2024

  8. [16]

    Anubhav Jain, Yuya Kobayashi, Takashi Shibuya, Yuhta Takida, Nasir Memon, Julian Togelius, and Yuki Mitsufuji. 2024. Trasce: Trajectory steering for concept erasure.arXiv preprint arXiv:2412.07658(2024)

    2024 arXiv

  9. [17]

    Nupur Kumari, Bingliang Zhang, Sheng-Yu Wang, Eli Shechtman, Richard Zhang, and Jun-Yan Zhu. 2023. Ablating concepts in text-to-image diffusion models. In Proceedings of the IEEE/CVF international conference on computer vision. 22691– 22702

    2023

  10. [18]

    Mingi Kwon, Jaeseok Jeong, and Youngjung Uh. 2022. Diffusion models already have a semantic latent space.arXiv preprint arXiv:2210.10960(2022)

    2022 arXiv

  11. [19]

    LAION-AI. 2022. aesthetic-predictor. https://github.com/LAION-AI/aesthetic- predictor

    2022

  12. [20]

    Ouxiang Li, Yuan Wang, Xinting Hu, Houcheng Jiang, Tao Liang, Yanbin Hao, Guojun Ma, and Fuli Feng. 2025. Speed: Scalable, precise, and efficient concept erasure for diffusion models.arXiv preprint arXiv:2503.07392(2025)

    2025

  13. [21]

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. 2014. Microsoft coco: Common objects in context. InEuropean conference on computer vision. Springer, 740–755

    2014

  14. [22]

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. 2022. Flow matching for generative modeling. InThe eleventh international conference on learning representations

    2022

  15. [23]

    Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. 2024. Improved baselines with visual instruction tuning. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 26296–26306

    2024

  16. [24]

    Xingchao Liu, Chengyue Gong, and Qiang Liu. 2023. Flow straight and fast: Learn- ing to generate and transfer data with rectified flow. InInternational conference on learning representations (ICLR)

    2023

  17. [25]

    Andong Lu, Wanyu Wang, Chenglong Li, Jin Tang, and Bin Luo. 2024. Rgbt tracking via all-layer multimodal interactions with progressive fusion mamba. arXiv preprint arXiv:2408.08827(2024)

    2024 arXiv

  18. [26]

    Shilin Lu, Zilan Wang, Leyang Li, Yanzhu Liu, and Adams Wai-Kin Kong. 2024. Mace: Mass concept erasure in diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 6430–6440

    2024

  19. [27]

    Yiwei Lu, Matthew YR Yang, Zuoqiu Liu, Gautam Kamath, and Yaoliang Yu. 2024. Disguised copyright infringement of latent diffusion models.arXiv preprint arXiv:2404.06737(2024)

    2024 arXiv

  20. [28]

    Mengyao Lyu, Yuhong Yang, Haiwen Hong, Hui Chen, Xuan Jin, Yuan He, Hui Xue, Jungong Han, and Guiguang Ding. 2024. One-dimensional adapter to rule them all: Concepts diffusion models and erasing applications. InProceedings of the IEEE/CVF Conference on Computer Vision and Patt...

    2024

  21. [29]

    Byeonghu Na, Mina Kang, Jiseok Kwak, Minsang Park, Jiwoo Shin, SeJoon Jun, Gayoung Lee, Jin-Hwa Kim, and Il-Chul Moon. 2026. Training-free safe text embedding guidance for text-to-image diffusion models.Advances in Neural Information Processing Systems38 (2026), 85984–86014

    2026

  22. [30]

    Ji-Hoon Park, Yeong-Joon Ju, and Seong-Whan Lee. 2024. Explaining generative diffusion models via visual analysis for interpretable decision-making process. Expert Systems with Applications248 (2024), 123231

    2024

  23. [31]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. InInternational conference on machine learnin...

    2021

  24. [32]

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen

  25. [33]

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

    2022

  26. [34]

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. 2022. Photorealistic text-to-image diffusion models with deep language understanding.Advances in neural inform...

    2022

  27. [35]

    Patrick Schramowski, Manuel Brack, Björn Deiseroth, and Kristian Kersting

  28. [36]

    Jiaming Song, Chenlin Meng, and Stefano Ermon. 2020. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502(2020)

    2020 arXiv

  29. [37]

    Yang Song and Stefano Ermon. 2019. Generative modeling by estimating gradients of the data distribution.Advances in neural information processing systems32 (2019)

    2019

  30. [38]

    InProceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Safe latent diffusion: Mitigating inappropriate degeneration in diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 22522–22531

  31. [39]

    Jianyi Wang, Zongsheng Yue, Shangchen Zhou, Kelvin CK Chan, and Chen Change Loy. 2024. Exploiting diffusion prior for real-world image super- resolution.International Journal of Computer Vision132, 12 (2024), 5929–5949

    2024

  32. [40]

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. 2004. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing13, 4 (2004), 600–612

    2004

  33. [41]

    Z-Image Team. 2025. Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer.arXiv preprint arXiv:2511.22699(2025)

    2025 arXiv

  34. [42]

    Gong Zhang, Kai Wang, Xingqian Xu, Zhangyang Wang, and Humphrey Shi

  35. [43]

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang

  36. [44]

    Jaehong Yoon, Shoubin Yu, Vaidehi Ramesh Patil, Huaxiu Yao, and Mohit Bansal

  37. [45]

    Yang Zhang, Teoh Tze Tzun, Lim Wei Hern, and Kenji Kawaguchi. 2024. On copyright risks of text-to-image diffusion models. InECCV 2024 Workshop The Dark Side of Generative AIs and Beyond

    2024

  38. [46]

    Chenyi Zhuang, Ying Hu, and Pan Gao. 2024. Magnet: We never know how text-to-image diffusion models work, until we learn how vision-language models function.Advances in Neural Information Processing Systems37 (2024), 57115– 57149

    2024

  39. [50]

    Tong Zhang, Ru Zhang, Jianyi Liu, Zhen Yang, and Gongshen Liu. 2025. Beyond Fixed Anchors: Precisely Erasing Concepts with Sibling Exclusive Counterparts. arXiv preprint arXiv:2510.16342(2025)

    2025

  40. [2018]

    InProceedings of the IEEE conference on computer vision and pattern recognition

    The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE conference on computer vision and pattern recognition. 586–595

  41. [2022]

    Hierarchical text-conditional image generation with clip latents.arXiv preprint arXiv:2204.061251, 2 (2022), 3

    2022 arXiv

  42. [2023]

    InProceedings of the IEEE/CVF international conference on computer vision

    Erasing concepts from diffusion models. InProceedings of the IEEE/CVF international conference on computer vision. 2426–2436

  43. [2024]

    In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    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. 1755–1764

  44. [2025]

    InInternational Conference on Learning Representations, Vol

    Safree: Training-free and adaptive guard for safe text-to-image and video generation. InInternational Conference on Learning Representations, Vol. 2025. 56439–56465

    2025

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.