pith. sign in

arxiv: 2604.06063 · v2 · submitted 2026-04-04 · 💻 cs.CV · cs.MM

EDGE-Shield: Efficient Denoising-staGE Shield for Violative Content Filtering via Scalable Reference-Based Matching

Takara Taniguchi , Ryohei Shimizu , Duc Minh Vo , Kota Izumi , Shiqi Yang , Teppei Suzuki This is my paper

Pith reviewed 2026-05-13 17:48 UTC · model grok-4.3

classification 💻 cs.CV cs.MM
keywords content filteringtext-to-image generationdenoising processreference-based matchingviolative contentlatency reductionembedding matchingx-pred transformation
0
0 comments X

The pith

EDGE-Shield filters violative content in text-to-image models by matching reference embeddings on transformed early-stage latents instead of waiting for finished images.

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

Text-to-image generators can produce copyrighted or harmful images, but existing reference-based filters must wait until generation completes and struggle when many references are involved. EDGE-Shield performs the check inside the denoising process itself by comparing embeddings after an x-pred transformation that turns noisy latents into pseudo-clean estimates. This change allows detection at earlier steps while keeping the same accuracy as full-image checks. The result is much lower latency across different generator architectures. A sympathetic reader would care because the method makes scalable, up-to-date protection against new copyrighted works practical without adding noticeable delay to generation.

Core claim

The paper presents EDGE-Shield as a denoising-stage content filter that applies embedding-based reference matching to x-pred transformed noisy latents. The x-pred step converts each intermediate noisy representation into an estimate of the clean latent that would appear at a later denoising step. Experiments on Z-Image-Turbo and Qwen-Image show this yields an approximate 79 percent reduction in processing time for the first model and 50 percent for the second while preserving filtering accuracy.

What carries the argument

The x-pred transformation, which converts a model's noisy intermediate latent into a pseudo-estimated clean latent from a later denoising stage, paired with embedding-based reference matching performed at early steps.

If this is right

  • Filtering can occur without completing image generation, so violative outputs are blocked before they are produced.
  • The method scales to large numbers of reference images because embedding comparisons replace slower pixel-level or feature-level checks.
  • Accuracy holds across different generator architectures, so the same filter can protect multiple models.
  • Protection remains current as new copyrighted works appear, since references can be added without retraining.

Where Pith is reading between the lines

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

  • The approach could be extended to video or audio generators by applying the same early-stage transformation and matching logic.
  • Integration into live generation APIs might allow real-time policy enforcement without user-visible slowdown.
  • If early detection proves stable, similar latent-space checks could be tested for other safety properties beyond copyright, such as prohibited visual styles.

Load-bearing premise

Matching embeddings on the x-pred transformed noisy latent at early denoising stages can detect violative content reliably without needing the completed image or producing too many false positives.

What would settle it

A direct comparison on a held-out set of violative and non-violative prompts showing that early-stage x-pred matching either misses a substantial fraction of violations or produces materially higher false-positive rates than waiting for the finished image.

Figures

Figures reproduced from arXiv: 2604.06063 by Duc Minh Vo, Kota Izumi, Ryohei Shimizu, Shiqi Yang, Takara Taniguchi, Teppei Suzuki.

Figure 1
Figure 1. Figure 1: Comparison of our EDGE-Shield with the existing content filters [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of embedding caching and the detail component of EDGE-Shield. Left: We compute the embeddings of target references of violative content. Right: we transform the intermediate latent into the clean latent by using our x-pred transforma￾tion. EDGE-Shield calculates the similarity score among the embedding of the decoded clean latent and cached embeddings. 4.1 Task Definition We formulate the safety a… view at source ↗
Figure 3
Figure 3. Figure 3: Evaluation results of ROC-AUC, Latency, and the number of references on Z-Image-Turbo. (a) Relationship between the number of references and the ROC-AUC score. (b) Relationship between the number of references and the latency time. 0.0 0.2 0.4 0.6 0.8 1.0 Recall 0.2 0.4 0.6 0.8 1.0 Precision LLaVa-NEXT InternVL3.5 Qwen3-VL Ours (Q3VLEmbed) Random Classifier Baseline (a) Z-Image-Turbo 0.0 0.2 0.4 0.6 0.8 1.… view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of Precision-Recall curve. performance of the highly capable Qwen3-VL, demonstrating its superior ability to accurately identify violative content. More importantly, [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: ROC-AUC visualization of x-pred transformation ablation. This figure shows the ROC-AUC scores across each timesteps for three different models. Timestep (T) 1 2 3 4 5 6 7 8 9 Vanilla D(zt) x-pred D(xθ(zt, t)) [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualized intermediate latents of Z-Image-Turbo, comparing states with and without the x-pred transformation across 9 generation timesteps. ingful enhancement only becomes apparent between steps T = 20 and T = 30, rather than within the T = 1 to T = 10 range. Furthermore, a slight degradation in quality is observed in the final steps of the x-pred transformation. Encoder-wise Analysis [PITH_FULL_IMAGE:fi… view at source ↗
Figure 7
Figure 7. Figure 7: Visualized intermediate latents of Qwen-Image, comparing states with and without the x-pred transformation across 50 generation timesteps [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Accuracy across thresholds from 0.1 to 0.9. (a) Results on Z-Image-Turbo at timestep 1 (b) Results on Qwen-Image at timestep 25 [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualized intermediate latents of SD1.4, comparing states with and without the x-pred transformation across 40 generation timesteps [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
read the original abstract

The advent of Text-to-Image generative models poses significant risks of copyright violation and deepfake generation. Since the rapid proliferation of new copyrighted works and private individuals constantly emerges, reference-based training-free content filters are essential for providing up-to-date protection without the constraints of a fixed knowledge cutoff. However, existing reference-based approaches often lack scalability when handling numerous references and require waiting for finishing image generation. To solve these problems, we propose EDGE-Shield, a scalable content filter during the denoising process that maintains practical latency while effectively blocking violative content. We leverage embedding-based matching for efficient reference comparison. Additionally, we introduce an \textit{$x$}-pred transformation that converts the model's noisy intermediate latent into the pseudo-estimated clean latent at the later stage, enhancing classification accuracy of violative content at earlier denoising stages. We conduct experiments of violative content filtering against two generative models including Z-Image-Turbo and Qwen-Image. EDGE-Shield significantly outperforms traditional reference-based methods in terms of latency; it achieves an approximate $79\%$ reduction in processing time for Z-Image-Turbo and approximate $50\%$ reduction for Qwen-Image, maintaining the filtering accuracy across different model architectures.

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.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper describes an empirical engineering method for early-stage violative content filtering in diffusion models using embedding-based matching and an x-pred transformation on noisy latents. No equations, derivations, or first-principles claims are presented that reduce the reported latency reductions (79% for Z-Image-Turbo, 50% for Qwen-Image) or accuracy maintenance to fitted parameters, self-definitions, or self-citation chains by construction. The x-pred step is introduced as a practical heuristic to improve early-stage classification without any mathematical equivalence to the target performance metrics, and performance is validated experimentally across model architectures rather than derived from inputs. The contribution remains self-contained as an applied adaptation of reference-based matching.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, background axioms, or postulated entities; the x-pred transformation is presented as a technical device rather than a new physical or mathematical entity.

pith-pipeline@v0.9.0 · 5537 in / 1126 out tokens · 76188 ms · 2026-05-13T17:48:55.418750+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

47 extracted references · 47 canonical work pages

  1. [1]

    In: Computing Research Repository (CoRR) (2025)

    Bai, S., Cai, Y., Chen, R., Chen, K., Chen, X., Cheng, Z., Deng, L., Ding, W., Gao, C., Ge, C., Ge, W., Guo, Z., Huang, Q., Huang, J., Huang, F., Hui, B., Jiang, S., Li, Z., Li, M., Li, M., Li, K., Lin, Z., Lin, J., Liu, X., Liu, J., Liu, C., Liu, Y., Liu, D., Liu, S., Lu, D., Luo, R., Lv, C., Men, R., Meng, L., Ren, X., Ren, X., Song, S., Sun, Y., Tang, ...

    work page 2025

  2. [2]

    Biswas, S.D., Roy, A., Roy, K.: CURE: Concept unlearning via orthogonal rep- resentation editing in diffusion models. In: Adv. Neural Inform. Process. Syst. (NeurIPS) (2025)

    work page 2025

  3. [3]

    In: Proceedings of the 32nd USENIX Conference on Security Symposium (USENIX) (2023)

    Carlini, N., Hayes, J., Nasr, M., Jagielski, M., Sehwag, V., Tramèr, F., Balle, B., Ippolito, D., Wallace, E.: Extracting training data from diffusion models. In: Proceedings of the 32nd USENIX Conference on Security Symposium (USENIX) (2023)

    work page 2023

  4. [4]

    In: Computing Research Repository (CoRR) (2024)

    Chi, J., Karn, U., Zhan, H., Smith, E., Rando, J., Zhang, Y., Plawiak, K., Coudert, Z.D., Upasani, K., Pasupuleti, M.: Llama guard 3 vision: Safeguarding human-ai image understanding conversations. In: Computing Research Repository (CoRR) (2024)

    work page 2024

  5. [5]

    Cywiński, B., Deja, K.: SAeuron: Interpretable concept unlearning in diffusion models with sparse autoencoders. In: Int. Conf. Mach. Learn. (ICML) (2025)

    work page 2025

  6. [6]

    Gaintseva, T., Oncescu, A.M., Ma, C., Liu, Z., Benning, M., Slabaugh, G., Deng, J., Elezi, I.: CASteer: Cross-attention steering for controllable concept erasure. In: Int. Conf. Learn. Represent. (ICLR) (2026)

    work page 2026

  7. [7]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) (2024)

    Gandikota, R., Orgad, H., Belinkov, Y., Materzyńska, J., Bau, D.: Unified concept editing in diffusion models. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) (2024)

    work page 2024

  8. [8]

    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: Int. Conf. Mach. Learn. (ICML) (2024)

    work page 2024

  9. [9]

    Helff, L., Friedrich, F., Brack, M., Kersting, K., Schramowski, P.: Llavaguard: An open VLM-based framework for safeguarding vision datasets and models. In: Int. Conf. Mach. Learn. (ICML) (2025)

    work page 2025

  10. [10]

    In: Computing Research Repository (CoRR) (2023)

    Inan, H., Upasani, K., Chi, J., Rungta, R., Iyer, K., Mao, Y., Tontchev, M., Hu, Q., Fuller, B., Testuggine, D., Khabsa, M.: Llama guard: Llm-based input-output safeguard for human-ai conversations. In: Computing Research Repository (CoRR) (2023)

    work page 2023

  11. [11]

    In: Computing Research Repository (CoRR) (2025)

    Kim, D., Ghadiyaram, D.: Concept steerers: Leveraging k-sparse autoencoders for test-time controllable generations. In: Computing Research Repository (CoRR) (2025)

    work page 2025

  12. [12]

    Kim, M., Kim, D., Yusuf, A., Ermon, S., Park, M.: Training-free safe denoisers for safe use of diffusion models. In: Adv. Neural Inform. Process. Syst. (NeurIPS) (2025)

    work page 2025

  13. [13]

    In: Proceedings of the 37th Interna- tional Conference on Neural Information Processing Systems (ICONIP) (2023)

    Kotar, K., Tian, S., Yu, H.X., Yamins, D.L.K., Wu, J.: Are these the same apple? comparing images based on object intrinsics. In: Proceedings of the 37th Interna- tional Conference on Neural Information Processing Systems (ICONIP) (2023)

    work page 2023

  14. [14]

    In: Symposium on Operating Systems Principles (SOSP) (2023)

    Kwon, W., Li, Z., Zhuang, S., Sheng, Y., Zheng, L., Yu, C.H., Gonzalez, J., Zhang, H., Stoica, I.: Efficient memory management for large language model serving with pagedattention. In: Symposium on Operating Systems Principles (SOSP) (2023)

    work page 2023

  15. [15]

    In: Computing Research Repository (CoRR) (2025) 16 F

    Labs,B.F.,Batifol,S.,Blattmann,A.,Boesel,F.,Consul,S.,Diagne,C.,Dockhorn, T., English, J., English, Z., Esser, P., Kulal, S., Lacey, K., Levi, Y., Li, C., Lorenz, D., Müller, J., Podell, D., Rombach, R., Saini, H., Sauer, A., Smith, L.: Flux.1 kontext: Flow matching for in-context image generation and editing in latent space. In: Computing Research Reposi...

    work page 2025

  16. [16]

    In: IEEE Conf

    Li, L., Shi, Z., Hu, X., Dong, B., Qin, Y., Liu, X., Sheng, L., Shao, J.: T2isafety: Benchmark for assessing fairness, toxicity, and privacy in image generation. In: IEEE Conf. Comput. Vis. Pattern Recog. (CVPR) (2025)

    work page 2025

  17. [17]

    In: Computing Research Repository (CoRR) (2026)

    Li, M., Zhang, Y., Long, D., Chen, K., Song, S., Bai, S., Yang, Z., Xie, P., Yang, A., Liu, D., Zhou, J., Lin, J.: Qwen3-vl-embedding and qwen3-vl-reranker: A unified framework for state-of-the-art multimodal retrieval and ranking. In: Computing Research Repository (CoRR) (2026)

    work page 2026

  18. [18]

    In: Com- puting Research Repository (CoRR) (2026)

    Li, T., He, K.: Back to basics: Let denoising generative models denoise. In: Com- puting Research Repository (CoRR) (2026)

    work page 2026

  19. [19]

    In: Computing Research Repository (CoRR) (2024)

    Liu, H., Li, C., Li, Y., Li, B., Zhang, Y., Shen, S., Lee, Y.J.: Llava-next: Improved reasoning, ocr, and world knowledge. In: Computing Research Repository (CoRR) (2024)

    work page 2024

  20. [20]

    Liu,H.,Li,C.,Wu,Q.,Lee,Y.J.:Visualinstructiontuning.In:Adv.NeuralInform. Process. Syst. (NeurIPS) (2023)

    work page 2023

  21. [21]

    In: Computing Research Repository (CoRR) (2026)

    Liu, M., Zhang, S., Long, C.: Wukong framework for not safe for work detection in text-to-image systems. In: Computing Research Repository (CoRR) (2026)

    work page 2026

  22. [22]

    Liu, R., Khakzar, A., Gu, J., Chen, Q., Torr, P., Pizzati, F.: Latent guard: A safety framework for text-to-image generation. In: Eur. Conf. Comput. Vis. (ECCV) (2024)

    work page 2024

  23. [23]

    In: ACM Int

    Liu, S., Shi, Z., Lyu, L., Jin, Y., Faltings, B.: Copyjudge: Automated copyright infringement identification and mitigation in text-to-image diffusion models. In: ACM Int. Conf. Multimedia (ACMMM) (2025)

    work page 2025

  24. [24]

    In: IEEE Conf

    Lu, S., Wang, Z., Li, L., Liu, Y., Kong, A.W.K.: Mace: Mass concept erasure in diffusion models. In: IEEE Conf. Comput. Vis. Pattern Recog. (CVPR) (2024)

    work page 2024

  25. [25]

    In: Computing Research Repository (CoRR) (2024)

    Ma, R., Zhou, Q., Jin, Y., Zhou, D., Xiao, B., Li, X., Qu, Y., Singh, A., Keutzer, K., Hu, J., Xie, X., Dong, Z., Zhang, S., Zhou, S.: A dataset and benchmark for copy- right infringement unlearning from text-to-image diffusion models. In: Computing Research Repository (CoRR) (2024)

    work page 2024

  26. [26]

    In: Association for the Advancement of Artificial Intelligence (AAAI) (2023)

    Markov, T., Zhang, C., Agarwal, S., Eloundou Nekoul, F., Lee, T., Adler, S., Jiang, A., Weng, L.: A holistic approach to undesired content detection in the real world. In: Association for the Advancement of Artificial Intelligence (AAAI) (2023)

    work page 2023

  27. [27]

    Moon, S., Lee, M., Park, S., Kim, D.: Holistic unlearning benchmark: A multi- facetedevaluationfortext-to-imagediffusionmodelunlearning.In:Int.Conf.Com- put. Vis. (ICCV) (2025)

    work page 2025

  28. [28]

    In: Computing Research Repository (CoRR) (2024)

    OpenAI, :, Hurst, A., Lerer, A., Goucher, A.P., Perelman, A., Ramesh, A., Clark, A., Ostrow, A., Welihinda, A., Hayes, A., Radford, A., Mądry, A., Baker- Whitcomb,A.,Beutel,A.,Borzunov,A.,Carney,A.,Chow,A.,Kirillov,A.,Nichol, A., Paino, A., Renzin, A., Passos, A.T., Kirillov, A., Christakis, A., Conneau, A., Kamali, A., Jabri, A., Moyer, A., Tam, A., Croo...

    work page 2024

  29. [29]

    In: IEEE Conf

    Pizzi, E., Roy, S.D., Ravindra, S.N., Goyal, P., Douze, M.: A self-supervised de- scriptor for image copy detection. In: IEEE Conf. Comput. Vis. Pattern Recog. 18 F. Author et al. (CVPR) (2022)

    work page 2022

  30. [30]

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., Müller, J., Penna, J., Rombach, R.: SDXL: Improving latent diffusion models for high-resolution im- age synthesis. In: Int. Conf. Learn. Represent. (ICLR) (2024)

    work page 2024

  31. [31]

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., Sutskever, I.: Learning transfer- able visual models from natural language supervision. In: Int. Conf. Mach. Learn. (ICML) (2021)

    work page 2021

  32. [32]

    In: Computing Research Repository (CoRR) (2022)

    Ramesh, A., Dhariwal, P., Nichol, A., Chu, C., Chen, M.: Hierarchical text- conditional image generation with clip latents. In: Computing Research Repository (CoRR) (2022)

    work page 2022

  33. [33]

    In: NeurIPS ML Safety Workshop (2022)

    Rando, J., Paleka, D., Lindner, D., Heim, L., Tramer, F.: Red-teaming the stable diffusion safety filter. In: NeurIPS ML Safety Workshop (2022)

    work page 2022

  34. [34]

    In: IEEE Conf

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: IEEE Conf. Comput. Vis. Pattern Recog. (CVPR) (2022)

    work page 2022

  35. [35]

    Schramowski, P., Brack, M., Deiseroth, B., Kersting, K.: Safe latent diffusion: Mitigatinginappropriatedegenerationindiffusionmodels.In:IEEEConf.Comput. Vis. Pattern Recog. (CVPR) (2023)

    work page 2023

  36. [36]

    Shi, Z., Yan, J., Tang, X., Lyu, L., Faltings, B.: Rlcp: A reinforcement learning- based copyright protection method for text-to-image diffusion model. In: Int. Conf. Multimedia and Expo (ICME) (2025)

    work page 2025

  37. [37]

    Song, Y., Liu, X., Shou, M.Z.: Diffsim: Taming diffusion models for evaluating visual similarity. In: Int. Conf. Comput. Vis. (ICCV) (2025)

    work page 2025

  38. [38]

    In: Computing Research Repository (CoRR) (2025)

    Team, I., Cai, H., Cao, S., Du, R., Gao, P., Hoi, S., Hou, Z., Huang, S., Jiang, D., Jin, X., Li, L., Li, Z., Li, Z.Y., Liu, D., Liu, D., Shi, J., Wu, Q., Yu, F., Zhang, C., Zhang, S., Zhou, S.: Z-image: An efficient image generation foundation model with single-stream diffusion transformer. In: Computing Research Repository (CoRR) (2025)

    work page 2025

  39. [39]

    In: Computing Research Repository (CoRR) (2025)

    Tschannen, M., Gritsenko, A., Wang, X., Naeem, M.F., Alabdulmohsin, I., Parthasarathy, N., Evans, T., Beyer, L., Xia, Y., Mustafa, B., Hénaff, O., Harm- sen, J., Steiner, A., Zhai, X.: Siglip 2: Multilingual vision-language encoders with improved semantic understanding, localization, and dense features. In: Computing Research Repository (CoRR) (2025)

    work page 2025

  40. [40]

    In: Computing Research Repository (CoRR) (2025)

    Wang, W., Gao, Z., Gu, L., Pu, H., Cui, L., Wei, X., Liu, Z., Jing, L., Ye, S., Shao, J., Wang, Z., Chen, Z., Zhang, H., Yang, G., Wang, H., Wei, Q., Yin, J., Li, W., Cui, E., Chen, G., Ding, Z., Tian, C., Wu, Z., Xie, J., Li, Z., Yang, B., Duan, Y., Wang, X., Hou, Z., Hao, H., Zhang, T., Li, S., Zhao, X., Duan, H., Deng, N., Fu, B., He, Y., Wang, Y., He,...

    work page 2025

  41. [41]

    Wang, W., Sun, Y., Tan, Z., Yang, Y.: Image copy detection for diffusion models. In: Adv. Neural Inform. Process. Syst. (NeurIPS) (2024)

    work page 2024

  42. [42]

    In: Computing Research Repository (CoRR) (2025)

    Wu, C., Li, J., Zhou, J., Lin, J., Gao, K., Yan, K., ming Yin, S., Bai, S., Xu, X., Chen, Y., Chen, Y., Tang, Z., Zhang, Z., Wang, Z., Yang, A., Yu, B., Cheng, C., Liu, D., Li, D., Zhang, H., Meng, H., Wei, H., Ni, J., Chen, K., Cao, K., Peng, L., Abbreviated paper title 19 Qu, L., Wu, M., Wang, P., Yu, S., Wen, T., Feng, W., Xu, X., Wang, Y., Zhang, Y., ...

    work page 2025

  43. [43]

    In: Computing Research Repository (CoRR) (2025)

    Yang, F., Huang, Y., Zhu, J., Shi, L., Pu, G., Dong, J.S., Wang, K.: Seeing it before it happens: In-generation NSFW detection for diffusion-based text-to-image models. In: Computing Research Repository (CoRR) (2025)

    work page 2025

  44. [44]

    Yang, Y., Gao, R., Yang, X., Zhong, J., Xu, Q.: Guardt2i: Defending text-to-image models from adversarial prompts. In: Adv. Neural Inform. Process. Syst. (NeurIPS) (2024)

    work page 2024

  45. [45]

    Yoon, J., Yu, S., Patil, V., Yao, H., Bansal, M.: Safree: Training-free and adaptive guard for safe text-to-image and video generation. In: Int. Conf. Learn. Represent. (ICLR) (2025)

    work page 2025

  46. [46]

    Zhai, X., Mustafa, B., Kolesnikov, A., Beyer, L.: Sigmoid loss for language image pre-training. In: Int. Conf. Comput. Vis. (ICCV) (2023)

    work page 2023

  47. [47]

    In: IEEE Conf

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: IEEE Conf. Comput. Vis. Pattern Recog. (CVPR) (2018) A Additional Results A.1 Application to noise-based generative models Noise-based Models.In accordance with Eq. 1 in the main draft, noise-based models employ a linear ...

    work page arXiv 2018