pith. sign in

arxiv: 2605.29390 · v1 · pith:ZQF75BDEnew · submitted 2026-05-28 · 💻 cs.CV

Orthogonal Negative Guidance in Attention Feature Space for Text-to-Image Generation

Jungmin Ko , Jungwon Park , Jimyeong Kim , Changin Choi , Wonseok Lee , Wonjong Rhee This is my paper

Pith reviewed 2026-06-29 08:27 UTC · model grok-4.3

classification 💻 cs.CV
keywords text-to-image generationnegative guidanceattention featuresconcept suppressionMM-DiTorthogonal projectionFLUX modeldiffusion transformers
0
0 comments X

The pith

Orthogonal negative guidance in attention feature space suppresses unwanted concepts in text-to-image generation while preserving desired semantics.

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

The paper proposes a training-free method for enforcing the absence of specified objects or attributes in images generated by text-to-image models. It works by orthogonalizing the attention features from a negative prompt relative to those from the positive prompt in the output space of MM-DiT transformers, then subtracting only the orthogonal component. This approach aims to improve upon existing methods that either fail to remove the target concept or degrade image quality. Experiments on FLUX models demonstrate better trade-offs between concept suppression, prompt alignment, and quality, with human evaluations showing an 18.78% improvement over the second-best baseline. The method also supports suppressing multiple concepts and allows adjustable levels of suppression.

Core claim

By orthogonalizing negative-prompt attention features with respect to positive-prompt features in the attention output space and subtracting only the orthogonal component, the method suppresses unwanted concepts while preserving desired semantics in MM-DiT-based text-to-image transformers.

What carries the argument

Orthogonal Negative Guidance, which orthogonalizes negative-prompt attention features to positive-prompt features and subtracts only the orthogonal component from the positive features.

If this is right

  • It achieves favorable trade-offs between concept suppression, prompt alignment, and image quality on FLUX-dev and FLUX-schnell.
  • Human evaluation shows it outperforms the second-best baseline by 18.78%.
  • It supports multi-concept suppression.
  • It allows adjustable concept suppression.

Where Pith is reading between the lines

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

  • This selective subtraction could be tested in other transformer-based generative architectures to see if the orthogonality principle generalizes.
  • Adjustable suppression levels suggest applications in fine-tuning user control over generated content attributes.
  • The method implies that feature space projections can isolate semantic directions more precisely than simple prompt negation.

Load-bearing premise

Subtracting only the orthogonal component of negative-prompt attention features from positive-prompt features in the output space of MM-DiT transformers will suppress the target concept without collateral damage to desired semantics or image quality.

What would settle it

Generating images with the method and checking if the unwanted concept still appears at similar rates as standard negative prompting, or if image quality metrics drop significantly compared to baselines.

Figures

Figures reproduced from arXiv: 2605.29390 by Changin Choi, Jimyeong Kim, Jungmin Ko, Jungwon Park, Wonjong Rhee, Wonseok Lee.

Figure 1
Figure 1. Figure 1: Images generated without negative guidance (left image of each pair) and with our Orthogonal Negative Guidance (right image of each pair) on FLUX-dev. Black and red text indicate the positive and negative prompts, respectively. Our method effectively suppresses unwanted concepts across diverse scenarios and supports multi￾concept suppression (bottom left) and adjustable concept suppression (bottom right). … view at source ↗
Figure 2
Figure 2. Figure 2: Orthogonal Negative Guidance in attention feature space. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Diverse Concept Suppression Benchmark (DCS-Bench) [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison with negative guidance baselines on FLUX [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison with generate-and-edit baselines on FLUX [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Quantitative comparison on FLUX-dev (left two plots) and FLUX [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: shows that removing feature sharing leads to spatially misaligned guidance signals, which appear as handbag-shaped and lamppost￾like distortions in the first and second rows, re￾spectively. These results indicate that sharing image-side features between branches is impor￾tant for constructing spatially aligned negative guidance signals. Effect of Orthogonalization. As shown in Eq. (9), our method subtracts… view at source ↗
Figure 8
Figure 8. Figure 8: Effect of Orthogonalization. (a) Qualitative comparison. (b) Quantitative comparison [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Examples of adjustable concept suppression. [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Examples of multi-concept suppression. Our method suppresses multiple concepts simultaneously by specifying all target concepts in a single negative prompt. The leftmost images correspond to generation without negative guidance. 7 Limitations and Conclusion “A professional photography studio” – “softbox lights” FLUX-dev Ours [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Failure case. Our method has several limitations. First, because it suppresses concepts during the generation pro￾cess, it is not well suited for sequential concept suppression, which is better suited to generate-and￾edit pipelines. Second, our method occasionally fails to suppress certain concepts, with relatively lower suppression performance in the Place/Scene and Event/Action categories, as shown in … view at source ↗
Figure 12
Figure 12. Figure 12: Overview of the DCS-Bench dataset construction pipeline. [PITH_FULL_IMAGE:figures/full_fig_p020_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Representative examples from the VLM-based evaluation across the [PITH_FULL_IMAGE:figures/full_fig_p024_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Prompt template for the Negative Concept Suppression metric. [PITH_FULL_IMAGE:figures/full_fig_p025_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Prompt template for the Prompt Alignment metric. [PITH_FULL_IMAGE:figures/full_fig_p026_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Prompt template for the Image Quality metric. [PITH_FULL_IMAGE:figures/full_fig_p027_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Screenshot of the survey interface used in the human preference [PITH_FULL_IMAGE:figures/full_fig_p030_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Additional qualitative comparison with negative guidance methods [PITH_FULL_IMAGE:figures/full_fig_p033_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Additional qualitative comparison with generate-and-edit baselines [PITH_FULL_IMAGE:figures/full_fig_p034_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Additional qualitative comparison with negative guidance methods [PITH_FULL_IMAGE:figures/full_fig_p035_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Additional qualitative comparison with generate-and-edit baselines [PITH_FULL_IMAGE:figures/full_fig_p036_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Additional qualitative comparison with negative guidance methods [PITH_FULL_IMAGE:figures/full_fig_p037_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Additional qualitative comparison with generate-and-edit baselines [PITH_FULL_IMAGE:figures/full_fig_p038_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Additional qualitative results on suppression scenarios not included [PITH_FULL_IMAGE:figures/full_fig_p039_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Per-category quantitative comparison on FLUX-dev. [PITH_FULL_IMAGE:figures/full_fig_p040_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Per-category quantitative comparison on FLUX-schnell. [PITH_FULL_IMAGE:figures/full_fig_p041_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: Qualitative Comparison with prompt-level negation on FLUX-dev. [PITH_FULL_IMAGE:figures/full_fig_p043_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: Ablation on attention features for negative guidance. [PITH_FULL_IMAGE:figures/full_fig_p043_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: Additional examples of adjustable concept suppression. [PITH_FULL_IMAGE:figures/full_fig_p044_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: Additional examples of multi-concept suppression. [PITH_FULL_IMAGE:figures/full_fig_p045_30.png] view at source ↗
read the original abstract

Text-to-image (T2I) models have become increasingly capable of generating high-quality images. Yet, enforcing the explicit absence of a specified object or attribute remains a fundamentally challenging problem. Existing approaches, including prompt negation, post-hoc editing, and negative guidance, remain insufficient for explicit concept suppression, often failing to remove the target concept or degrading overall image quality. To this end, we propose Orthogonal Negative Guidance in attention feature space, a training-free method that operates in the attention output space of MM-DiT-based T2I transformers. Our method orthogonalizes negative-prompt attention features with respect to positive-prompt features and subtracts only the orthogonal component, suppressing unwanted concepts while preserving desired semantics. Experiments on FLUX-dev and FLUX-schnell show that our method achieves favorable trade-offs between concept suppression, prompt alignment, and image quality. In human evaluation, our method outperforms the second-best baseline by 18.78%. We further show that our method supports multi-concept suppression and adjustable concept suppression.

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

1 major / 1 minor

Summary. The paper proposes Orthogonal Negative Guidance, a training-free method for explicit concept suppression in text-to-image generation. It operates directly in the attention output space of MM-DiT transformers by orthogonalizing negative-prompt attention features with respect to positive-prompt features and subtracting only the orthogonal component. This is claimed to suppress unwanted concepts while preserving desired semantics. Experiments on FLUX-dev and FLUX-schnell are reported to achieve favorable trade-offs among suppression, prompt alignment, and image quality, with an 18.78% win rate in human evaluation over the second-best baseline; the method is further shown to support multi-concept and adjustable suppression.

Significance. If the results hold, the work supplies a parameter-free geometric construction for negative guidance that requires no additional training or fitted parameters. The direct orthogonality operation in attention feature space, combined with the reported human-evaluation gain and multi-concept support, would represent a practical advance for controllable T2I generation.

major comments (1)
  1. [§4] §4 (Experiments): The central empirical claim—an 18.78% human-evaluation win and favorable trade-offs on FLUX-dev/schnell—lacks accompanying quantitative metrics, baseline definitions, number of evaluators, statistical tests, or implementation details in the reported results. This information is load-bearing for verifying the suppression-without-collateral-damage claim.
minor comments (1)
  1. [§3] A diagram or explicit projection formula in the method section would clarify how orthogonality is computed in the high-dimensional attention output space.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The point raised about experimental details is valid, and we will incorporate the requested information in the revision to improve clarity and verifiability.

read point-by-point responses

  1. Referee: [§4] §4 (Experiments): The central empirical claim—an 18.78% human-evaluation win and favorable trade-offs on FLUX-dev/schnell—lacks accompanying quantitative metrics, baseline definitions, number of evaluators, statistical tests, or implementation details in the reported results. This information is load-bearing for verifying the suppression-without-collateral-damage claim.

    Authors: We agree that the human evaluation section requires additional supporting details to allow independent verification. In the revised manuscript, we will expand §4 to include: (1) quantitative metrics such as CLIP-based prompt alignment scores, concept suppression rates via classifier probes, and FID for image quality; (2) explicit definitions and hyperparameter settings for all baselines (e.g., negative prompting, classifier-free guidance variants); (3) the number of human evaluators, their recruitment criteria, and the exact evaluation protocol (pairwise comparisons with randomized presentation); (4) statistical tests including p-values and confidence intervals for the 18.78% win rate; and (5) full implementation details such as the precise negative/positive prompt templates, diffusion steps, guidance scales, and hardware used. These additions will directly address the concern about verifying the claimed trade-offs. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper presents a training-free geometric operation—orthogonalizing negative-prompt attention features w.r.t. positive-prompt features in MM-DiT output space and subtracting only the orthogonal component—without any derivation, fitted parameters, or self-referential definitions that reduce the claimed result to its inputs by construction. The central claim is an explicit algorithmic step whose empirical behavior (concept suppression vs. quality trade-offs on FLUX-dev/schnell, human eval gains) is tested directly; no equations or self-citations are invoked as load-bearing premises that would create circularity. This is the common case of a self-contained empirical method.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no free parameters, axioms, or invented entities are stated or derivable from the provided text.

pith-pipeline@v0.9.1-grok · 5720 in / 1123 out tokens · 33947 ms · 2026-06-29T08:27:43.833336+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

57 extracted references · 19 canonical work pages · 13 internal anchors

  1. [1]

    AI, S.: Introducing stable diffusion 3.5 (2024),https://stability.ai/news/ introducing-stable-diffusion-3-5/

    2024

  2. [2]

    Building Normalizing Flows with Stochastic Interpolants

    Albergo, M.S., Vanden-Eijnden, E.: Building normalizing flows with stochastic interpolants. arXiv preprint arXiv:2209.15571 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  3. [3]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Alhamoud, K., Alshammari, S., Tian, Y., Li, G., Torr, P.H., Kim, Y., Ghassemi, M.: Vision-language models do not understand negation. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 29612–29622 (2025)

    2025

  4. [4]

    In: SIGGRAPH Asia 2023 Conference Papers

    Avrahami, O., Aberman, K., Fried, O., Cohen-Or, D., Lischinski, D.: Break-a- scene: Extracting multiple concepts from a single image. In: SIGGRAPH Asia 2023 Conference Papers. pp. 1–12 (2023)

    2023

  5. [5]

    Qwen3-VL Technical Report

    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 internal anchor Pith review Pith/arXiv arXiv 2025

  6. [6]

    Bar-Tal, O., Yariv, L., Lipman, Y., Dekel, T.: Multidiffusion: Fusing diffusion paths for controlled image generation (2023)

    2023

  7. [7]

    Advances in Neural Information Processing Systems36, 25365–25389 (2023)

    Brack, M., Friedrich, F., Hintersdorf, D., Struppek, L., Schramowski, P., Kersting, K.: Sega: Instructing text-to-image models using semantic guidance. Advances in Neural Information Processing Systems36, 25365–25389 (2023)

    2023

  8. [8]

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

    Brooks, T., Holynski, A., Efros, A.A.: Instructpix2pix: Learning to follow image editing instructions. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 18392–18402 (2023)

    2023

  9. [9]

    arXiv preprint arXiv:2510.14376 (2025)

    Byun, D., Park, J., Ko, J., Choi, C., Rhee, W.: Dos: Directional object sep- aration in text embeddings for multi-object image generation. arXiv preprint arXiv:2510.14376 (2025)

    work page arXiv 2025

  10. [10]

    ACM trans- actions on Graphics (TOG)42(4), 1–10 (2023)

    Chefer, H., Alaluf, Y., Vinker, Y., Wolf, L., Cohen-Or, D.: Attend-and-excite: Attention-based semantic guidance for text-to-image diffusion models. ACM trans- actions on Graphics (TOG)42(4), 1–10 (2023)

    2023

  11. [11]

    arXiv preprint arXiv:2505.21179 (2025)

    Chen, D.Y., Bandyopadhyay, H., Zou, K., Song, Y.Z.: Normalized atten- tion guidance: Universal negative guidance for diffusion model. arXiv preprint arXiv:2505.21179 (2025)

    work page arXiv 2025

  12. [12]

    In: Forty-first international conference on machine learning (2024)

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, 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

  13. [13]

    An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion

    Gal, R., Alaluf, Y., Atzmon, Y., Patashnik, O., Bermano, A.H., Chechik, G., Cohen-Or, D.: An image is worth one word: Personalizing text-to-image gener- ation using textual inversion. arXiv preprint arXiv:2208.01618 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  14. [14]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Gandikota, R., Materzynska, J., Fiotto-Kaufman, J., Bau, D.: Erasing concepts from diffusion models. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 2426–2436 (2023)

    2023

  15. [15]

    In: European Conference on Computer Vision

    Gandikota, R., Materzyńska, J., Zhou, T., Torralba, A., Bau, D.: Concept sliders: Lora adaptors for precise control in diffusion models. In: European Conference on Computer Vision. pp. 172–188. Springer (2024) 16 J. Ko et al

    2024

  16. [16]

    In: Proceedings of the IEEE/CVF winter conference on applications of computer vision

    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. pp. 5111–5120 (2024)

    2024

  17. [17]

    arXiv preprint arXiv:2508.10931 (2025)

    Guo, W., Du, S.: Vsf: Simple, efficient, and effective negative guidance in few-step image generation models by value sign flip. arXiv preprint arXiv:2508.10931 (2025)

    work page arXiv 2025

  18. [18]

    In: Proceedings of the IEEE/CVF inter- national conference on computer vision

    Han, L., Li, Y., Zhang, H., Milanfar, P., Metaxas, D., Yang, F.: Svdiff: Compact parameter space for diffusion fine-tuning. In: Proceedings of the IEEE/CVF inter- national conference on computer vision. pp. 7323–7334 (2023)

    2023

  19. [19]

    Prompt-to-Prompt Image Editing with Cross Attention Control

    Hertz, A., Mokady, R., Tenenbaum, J., Aberman, K., Pritch, Y., Cohen-Or, D.: Prompt-to-prompt image editing with cross attention control. arXiv preprint arXiv:2208.01626 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  20. [20]

    Classifier-Free Diffusion Guidance

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

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  21. [21]

    GPT-4o System Card

    Hurst, A., Lerer, A., Goucher, A.P., Perelman, A., Ramesh, A., Clark, A., Os- trow, A., Welihinda, A., Hayes, A., Radford, A., et al.: Gpt-4o system card. arXiv preprint arXiv:2410.21276 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  22. [22]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Kim, J., Park, J., Rhee, W.: Selectively informative description can reduce unde- sired embedding entanglements in text-to-image personalization. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 8312–8322 (2024)

    2024

  23. [23]

    arXiv preprint arXiv:2507.01496 (2025)

    Kim, J., Park, J., Song, Y., Kwak, N., Rhee, W.: Reflex: Text-guided editing of real images in rectified flow via mid-step feature extraction and attention adaptation. arXiv preprint arXiv:2507.01496 (2025)

    work page arXiv 2025

  24. [24]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Kulikov, V., Kleiner, M., Huberman-Spiegelglas, I., Michaeli, T.: Flowedit: Inversion-free text-based editing using pre-trained flow models. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 19721–19730 (2025)

    2025

  25. [25]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Kumari, N., Zhang, B., Wang, S.Y., Shechtman, E., Zhang, R., Zhu, J.Y.: Ablat- ing concepts in text-to-image diffusion models. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 22691–22702 (October 2023)

    2023

  26. [26]

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

    Kumari, N., Zhang, B., Zhang, R., Shechtman, E., Zhu, J.Y.: Multi-concept cus- tomization of text-to-image diffusion. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 1931–1941 (2023)

    1931

  27. [27]

    Labs, B.F.: Announcing black forest labs (2024),https://blackforestlabs.ai/ announcing-black-forest-labs/

    2024

  28. [28]

    Labs, B.F.: Flux.https://github.com/black-forest-labs/flux(2024)

    2024

  29. [29]

    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., Dock- horn, T., English, J., English, Z., Esser, P., et al.: Flux. 1 kontext: Flow match- ing 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

  30. [30]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Li, J., Hu, L., Zhang, J., Zheng, T., Zhang, H., Wang, D.: Fair text-to-image diffusion via fair mapping. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 39, pp. 26256–26264 (2025)

    2025

  31. [31]

    arXiv preprint arXiv:2402.05375 (2024)

    Li, S., van de Weijer, J., Hu, T., Khan, F.S., Hou, Q., Wang, Y., Yang, J.: Get what you want, not what you don’t: Image content suppression for text-to-image diffusion models. arXiv preprint arXiv:2402.05375 (2024)

    work page arXiv 2024

  32. [32]

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

    Li,Y.,Liu,H.,Wu,Q.,Mu,F.,Yang,J.,Gao,J.,Li,C.,Lee,Y.J.:Gligen:Open-set grounded text-to-image generation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 22511–22521 (2023) Orthogonal Negative Guidance in Attention Feature Space 17

    2023

  33. [33]

    In: European conference on computer vision

    Lin, T.Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., Zitnick, C.L.: Microsoft coco: Common objects in context. In: European conference on computer vision. pp. 740–755. Springer (2014)

    2014

  34. [34]

    Flow Matching for Generative Modeling

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

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  35. [35]

    In: European conference on computer vision

    Liu, N., Li, S., Du, Y., Torralba, A., Tenenbaum, J.B.: Compositional visual gen- eration with composable diffusion models. In: European conference on computer vision. pp. 423–439. Springer (2022)

    2022

  36. [36]

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

    Liu, X., Gong, C., Liu, Q.: Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  37. [37]

    SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations

    Meng, C., He, Y., Song, Y., Song, J., Wu, J., Zhu, J.Y., Ermon, S.: Sdedit: Guided image synthesis and editing with stochastic differential equations. arXiv preprint arXiv:2108.01073 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  38. [38]

    In: Proceedings of the AAAI conference on artificial intelligence

    Mou, C., Wang, X., Xie, L., Wu, Y., Zhang, J., Qi, Z., Shan, Y.: T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In: Proceedings of the AAAI conference on artificial intelligence. vol. 38, pp. 4296–4304 (2024)

    2024

  39. [39]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Nguyen, V., Nguyen, A., Dao, T., Nguyen, K., Pham, C., Tran, T., Tran, A.: Supercharged one-step text-to-image diffusion models with negative prompts. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 18004–18013 (2025)

    2025

  40. [40]

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

    Parihar, R., Bhat, A., Basu, A., Mallick, S., Kundu, J.N., Babu, R.V.: Balanc- ing act: Distribution-guided debiasing in diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 6668–6678 (2024)

    2024

  41. [41]

    In: The Thirteenth International Conference on Learning Representations (2024)

    Park, J., Ko, J., Byun, D., Suh, J., Rhee, W.: Cross-attention head position pat- terns can align with human visual concepts in text-to-image generative models. In: The Thirteenth International Conference on Learning Representations (2024)

    2024

  42. [42]

    Peebles,W.,Xie,S.:Scalablediffusionmodelswithtransformers.In:Proceedingsof the IEEE/CVF international conference on computer vision. pp. 4195–4205 (2023)

    2023

  43. [43]

    SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., Müller, J., Penna, J., Rombach, R.: Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  44. [44]

    arXiv preprint arXiv:2210.04610 (2022)

    Rando, J., Paleka, D., Lindner, D., Heim, L., Tramèr, F.: Red-teaming the stable diffusion safety filter. arXiv preprint arXiv:2210.04610 (2022)

    work page arXiv 2022

  45. [45]

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

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., 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

  46. [46]

    Ruiz, N., Li, Y., Jampani, V., Pritch, Y., Rubinstein, M., Aberman, K.: Dream- booth:Finetuningtext-to-imagediffusionmodelsforsubject-drivengeneration.In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition. pp. 22500–22510 (2023)

    2023

  47. [47]

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

    Schramowski, P., Brack, M., Deiseroth, B., 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

  48. [48]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Shin, J., Hwang, A., Kim, Y., Kim, D., Park, J.: Exploring multimodal diffu- sion transformers for enhanced prompt-based image editing. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 19492–19502 (2025) 18 J. Ko et al

    2025

  49. [49]

    In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition

    Tumanyan, N., Geyer, M., Bagon, S., Dekel, T.: Plug-and-play diffusion features for text-driven image-to-image translation. In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition. pp. 1921–1930 (2023)

    1921

  50. [50]

    Advances in neural information pro- cessing systems30(2017)

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need. Advances in neural information pro- cessing systems30(2017)

    2017

  51. [51]

    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., Qu, L., Wu, M., Wang, P., Yu, S., Wen, T., Feng, W., Xu, X., Wang, Y., Zhang, Y., Zhu, Y., Wu, Y., Cai, Y., L...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  52. [52]

    OmniGen2: Towards Instruction-Aligned Multimodal Generation

    Wu, C., Zheng, P., Yan, R., Xiao, S., Luo, X., Wang, Y., Li, W., Jiang, X., Liu, Y., Zhou, J., et al.: Omnigen2: Exploration to advanced multimodal generation. arXiv preprint arXiv:2506.18871 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  53. [53]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Xie,J.,Li,Y.,Huang,Y.,Liu,H.,Zhang,W.,Zheng,Y.,Shou,M.Z.:Boxdiff:Text- to-image synthesis with training-free box-constrained diffusion. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 7452–7461 (2023)

    2023

  54. [54]

    In: 2024 IEEE symposium on security and privacy (SP)

    Yang, Y., Hui, B., Yuan, H., Gong, N., Cao, Y.: Sneakyprompt: Jailbreaking text- to-image generative models. In: 2024 IEEE symposium on security and privacy (SP). pp. 897–912. IEEE (2024)

    2024

  55. [55]

    a bunch of animals that are in a field

    Zhang, L., Rao, A., Agrawala, M.: Adding conditional control to text-to-image diffusion models. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 3836–3847 (2023) Orthogonal Negative Guidance in Attention Feature Space 19 A Pseudocode for Orthogonal Negative Guidance Algorithm 1Orthogonal Negative Guidance Require:P +: Posit...

    2023

  56. [56]

    Object: "roses"

    Object Absence: The image must NOT contain the object mentioned below. Object: "roses"

  57. [57]

    A vase of flowers is partially in the dark

    Description Match: The image matches the given description. Description: "A vase of flowers is partially in the dark .. " 0 Image 1 0 lmage3 0 Images Back Next 0 lmage2 0 lmage4 D No image satisfies both requirements. Clear form Fig.17: Screenshot of the survey interface used in the human preference study.Participants selected all images that satisfied bo...