pith. sign in

arxiv: 2606.18906 · v1 · pith:BUR5GVEJnew · submitted 2026-06-17 · 💻 cs.CV

BindEdit: Taming Attention Leakage for Precise Multi-Object Image Editing

Chaewon Park , Soyoon Lee , Naeun Lee , Minjung Shin , Seogkyu Jeon , Kibeom Hong This is my paper

Pith reviewed 2026-06-26 21:14 UTC · model grok-4.3

classification 💻 cs.CV
keywords image editingdiffusion modelsattention leakagemulti-object editingcross-attentionself-attention
0
0 comments X

The pith

BindEdit suppresses Edit-Token Leakage and Source Dominance Leakage to enable precise multi-object image edits in one diffusion trajectory.

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

The paper sets out to show that failures like semantic blending and incomplete edits in multi-object image editing stem from attention signals becoming entangled across regions and tokens during denoising. It introduces BindEdit to enforce binding of target token groups to spatial regions, re-balance cross-attention to favor target tokens, and apply a region fidelity term for coherent concept expression. A sympathetic reader would care because the approach promises reliable edits of several objects at once without running multiple separate diffusion processes. Experiments on a new benchmark with varying object counts support that the method maintains performance where prior techniques degrade.

Core claim

The authors identify Edit-Token Leakage, caused by ambiguous token-region alignment, and Source Dominance Leakage, where unchanged source tokens overwhelm target attention. BindEdit counters these by jointly regularizing cross- and self-attention to bind each target token group to its region while preserving instance separation, applying cross-attention re-balancing to amplify target influence inside editable regions, and adding a region fidelity term that ensures each target concept appears coherently across the full editing mask. This produces consistent gains over existing methods on both single- and multi-object tasks when run inside a single diffusion trajectory.

What carries the argument

BindEdit, the set of attention-level constraints (joint cross- and self-attention regularization, cross-attention re-balancing, and region fidelity term) that bind target tokens to spatial regions and reduce source dominance inside one diffusion trajectory.

If this is right

  • Precise edits become possible for scenes containing multiple objects without semantic blending or duplication.
  • Performance stays robust when the number of objects or categories increases.
  • All operations remain inside one diffusion trajectory, avoiding the cost of repeated denoising passes.
  • A dedicated multi-object benchmark now exists for systematic comparison of editing methods.

Where Pith is reading between the lines

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

  • If attention leakage is the dominant failure mode, similar binding and re-balancing steps could be tested on related generative tasks that rely on cross-attention.
  • The region fidelity term might be adapted to enforce consistency across time steps in video editing pipelines.
  • The benchmark's diversity in object counts offers a concrete way to quantify how leakage scales with scene complexity.

Load-bearing premise

The failures seen in complex multi-object editing are caused primarily by the two identified forms of attention leakage rather than other factors such as model capacity.

What would settle it

Measure whether applying the three attention constraints alone closes the performance gap on the multi-object benchmark while leaving the underlying diffusion model and trajectory count unchanged.

Figures

Figures reproduced from arXiv: 2606.18906 by Chaewon Park, Kibeom Hong, Minjung Shin, Naeun Lee, Seogkyu Jeon, Soyoon Lee.

Figure 1
Figure 1. Figure 1: (a) Edit-Token Leakage: when editing two birds into a teddy bear and a cat, the cross-attention maps of target tokens both activate across unintended regions without our binding guid￾ance, causing one object to go missing. BindEdit precisely binds each target token group to its designated mask region. (b) Source Dominance Leakage: when editing three of four dog paintings into two birds and a rabbit, the so… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed BindEdit. BindEdit guides the denoising process through three attention-level losses within a single diffusion trajectory. First, Attention Binding Guidance (Lbind) jointly constrains cross- and self-attention to enforce semantic-level binding and instance-level separation. Second, Source Suppression Guidance (Lsupp) re-balances cross-attention within editable regions by strength￾e… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparisons. We present qualitative results on three benchmarks: LoMOE-Bench, OIR-Bench, and our Extended Multi-Object Benchmark. For each example, the edit masks indicate the regions to be edited, where masks sharing the same color corre￾spond to the same target token group. The target prompt is displayed below each example, with target tokens color-coded to match their corresponding edit mask… view at source ↗
Figure 4
Figure 4. Figure 4: Ablation studies. (Left) Qualitative effect of progres￾sively adding each proposed guidance(Lbind,Lsupp, and Lfidelity) from left to right. Each loss incrementally resolves object blend￾ing, suppresses source token dominance, and consolidates scat￾tered attention into coherent objects. (Right) Robustness to sup￾pression scale γ Both CLIP-object and CLIP-image remain stable across γ ∈ {6, 8, 10, 12, 14} [P… view at source ↗
read the original abstract

Real image editing enables precise manipulation of visual content, yet existing methods often fail in complex multi-object scenarios, causing semantic blending, object duplication, or incomplete edits. We attribute these failures to attention leakage, where signals across spatial regions and text tokens become entangled during the denoising process. Specifically, we identify two distinct forms of leakage: Edit-Token Leakage, where ambiguous token-region alignment leads to object blending, and Source Dominance Leakage, where tokens of unchanged source objects overwhelm the attention intended for target entities. To resolve these leakages, we propose \textbf{BindEdit}, which enforces attention-level constraints within a single diffusion trajectory. To suppress Edit-Token Leakage, BindEdit jointly regularizes cross- and self-attention so that each target token group is bound to its corresponding spatial region while maintaining instance-level separation. To suppress Source Dominance Leakage, a cross-attention re-balancing mechanism amplifies target token influence and attenuates residual source semantics within editable regions. Moreover, a region fidelity term ensures that each target concept is expressed coherently across the entire editing mask. Additionally, we propose a comprehensive multi-object benchmark encompassing diverse object counts and categories. Extensive experiments demonstrate that BindEdit consistently outperforms existing methods within a single diffusion trajectory, maintaining robust performance across both single- and multi-object editing scenarios.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper identifies two forms of attention leakage (Edit-Token Leakage and Source Dominance Leakage) as the cause of failures in multi-object diffusion-based image editing, and proposes BindEdit to suppress them via joint cross/self-attention regularization, cross-attention re-balancing, and a region fidelity term, all within a single denoising trajectory. It introduces a new multi-object editing benchmark and claims consistent outperformance over prior methods on both single- and multi-object tasks.

Significance. If the causal attribution and performance gains hold after proper isolation, the work would be a useful incremental advance in attention-constrained editing; the new benchmark is a clear positive contribution that enables future standardized comparisons.

major comments (2)
  1. [§4] §4 (Experiments) and abstract: the central claim that BindEdit succeeds specifically by taming the two named leakage forms rests on unverified assertions; no quantitative leakage proxies (token-to-region attention concentration or source-token residual mass inside edit masks) are reported on baselines versus BindEdit, and no component ablations remove one regularization term at a time to show both leakage reduction and performance drop.
  2. [§3.2] §3.2 (cross-attention re-balancing) and §3.3 (region fidelity): these mechanisms are presented as directly addressing Source Dominance Leakage and coherent concept expression, yet the manuscript supplies no direct measurement of residual source semantics inside editable regions before/after re-balancing, leaving the load-bearing status of each term for the multi-object robustness claim untested.
minor comments (2)
  1. [§3.1] Notation for the joint regularization loss is introduced without an explicit equation number or pseudocode listing the combined objective, making it difficult to reproduce the exact weighting between cross- and self-attention terms.
  2. [§4.1] The benchmark description lacks a table enumerating object counts, categories, and edit types across the proposed dataset splits.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the referee's constructive feedback. We address each major comment below and will revise the manuscript to include the requested quantitative validations.

read point-by-point responses

  1. Referee: [§4] §4 (Experiments) and abstract: the central claim that BindEdit succeeds specifically by taming the two named leakage forms rests on unverified assertions; no quantitative leakage proxies (token-to-region attention concentration or source-token residual mass inside edit masks) are reported on baselines versus BindEdit, and no component ablations remove one regularization term at a time to show both leakage reduction and performance drop.

    Authors: We agree that direct quantitative proxies and isolated component ablations would strengthen the causal attribution. In the revised manuscript we will report token-to-region attention concentration and source-token residual mass inside edit masks for baselines versus BindEdit, together with ablations that remove each regularization term individually and measure the resulting changes in both leakage proxies and editing metrics. revision: yes

  2. Referee: [§3.2] §3.2 (cross-attention re-balancing) and §3.3 (region fidelity): these mechanisms are presented as directly addressing Source Dominance Leakage and coherent concept expression, yet the manuscript supplies no direct measurement of residual source semantics inside editable regions before/after re-balancing, leaving the load-bearing status of each term for the multi-object robustness claim untested.

    Authors: We acknowledge the lack of explicit before/after measurements of residual source semantics. The revision will add quantitative metrics and visualizations that track source-token attention mass within editable regions before and after the re-balancing and region-fidelity terms, clarifying the contribution of each component to multi-object robustness. revision: yes

Circularity Check

0 steps flagged

No circularity; claims rest on empirical method description without self-referential reduction

full rationale

The provided abstract and description contain no equations, fitted parameters, predictions derived from subsets of data, or self-citations. The central claims attribute failures to two leakage types and propose regularization terms, re-balancing, and a fidelity term, but these are presented as novel constraints rather than reductions to prior self-defined quantities or inputs. No derivation chain exists that reduces by construction to the inputs, and the performance claims are framed as experimental results on a new benchmark. This matches the default expectation of no significant circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The approach implicitly assumes that attention maps are the dominant controllable factor in the diffusion trajectory.

pith-pipeline@v0.9.1-grok · 5784 in / 1143 out tokens · 17402 ms · 2026-06-26T21:14:27.926517+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

44 extracted references · 12 canonical work pages · 5 internal anchors

  1. [1]

    Stable flow: Vital layers for training-free image editing

    Omri Avrahami, Or Patashnik, Ohad Fried, Egor Nemchi- nov, Kfir Aberman, Dani Lischinski, and Daniel Cohen- Or. Stable flow: Vital layers for training-free image editing. 2025 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 7877–7888, 2024. 1

    2025

  2. [2]

    Ledits++: Limitless image editing using text-to-image models.2024 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 8861– 8870, 2023

    Manuel Brack, Felix Friedrich, Katharina Kornmeier, Linoy Tsaban, Patrick Schramowski, Kristian Kersting, and Apolin’ario Passos. Ledits++: Limitless image editing using text-to-image models.2024 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 8861– 8870, 2023. 3

    2024

  3. [3]

    Masactrl: Tuning-free mutual self- attention control for consistent image synthesis and editing

    Ming Cao, Xintao Wang, Zhongang Qi, Ying Shan, Xiaohu Qie, and Yinqiang Zheng. Masactrl: Tuning-free mutual self- attention control for consistent image synthesis and editing. 2023 IEEE/CVF International Conference on Computer Vi- sion (ICCV), pages 22503–22513, 2023. 2

    2023

  4. [4]

    Lomoe: Localized multi-object editing via multi-diffusion

    Goirik Chakrabarty, Aditya Chandrasekar, Ramya Hebbal- aguppe, and Prathosh AP. Lomoe: Localized multi-object editing via multi-diffusion. InProceedings of the 32nd ACM International Conference on Multimedia, pages 3342–3351,

  5. [5]

    Attend-and-excite: Attention-based se- mantic guidance for text-to-image diffusion models.ACM Transactions on Graphics (TOG), 42:1 – 10, 2023

    Hila Chefer, Yuval Alaluf, Yael Vinker, Lior Wolf, and Daniel Cohen-Or. Attend-and-excite: Attention-based se- mantic guidance for text-to-image diffusion models.ACM Transactions on Graphics (TOG), 42:1 – 10, 2023. 3, 4

    2023

  6. [6]

    Training- free layout control with cross-attention guidance.2024 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 5331–5341, 2023

    Minghao Chen, Iro Laina, and Andrea Vedaldi. Training- free layout control with cross-attention guidance.2024 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 5331–5341, 2023. 2

    2024

  7. [7]

    Region-aware text-to-image generation via hard binding and soft refinement.ArXiv, abs/2411.06558, 2024

    Zhennan Chen, Yajie Li, Haofan Wang, Zhibo Chen, Zhengkai Jiang, Jun Li, Qian Wang, Jian Yang, and Ying Tai. Region-aware text-to-image generation via hard binding and soft refinement.ArXiv, abs/2411.06558, 2024. 3

    work page arXiv 2024

  8. [8]

    DiffEdit: Diffusion-based semantic image editing with mask guidance

    Guillaume Couairon, Jakob Verbeek, Holger Schwenk, and Matthieu Cord. Diffedit: Diffusion-based semantic image editing with mask guidance.ArXiv, abs/2210.11427, 2022. 2

    work page Pith review arXiv 2022

  9. [9]

    Be yourself: Bounded attention for multi-subject text-to-image generation.In European Conference on Com- puter Vision (ECCV), 2024

    Omer Dahary, Or Patashnik, Kfir Aberman, and Daniel Cohen-Or. Be yourself: Bounded attention for multi-subject text-to-image generation.In European Conference on Com- puter Vision (ECCV), 2024. 2, 3, 4

    2024

  10. [10]

    Layeredit: Disentangled multi-object editing via conflict- aware multi-layer learning.The 40th Annual AAAI Confer- ence on Artificial Intelligence, 2026

    Fengyi Fu, Mengqi Huang, Lei Zhang, and Zhendong Mao. Layeredit: Disentangled multi-object editing via conflict- aware multi-layer learning.The 40th Annual AAAI Confer- ence on Artificial Intelligence, 2026. 2, 3, 6, 7, 8

    2026

  11. [11]

    Woojung Han, Yeonkyung Lee, Chanyoung Kim, Kwanghyun Park, and Seong Jae Hwang. Spatial transport optimization by repositioning attention map for training-free text-to-image synthesis.2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 18401–18410, 2025. 2, 3, 4

    2025

  12. [12]

    Prompt-to-Prompt Image Editing with Cross Attention Control

    Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-Or. Prompt-to-prompt im- age editing with cross attention control.arXiv preprint arXiv:2208.01626, 2022. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  13. [13]

    Delta de- noising score

    Amir Hertz, Kfir Aberman, and Daniel Cohen-Or. Delta de- noising score. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 2328–2337, 2023. 1, 2

    2023

  14. [14]

    Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020. 1, 2

    2020

  15. [15]

    Paralleledits: Efficient multi- aspect text-driven image editing with attention grouping.Ad- vances in Neural Information Processing Systems 37, 2024

    Mingzhen Huang, Jialing Cai, Shan Jia, Vishnu Suresh Lokhande, and Siwei Lyu. Paralleledits: Efficient multi- aspect text-driven image editing with attention grouping.Ad- vances in Neural Information Processing Systems 37, 2024. 2, 3, 6, 7, 8

    2024

  16. [16]

    An edit friendly ddpm noise space: Inversion and manipulations

    Inbar Huberman-Spiegelglas, Vladimir Kulikov, and Tomer Michaeli. An edit friendly ddpm noise space: Inversion and manipulations. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12469– 12478, 2024. 1, 2, 3, 6, 7, 8

    2024

  17. [17]

    Bounded editing: Multi-object image manipulation with region- specific control.Proceedings of the 40th ACM/SIGAPP Sym- posium on Applied Computing, 2025

    Min Gyu Kang, Keon Kim, and Yong Suk Choi. Bounded editing: Multi-object image manipulation with region- specific control.Proceedings of the 40th ACM/SIGAPP Sym- posium on Applied Computing, 2025. 3

    2025

  18. [18]

    Flexiedit: Frequency-aware latent refinement for enhanced non-rigid editing.In European Conference on Computer Vision (ECCV), 2024

    Gwanhyeong Koo, Sunjae Yoon, Jiajing Hong, and Chang- dong Yoo. Flexiedit: Frequency-aware latent refinement for enhanced non-rigid editing.In European Conference on Computer Vision (ECCV), 2024. 1, 2

    2024

  19. [19]

    Flowedit: Inversion-free text-based editing using pre-trained flow models.ArXiv, abs/2412.08629, 2024

    Vladimir Kulikov, Matan Kleiner, Inbar Huberman- Spiegelglas, and Tomer Michaeli. Flowedit: Inversion-free text-based editing using pre-trained flow models.ArXiv, abs/2412.08629, 2024. 1

    work page arXiv 2024

  20. [20]

    Black Forest Labs, Stephen Batifol, A. Blattmann, Frederic Boesel, Saksham Consul, Cyril Diagne, Tim Dockhorn, Jack English, Zion English, Patrick Esser, Sumith Kulal, Kyle Lacey, Yam Levi, Cheng Li, Dominik Lorenz, Jonas Muller, Dustin Podell, Robin Rombach, Harry Saini, Axel Sauer, and Luke Smith. Flux.1 kontext: Flow matching for in- context image gene...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  21. [21]

    Divide & bind your attention for improved generative seman- tic nursing.ArXiv, abs/2307.10864, 2023

    Yumeng Li, Margret Keuper, Dan Zhang, and Anna Khoreva. Divide & bind your attention for improved generative seman- tic nursing.ArXiv, abs/2307.10864, 2023. 2

    work page arXiv 2023

  22. [22]

    Moedit: On learning quantity perception for multi- object image editing.2025 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 2683– 2693, 2025

    Yanfeng Li, Ka-Hou Chan, Yue Sun, Chan Quan Loi Lam, Tong Tong, Zitong Yu, Keren Fu, Xiaohong Liu, and Tao Tan. Moedit: On learning quantity perception for multi- object image editing.2025 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 2683– 2693, 2025. 3

    2025

  23. [23]

    Bingyan Liu, Chengyu Wang, Tingfeng Cao, Kui Jia, and Jun Huang. Towards understanding cross and self-attention in stable diffusion for text-guided image editing.2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7817–7826, 2024. 1, 2

    2024

  24. [24]

    Sdedit: Guided image synthesis and editing with stochastic differential equa- tions

    Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jia- jun Wu, Jun-Yan Zhu, and Stefano Ermon. Sdedit: Guided image synthesis and editing with stochastic differential equa- tions. InInternational Conference on Learning Representa- tions, 2021. 2

    2021

  25. [25]

    Null-text inversion for editing real im- ages using guided diffusion models

    Ron Mokady, Amir Hertz, Kfir Aberman, Yael Pritch, and Daniel Cohen-Or. Null-text inversion for editing real im- ages using guided diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6038–6047, 2023. 1, 2

    2023

  26. [26]

    Contrastive denoising score for text-guided latent diffusion image editing.2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9192–9201, 2023

    Hyelin Nam, Gihyun Kwon, Geon Yeong Park, and Jong Chul Ye. Contrastive denoising score for text-guided latent diffusion image editing.2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9192–9201, 2023. 1

    2024

  27. [27]

    Cross-attention head position patterns can align with human visual concepts in text-to-image gener- ative models

    Jungwon Park, Jungmin Ko, Dongnam Byun, Jangwon Suh, and Wonjong Rhee. Cross-attention head position patterns can align with human visual concepts in text-to-image gener- ative models. InInternational Conference on Learning Rep- resentations, 2025. 2

    2025

  28. [28]

    Peebles and Saining Xie

    William S. Peebles and Saining Xie. Scalable diffusion mod- els with transformers.2023 IEEE/CVF International Confer- ence on Computer Vision (ICCV), pages 4172–4182, 2022. 1, 2

    2023

  29. [29]

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

    Dustin Podell, Zion English, Kyle Lacey, A. Blattmann, Tim Dockhorn, Jonas Muller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis.ArXiv, abs/2307.01952, 2023. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  30. [30]

    Self-cross diffu- sion guidance for text-to-image synthesis of similar subjects

    Weimin Qiu, Jieke Wang, and Meng Tang. Self-cross diffu- sion guidance for text-to-image synthesis of similar subjects. 2025 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 23528–23538, 2024. 3

    2025

  31. [31]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. InInternational Conference on Machine Learning, 2021. 6

    2021

  32. [32]

    Fds: Frequency-aware denois- ing score for text-guided latent diffusion image editing

    Yufan Ren, Zicong Jiang, Tong Zhang, Søren Forchham- mer, and Sabine S ¨usstrunk. Fds: Frequency-aware denois- ing score for text-guided latent diffusion image editing. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 2651–2660, 2025. 1, 2

    2025

  33. [33]

    Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer

    Robin Rombach, A. Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models.2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10674–10685, 2021. 1, 2

    2022

  34. [34]

    Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L. Denton, Seyed Kamyar Seyed Ghasemipour, Burcu Karagol Ayan, Seyedeh Sara Mah- davi, Raphael Gontijo Lopes, Tim Salimans, Jonathan Ho, David J. Fleet, and Mohammad Norouzi. Photorealistic text- to-image diffusion models with deep language understand- ing.ArXiv, abs/2205.11487, 2022. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  35. [35]

    Exploring multimodal diffusion trans- formers for enhanced prompt-based image editing.ArXiv, abs/2508.07519, 2025

    Joonghyuk Shin, Alchan Hwang, Yujin Kim, Daneul Kim, and Jaesik Park. Exploring multimodal diffusion trans- formers for enhanced prompt-based image editing.ArXiv, abs/2508.07519, 2025. 1

    work page arXiv 2025

  36. [36]

    Denoising Diffusion Implicit Models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. De- noising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  37. [37]

    Lin, and Ferhan Ture

    Raphael Tang, Akshat Pandey, Zhiying Jiang, Gefei Yang, Karun Kumar, Jimmy J. Lin, and Ferhan Ture. What the daam: Interpreting stable diffusion using cross attention. In Annual Meeting of the Association for Computational Lin- guistics, 2022. 4

    2022

  38. [38]

    Diffuse, attend, and segment: Unsupervised zero-shot segmentation using stable diffusion

    Junjiao Tian, Lavisha Aggarwal, Andrea Colaco, Zsolt Kira, and Mar Gonz ´alez-Franco. Diffuse, attend, and segment: Unsupervised zero-shot segmentation using stable diffusion. 2024 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 3554–3563, 2023. 4

    2024

  39. [39]

    Plug-and-play diffusion features for text-driven image-to-image translation

    Narek Tumanyan, Michal Geyer, Shai Bagon, and Tali Dekel. Plug-and-play diffusion features for text-driven image-to-image translation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1921–1930, 2023. 1, 2

    1921

  40. [40]

    Wei Wu, Qingnan Fan, Shuai Qin, Hong Gu, Ruoyu Zhao, and Antoni B. Chan. Freediff: Progressive frequency trun- cation for image editing with diffusion models.In European Conference on Computer Vision (ECCV), 2024. 1, 2

    2024

  41. [41]

    Object-aware inver- sion and reassembly for image editing.arXiv preprint arXiv:2310.12149, 2023

    Zhen Yang, Ganggui Ding, Wen Wang, Hao Chen, Bo- han Zhuang, and Chunhua Shen. Object-aware inver- sion and reassembly for image editing.arXiv preprint arXiv:2310.12149, 2023. 2, 3, 6, 7, 8

    work page arXiv 2023

  42. [42]

    Splitflow: Flow decomposition for inversion-free text-to- image editing.In Conference on Neural Information Pro- cessing Systems (NeurIPS), 2025

    Sunghoon Yoon, Minghan Li, Gaspard Beaudouin, Cong- cong Wen, Muhammad Rafay Azhar, and Mengyu Wang. Splitflow: Flow decomposition for inversion-free text-to- image editing.In Conference on Neural Information Pro- cessing Systems (NeurIPS), 2025. 3, 6, 7, 8

    2025

  43. [43]

    Efros, Eli Shecht- man, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric.2018 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 586–595, 2018. 6

    2018

  44. [44]

    Mde- edit: Masked dual-editing for multi-object image editing via diffusion models.ArXiv, abs/2505.05101, 2025

    Hongyang Zhu, Haipeng Liu, Bo Fu, and Yang Wang. Mde- edit: Masked dual-editing for multi-object image editing via diffusion models.ArXiv, abs/2505.05101, 2025. 3

    work page arXiv 2025