Diffusion Image Editing via Asynchronous Token Decoding

Jingchao Wang; Liangsi Lu; Minzhe Guo; Xuhang Chen; Yang Shi; Yanhui Chen; Yifeng Xie

REVIEW 4 major objections 4 minor 44 references

Diffusion Image Editing via Asynchronous Token Decoding

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read ATDEdit makes diffusion editing token-asynchronous, using a surprisal score to update only tokens that should change and reporting 27.44 dB PSNR on unedited regions.

desk verdict ATDEdit is a coherent and honestly scoped test-time editing recipe, but the target-latent coupling in Eq. (7) is off-manifold as written and the headline numbers need error bars and code. read the letter →

arxiv 2608.09322 v1 pith:XU7T7GTU submitted 2026-08-10 cs.CV

Yang Shi , Liangsi Lu , Minzhe Guo , Yifeng Xie , Yanhui Chen , Jingchao Wang , Xuhang Chen This is my paper · ORCID

classification cs.CV

keywords text-guidedimageeditingdiffusionmodelstransformersasynchronoustokendecodinginference-timebackgroundpreservationtoken-wiseconditionalsurprisalhardprojection

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

The paper proposes ATDEdit, an inference-time framework for text-guided image editing with Diffusion Transformers that treats each sampler step as a parallel but globally coupled update of a token matrix. It argues that switching the text condition for every token at once causes global drift, and that editing should be asynchronous: each token row follows its own condition-switching and update rule. The method discovers editable tokens online through Token-wise Conditional Surprisal, a same-state velocity-difference score, then constrains the remaining keep tokens by supplying source key/value memory and hard-projecting their latent rows back to the source. On PIE-Bench, ATDEdit reports the strongest background-preservation numbers in its comparison (27.44 dB PSNR, 0.055 LPIPS), without external masks, user masks, or fine-tuning; if correct, this gives a training-free way to keep local edits local in modern DiT-based editors.

What carries the argument

The load-bearing object is Token-wise Conditional Surprisal, a token-level score $s_i(t)$ that averages, over $n_{\mathrm{avg}}$ common-noise draws, the $\ell^1$ norm of the difference between predicted clean latents under target and source conditions at the same noisy latent; under rectified-flow parameterization this equals the magnitude of the same-state velocity difference. The score is smoothed by EMA, top-quantile selected in a mid-noise window, and optionally dilated to form the editable mask $E(t)$. The complementary machinery is the keep-token constraint: a partial attention memory boundary that substitutes source key/value rows at keep positions (leaving queries and attention weights target-conditioned) and a Euclidean projection that resets keep-token latent rows to their source values. Together they implement the asynchronous update, and the paper proves the projection is the exact Euclidean projection onto the affine keep-token constraint set.

What would settle it

Run ATDEdit on a global edit (e.g., 'turn day to night' where nearly every token should change) and compare CLIP-Edited against a synchronous SD3 baseline under the same step budget; if the local-editing framework freezes tokens that should change and CLIP-Edited drops substantially below the synchronous baseline, that outcome would contradict the claim that surprisal reliably identifies the editable set.

Watch

Extended reading notes

Core claim

ATDEdit reformulates diffusion editing as asynchronous token decoding: instead of a synchronous target-conditioned update applied to all tokens, the framework estimates an editable token set $E(t)$ at each timestep using Token-wise Conditional Surprisal, defined as the Monte Carlo average of the $\ell^1$ magnitude of the same-state $\hat{x}_0$ prediction difference when only the condition switches from $c_{\mathrm{src}}$ to $c_{\mathrm{tar}}$. Selected editable tokens receive target-conditioned corrections under a common-noise parallelogram coupling, while keep tokens are constrained by an attention memory boundary that replaces their keys/values with source-branch values and by a hard projection that exactly resets their latent rows to the source state. The paper is explicit about the scope of its formal results: the projection gives a token-state zero-drift guarantee in exact arithmetic, common noise reduces estimator variance under nonnegative covariance, and top-quantile mask selection optimizes only captured surprisal mass under a fixed cardinality budget. The empirical claim is that on PIE-Bench, this yields the strongest reported preservation metrics in the comparison — PSNR 27.44 dB, MSE 2.70 $\times 10^{-3}$, SSIM 0.9303, LPIPS 0.055 — while CLIP-Edited remains competitive with the same-backbone SD3 baselines.

Load-bearing premise

The load-bearing premise is that token-wise conditional surprisal, computed from same-state source/target velocity differences, reliably identifies the tokens that should change under the new prompt; if the DiT velocity field is miscalibrated or the surprisal ranking is wrong, the hard projection will freeze semantically editable regions or allow background drift, and the reported PSNR and LPIPS gains will not transfer to new tasks.

Editorial extensions

If this is right

Background preservation on PIE-Bench improves across Structure Distance, PSNR, MSE, SSIM, and LPIPS without any ground-truth or user-provided mask at inference time.
The online, time-varying mask lets the protected region adapt as denoising evolves, addressing a limitation of static mask-based editors like DiffEdit.
Because ATDEdit changes only the inference dynamics and not the backbone, it can be applied to any DiT-based rectified-flow sampler without fine-tuning.
The hard projection provides an exact latent-row zero-drift guarantee for selected keep tokens, making the preservation mechanism directly controllable in token space even though pixel-level invariance is explicitly not claimed.
The common-noise parallelogram coupling is a reusable variance-reduction device for any estimator built from paired source/target DiT evaluations.

Reading between the lines

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

A testable extension is to let the user steer the editable fraction $\rho(t)$ per region or per semantic class, which would make the surprisal-based mask a controllable editing dial rather than a fully automatic score.
The same asynchronous-token formulation could be carried to video editing, where the token grid is replaced by a spatiotemporal token set; the variance-reduction and zero-drift arguments would apply equally, but the mask discovery would need to handle motion.
Because the paper's formal guarantees are scoped to latent rows, a natural diagnostic is to compare pixel-level fidelity against the zero-drift guarantee: a large mismatch would localize how much of the background-preservation gain comes from the projection versus the KV memory boundary.
The robustness of the method hinges on whether surprisal ranking stays accurate under distribution shifts; stress-testing with prompts that are semantically similar to the source (e.g., near-synonym swaps) would show whether the editability signal degrades before the pixel metrics do.

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.

Desk Editor's Note

read the letter

ATDEdit is a genuinely new combination of known pieces: online token-wise conditional surprisal for locating edits, source key/value memory substitution, and hard projection of keep-token rows. The writing is clear and the formal claims are honestly scoped. The projection statement is exactly the Euclidean projection onto the keep-set constraint, and the top-quantile claim is just fixed-budget score maximization; the paper says both explicitly and does not oversell them. That is a real strength.

The empirical story is plausible but not fully secured. Table 1 mixes backbones; the same-backbone comparison against FlowEdit and FlowAlign is the only fair one, and ATDEdit wins preservation metrics there. But there are no error bars, no code or exact hyperparameters, and the ablation table is point estimates with a note that differences should be read descriptively. For a claimed SOTA result, that is thin. Figure 5 is a one-sample sensitivity sweep, so the window parameters are not really validated.

The stress-test on Eq. (7) is correct. The paper couples the target latent as z_tar_t = z_src_t + (x_edit_t - x_src), but same-noise rectified-flow interpolation would give z_src_t + (1-t)(x_edit_t - x_src). As written, z_tar_t is off the interpolation manifold by t(x_edit_t - x_src), and the target velocity is evaluated at those off-manifold points. The paper gives no argument or empirical check that the SD3 velocity field stays calibrated there. This does not invalidate the variance-reduction claim, since the same noise realization is still shared, but it does mean the steering update has no correctness guarantee and the method's stated mechanism is not fully established. This is a real internal inconsistency, not a matter of taste.

The self-citations to the authors' earlier editing methods are not a problem by themselves; what is more noticeable is that those methods are cited but not compared, which is easy for a reader to flag. The conclusion also honestly lists limitations: computational overhead, reliance on DiT calibration, and poor fit for global edits.

Who is this for? Anyone working on training-free DiT image editing. It deserves peer review: the core idea is coherent, the write-up is honest, and the off-manifold coupling is either fixable or at least addressable with an experiment showing it does not matter. I would not desk-reject it. With code, error bars, and a response on the Eq. (7) issue, it could be a solid contribution.

Referee Report

4 major / 4 minor

Summary. This paper proposes ATDEdit, a training-free, inference-time method for text-guided editing with diffusion transformers. The method computes a token-level "conditional surprisal" from same-state source/target velocity predictions to identify editable tokens, applies target-conditioned updates only to those tokens, substitutes source key/value memory at keep-token positions, and hard-resets keep-token latent rows to their source values. The authors derive scoped theoretical statements: common-noise coupling reduces estimator variance under nonnegative covariance, hard projection yields exact zero drift on selected keep-token rows in exact arithmetic, and top-quantile selection maximizes captured score mass under a fixed budget. On PIE-Bench, they report the strongest background-preservation numbers in their Table 1 comparison, including 27.44 dB PSNR and 0.055 LPIPS, while maintaining competitive semantic alignment.

Significance. If the reported results are robust, the contribution is practically useful: a mask-free, fine-tuning-free editing procedure that works with a modern rectified-flow DiT and improves background preservation. The manuscript is unusually careful about the scope of its formal claims: it repeatedly states that hard projection is a token-state guarantee, not pixel-level invariance; that common-noise variance reduction requires nonnegative covariance; and that top-quantile selection is only a fixed-budget score-mass statement. These explicit scope limitations are a strength. The main significance is nonetheless conditional on the empirical evaluation being reproducible and on the off-manifold target latents not degrading model calibration.

major comments (4)

[§3.3, Eq. (7)] The coupled target latent is defined as z^{tar}_t = z^{src}_t + (x^{edit}_t - x^{src}_t). For the rectified-flow interpolation in Eq. (6), the same-noise target latent corresponding to the current editable state x^{edit}_t would be (1-t)x^{edit}_t + tε = z^{src}_t + (1-t)(x^{edit}_t - x^{src}_t). Eq. (7) therefore evaluates v_θ at points that are not the standard interpolation for x^{edit}_t; the deviation is t(x^{edit}_t - x^{src}_t). The velocity difference in Eq. (13), which drives the update in Eq. (14), is thus obtained from inputs where the model may be miscalibrated. The common-noise variance bound in Section 3.7 concerns variance, not bias, and does not cover this. Since the central empirical claim depends on this update, the paper should either provide evidence that the model remains calibrated at these inputs (e.g., compare against the consistent interpolation from Eq. (6), or measure clean-estimate error on typical edit states during sampling) or explicitly state the construction as an unvalidated heuristic and temper the corresponding claims.
[§4.1 and Table 1] The method's free parameters—the editable-fraction schedule ρ(t), EMA coefficient β, detection window [t_lo, t_hi], mask-update interval κ, Monte Carlo draws n_avg, and dilation rounds r—are not reported anywhere in the manuscript. Figure 5 studies only [t_lo, t_hi], on a single "Chicken → Duck" example, and the caption explicitly states that the plotted values are not dataset-level PIE-Bench averages. Without the exact hyperparameter values used for Table 1, the results cannot be reproduced, and the reported preservation gains could depend on choices tuned on a single example. Please report the full hyperparameter configuration and provide a dataset-level sensitivity study, or at least a small validation split, for the most influential parameters.
[§3.4 and Table 1] The core mechanism is the online editable-token mask derived from Token-wise Conditional Surprisal, but the manuscript never reports any direct measure of mask quality. PIE-Bench provides ground-truth edited-region masks; Section 4.1 says these are used only for evaluation metrics. Reporting, for example, the IoU between the final keep/edit mask and the ground-truth edited region, or a localization precision/recall curve across time, would show whether the surprisal ranking is actually finding the editable region rather than indirectly inferring this from PSNR/LPIPS. The current evidence leaves open the possibility that the preservation gains come from freezing a large fraction of tokens rather than from correct localization. Section 5 correctly admits the method's reliance on DiT calibration; a direct mask-quality metric would address this admitted limitation.
[Abstract and Table 1] The abstract and conclusion claim the "strongest reported preservation metrics" without qualification, but Table 1 mixes backbones: DiffEdit and InfEdit run on SD1.4, KVEdit and Stable Flow on FLUX, and only FlowEdit, FlowAlign, and ATDEdit on SD3. The authors acknowledge this in the body ("we treat FlowEdit and FlowAlign as the primary same-backbone SD3 comparisons"), but the headline claim is still stated without that caveat. Please qualify the abstract and conclusion, e.g., "strongest among the compared methods on this benchmark" or "strongest among the same-backbone SD3 comparisons," and ideally add a same-backbone re-run of at least one representative baseline. The table also reports point estimates without error bars; reporting variance over Monte Carlo seeds or small hyperparameter perturbations would strengthen the comparison.

minor comments (4)

[§3.2] The text references "Eq. (15)" as the Euclidean projection several paragraphs before Eq. (15) is introduced in Section 3.6; please use a forward-reference or restate the projection formula locally.
[§3.5] Please state explicitly whether the source memory (K^src, V^src) is computed at z^src_t or at the target latent z^tar_t; this affects the interpretation of Eq. (12).
[Figure 5] The axes are labeled "Tlo" and "Thi" while the text uses t_lo and t_hi; please unify the notation.
[Eq. (9)] The score divides by d (the feature dimension) but not by the number of Monte Carlo draws; please clarify whether the 1/d factor is an intentional feature-channel normalization.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's stated formal guarantees are explicitly scoped tautologies, and the headline empirical claim rests on external PIE-Bench evaluation rather than on the method's own construction.

full rationale

I examined the derivation chain for reductions to inputs. The hard-projection zero-drift statement (Eq. 15) is a direct consequence of the constraint set C(t) in Eq. (4), and the paper explicitly scopes it: 'the resulting zero-drift statement is exact for the selected latent rows in exact arithmetic, but it does not by itself imply pixel-level invariance after decoding.' Similarly, the top-quantile selection claim is a fixed-cardinality tautology that the paper labels as such: 'selecting the largest scores maximizes only the sum of selected surprisal values; it is not a claim of global optimality.' The Token-wise Conditional Surprisal is transparently defined as a scaled same-state velocity difference (Eq. 10), not presented as an information-theoretic quantity. The common-noise variance bound (Eqs. 16-17) is a standard variance identity, not a fitted result. The headline preservation metrics are evaluated on the external PIE-Bench benchmark using the ground-truth mask only for evaluation, and the authors disclose that cross-backbone comparisons are not isolated from backbone effects. The only self-citations ([20], [21]) appear in contextual related-work claims about the difficulty of preserving background under synchronous conditioning and about static masks not adapting to time-varying dynamics; these are not load-bearing for the method's construction or for Table 1. The reviewer-raised off-manifold concern about Eq. (7) is a correctness or calibration issue, not a circularity, since it concerns whether the model is well-behaved at the coupled latents rather than an equation reducing to its own input.

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The method relies on SD3 as a calibrated model, on the rectified-flow x0 estimate, and on the unproven transfer from latent token preservation to pixel-level background preservation. The paper states the last point explicitly as not guaranteed, which is honest but means the central empirical claim rests on it. No true free parameters are fitted to the benchmark in the derivation sense, but several hyperparameters (rho(t), beta, kappa, t_lo, t_hi, n_avg, r) are left unreported.

free parameters (6)

Editable fraction schedule rho(t) = not reported
Controls how many tokens are edited per step; hand-chosen and not given a value in the paper.
EMA coefficient beta = not reported
Smooths surprisal estimates over time; value not reported.
Detection window t_lo and t_hi = not reported
Define when mask updates happen; Figure 5 shows they trade PSNR against CLIP, but exact values and validation protocol are not given.
Mask update interval kappa = not reported
Sets how often the editable mask is recomputed to save compute; value not reported.
Coupled Monte Carlo draws n_avg = not reported
Number of shared-noise draws used to estimate surprisal and steering; value not reported.
Dilation rounds r = not reported
Optional morphological dilation applied to the 2D token mask; value not reported.

assumptions (5)

domain assumption Rectified-flow parameterization with clean estimate x0_hat = z_t - t * v_theta(z_t, t, c) is valid for SD3.
Invoked in Eq. (3) and used throughout Section 3 to define surprisal and updates; SD3 is released as a rectified-flow model.
domain assumption Pretrained SD3's velocity field is calibrated enough that token-wise surprisal ranks editability.
The entire mask discovery in Section 3.4 relies on this; the paper explicitly calls the score a heuristic, not a correctness certificate.
domain assumption Latent keep-token preservation transfers to pixel-level background preservation.
The projection guarantee in Eq. (15) is latent-only; Section 3.7 states it does not imply pixel-level invariance. The method's empirical success depends on this transfer.
standard math Common-noise variance inequality requires finite second moments and nonnegative covariance.
The variance decomposition in Section 3.7, Eqs. (16)-(17), is exact under those conditions.
standard math Frobenius projection separates over token rows.
Used to justify Eq. (15) as the Euclidean projection onto C(t).

invented entities (1)

Token-wise Conditional Surprisal
purpose: Select editable tokens online by measuring model prediction change under a condition switch.
Defined in Eq. (9) as a normalized L1 velocity difference; it has no falsifiable handle outside the SD3 model itself and the paper calls it a heuristic, so it carries no independent validation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Diffusion Image Editing via Asynchronous Token Decoding." pith.science (2026). https://pith.science/paper/XU7T7GTU

@misc{pith2026260809322,
  author       = {Pith},
  title        = {Pith review of: Diffusion Image Editing via Asynchronous Token Decoding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XU7T7GTU}},
  note         = {Machine review of arXiv:2608.09322}
}

read the original abstract

Text-guided diffusion image editing aims to modify semantic attributes of an image while preserving its identity, layout, and background. However, na\"ively switching the text condition during sampling often causes global drift, as denoising dynamics propagate changes across tokens and can disrupt unedited regions. To address this issue, we propose \textbf{A}synchronous \textbf{T}oken \textbf{D}ecoding \textbf{Edit} (ATDEdit), an inference-time framework that views each sampler step as a parallel update of a globally coupled token matrix and enables token-indexed condition switching with differentiated update policies. Instead of applying synchronous target-conditioned updates to all tokens, ATDEdit estimates editable locations using token-wise conditional surprisal and applies target-conditioned corrections to the selected token set. It supplies source key/value memory at keep-token positions and projects selected keep-token latent rows back to their source values; these operations promote background preservation but do not constitute a pixel-level invariance guarantee. This approach combines local editing and background preservation without external or user-provided spatial masks and without model fine-tuning. On PIE-Bench, ATDEdit achieves the strongest reported preservation metrics, including 27.44~dB PSNR and 0.055 LPIPS, while retaining competitive semantic alignment.

Figures

Figures reproduced from arXiv: 2608.09322 by the authors.

**Figure 1.** ATDEdit. These examples demonstrate our inference-time, training-free framework operating on Diffusion Transform [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗

**Figure 2.** Comparison between Synchronous and Asynchro [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** Overview of ATDEdit for transforming a fox into a dog. (Top) Same-state source/target evaluations produce Token-wise [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Comparison of edited results. Real images are in the first column. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: Sensitivity analysis. (a) Binary editable-token masks [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 11 canonical work pages

[1]

Omri Avrahami, Or Patashnik, Ohad Fried, Egor Nemchinov, Kfir Aberman, Dani Lischinski, and Daniel Cohen-Or. 2025. Stable flow: Vital layers for training-free image editing. InProceedings of the Computer Vision and Pattern Recognition Conference. 7877–7888

work page 2025
[2]

Tim Brooks, Aleksander Holynski, and Alexei A Efros. 2023. Instructpix2pix: Learning to follow image editing instructions. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 18392–18402

2023
[3]

Mingdeng Cao, Xintao Wang, Zhongang Qi, Ying Shan, Xiaohu Qie, and Yinqiang Zheng. 2023. Masactrl: Tuning-free mutual self-attention control for consistent image synthesis and editing. InProceedings of the IEEE/CVF international confer- ence on computer vision. 22560–22570

2023
[4]

Huiwen Chang, Han Zhang, Jarred Barber, AJ Maschinot, Jose Lezama, Lu Jiang, Ming-Hsuan Yang, Kevin Murphy, William T Freeman, Michael Rubinstein, et al

work page
[5]

Guillaume Couairon, Jakob Verbeek, Holger Schwenk, and Matthieu Cord. 2022. Diffedit: Diffusion-based semantic image editing with mask guidance.arXiv preprint arXiv:2210.11427(2022)

arXiv 2022
[6]

Aleksandar Cvejic, Abdelrahman Eldesokey, and Peter Wonka. 2025. Partedit: fine-grained image editing using pre-trained diffusion models. InProceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers. 1–11

work page 2025
[7]

Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. 2024. Scaling rectified flow transformers for high-resolution image synthesis. InForty- first international conference on machine learning

2024
[8]

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

arXiv 2022

Show all 44 references

[9]

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

2020
[10]

Xuan Ju, Ailing Zeng, Yuxuan Bian, Shaoteng Liu, and Qiang Xu. 2023. Direct inversion: Boosting diffusion-based editing with 3 lines of code.arXiv preprint arXiv:2310.01506(2023)

2023 arXiv
[11]

Ronen Kamenetsky, Sara Dorfman, Daniel Garibi, Roni Paiss, Or Patashnik, and Daniel Cohen-Or. 2025. SAEdit: Token-level control for continuous image editing via Sparse AutoEncoder.arXiv preprint arXiv:2510.05081(2025)

2025
[12]

Bahjat Kawar, Shiran Zada, Oran Lang, Omer Tov, Huiwen Chang, Tali Dekel, Inbar Mosseri, and Michal Irani. 2023. Imagic: Text-based real image editing with diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 6007–6017

2023
[13]

Jeongsol Kim, Yeobin Hong, Jonghyun Park, and Jong Chul Ye. 2025. Flowalign: Trajectory-regularized, inversion-free flow-based image editing.arXiv preprint arXiv:2505.23145(2025)

2025 arXiv
[14]

Vladimir Kulikov, Matan Kleiner, Inbar Huberman-Spiegelglas, and Tomer Michaeli. 2025. Flowedit: Inversion-free text-based editing using pre-trained flow models. InProceedings of the IEEE/CVF International Conference on Computer Vision. 19721–19730

2025
[15]

Zixu Li, Yupeng Hu, Zhiwei Chen, Qinlei Huang, Guozhi Qiu, Zhiheng Fu, and Meng Liu. 2026. ReTrack: Evidence-Driven Dual-Stream Directional Anchor Calibration Network for Composed Video Retrieval. InAAAI, Vol. 40. 23373– 23381

2026
[16]

Zixu Li, Yupeng Hu, Zhiwei Chen, Haokun Wen, Xuemeng Song, and Liqiang Nie. 2026. COMBINER: Composed Image Retrieval Guided by Attribute-based Neighbor Relations.IEEE TIP(2026)

2026
[17]

Zixu Li, Yupeng Hu, Zhiwei Chen, Mingyu Zhang, Zhiheng Fu, and Liqiang Nie
[18]

Qiang Liu. 2022. Rectified flow: A marginal preserving approach to optimal transport.arXiv preprint arXiv:2209.14577(2022)

2022 arXiv
[19]

Xingchao Liu, Chengyue Gong, and Qiang Liu. 2022. Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003(2022)

2022 arXiv
[20]

Liangsi Lu, Xuhang Chen, Minzhe Guo, Shichu Li, Jingchao Wang, and Yang Shi
[21]

Liangsi Lu, Minzhe Guo, Xuhang Chen, and Yang Shi. 2026. Semantic Granularity Navigation in Image Editing.arXiv preprint arXiv:2605.21190(2026)

2026 arXiv
[22]

Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon. 2021. Sdedit: Guided image synthesis and editing with stochastic differential equations.arXiv preprint arXiv:2108.01073(2021)

2021 arXiv
[23]

Chordedit: One-step low-energy transport for image editing.arXiv preprint arXiv:2602.19083(2026)

2026
[24]

William Peebles and Saining Xie. 2023. Scalable diffusion models with transform- ers. InProceedings of the IEEE/CVF international conference on computer vision. 4195–4205

2023
[25]

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
[26]

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

2023
[27]

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

2020 arXiv
[28]

Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. 2020. Score-based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456(2020)

2020 arXiv
[29]

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
[30]

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need.Advances in neural information processing systems30 (2017)

2017
[31]

Yihao Wang, Yuheng Ji, Mingyu Cao, Yanqing Shen, Runze Xiao, Huaihai Lyu, Senwei Xie, Euan Liu, Klara Tian, Tianfeng Long, et al. 2026. Orca: The World is in Your Mind.arXiv preprint arXiv:2606.30534(2026)

2026 arXiv
[32]

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

2023
[33]

Zirui Wang, Zhizhou Sha, Zheng Ding, Yilin Wang, and Zhuowen Tu. 2024. Tokencompose: Text-to-image diffusion with token-level supervision. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 8553–8564

2024
[34]

Sihan Xu, Yidong Huang, Jiayi Pan, Ziqiao Ma, and Joyce Chai. 2023. Inversion- free image editing with natural language.arXiv preprint arXiv:2312.04965(2023)

2023 arXiv
[35]

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
[36]

Renye Yan, Jikang Cheng, Shikun Sun, Yi Sun, You Wu, Wei Peng, Zongwei Wang, Ling Liang, Junliang Xing, and Yimao Cai. 2026. Do Less, Achieve More: Do We Need Every-Step Optimization for RL Fine-tuning of Diffusion Models?arXiv preprint arXiv:2605.15855(2026)

2026 arXiv
[37]

Renye Yan, Jikang Cheng, You Wu, Ling Liang, Wei Peng, Athanasios V Vasilakos, Qingyu Zhao, Yu Zhang, Ehsan Adeli, Kilian M Pohl, et al . 2026. Pixel-Space Diffusion Transformers.arXiv preprint arXiv:2607.17585(2026)

2026 arXiv
[38]

Renye Yan, Jikang Cheng, Yaozhong Gan, Shikun Sun, You Wu, Yunfan Yang, Liang Ling, Jinlong Lin, Yeshuang Zhu, Jie Zhou, et al. 2025. Entropy-adaptive diffusion policy optimization with dynamic step alignment. InProceedings of the IEEE/CVF International Conference on Computer ...

2025
[39]

Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang
[40]

Tianrui Zhu, Shiyi Zhang, Jiawei Shao, and Yansong Tang. 2025. Kv-edit: Training-free image editing for precise background preservation.arXiv preprint arXiv:2502.17363(2025)

2025 arXiv
[41]

Wei-Ning Yang and Barry L Nelson. 1991. Using common random numbers and control variates in multiple-comparison procedures.Operations Research39, 4 (1991), 583–591

1991
[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
[2023]

Muse: Text-to-image generation via masked generative transformers.arXiv preprint arXiv:2301.00704(2023)

2023 arXiv
[2026]

InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Conesep: Cone-based robust noise-unlearning compositional network for composed image retrieval. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 16897–16909

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Omri Avrahami, Or Patashnik, Ohad Fried, Egor Nemchinov, Kfir Aberman, Dani Lischinski, and Daniel Cohen-Or. 2025. Stable flow: Vital layers for training-free image editing. InProceedings of the Computer Vision and Pattern Recognition Conference. 7877–7888

work page 2025

[2] [2]

Tim Brooks, Aleksander Holynski, and Alexei A Efros. 2023. Instructpix2pix: Learning to follow image editing instructions. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 18392–18402

2023

[3] [3]

Mingdeng Cao, Xintao Wang, Zhongang Qi, Ying Shan, Xiaohu Qie, and Yinqiang Zheng. 2023. Masactrl: Tuning-free mutual self-attention control for consistent image synthesis and editing. InProceedings of the IEEE/CVF international confer- ence on computer vision. 22560–22570

2023

[4] [4]

Huiwen Chang, Han Zhang, Jarred Barber, AJ Maschinot, Jose Lezama, Lu Jiang, Ming-Hsuan Yang, Kevin Murphy, William T Freeman, Michael Rubinstein, et al

work page

[5] [5]

Guillaume Couairon, Jakob Verbeek, Holger Schwenk, and Matthieu Cord. 2022. Diffedit: Diffusion-based semantic image editing with mask guidance.arXiv preprint arXiv:2210.11427(2022)

arXiv 2022

[6] [6]

Aleksandar Cvejic, Abdelrahman Eldesokey, and Peter Wonka. 2025. Partedit: fine-grained image editing using pre-trained diffusion models. InProceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers. 1–11

work page 2025

[7] [7]

Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. 2024. Scaling rectified flow transformers for high-resolution image synthesis. InForty- first international conference on machine learning

2024

[8] [8]

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

arXiv 2022