- Rating
- 3(+1)
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper introduces PersistEdit, a training-free and inversion-free image editing method built upon FlowEdit. The paper argues that independently resampling noise at each step leads to inconsistent realization-specific editing tendencies. To address this, PersistEdit introduces noise persistence to correlate noise across steps within the editing region, and semantic steering to select target-favorable noise innovations based on attention responses. It additionally applies a source-latent preservation constraint outside the editing mask. Experiments on part of the PIE-Bench show promising editing and background-preservation trade-off.
Critical strengths
-
The paper is easy to follow.
-
The paper provides a clear motivation for studying cross-step noise dependence in inversion-free flow editing and proposes a targeted solution. The analysis of persistent noise offers an interesting perspective on the realization-specific variations in FlowEdit.
Critical weaknesses
-
The mask estimation pipeline and its computational cost are unclear. According to Algorithm 1, the complete editing mask must be estimated before the editing process. However, TAM aggregates attention responses across steps 5–28 and multiple layers, which appears to require additional multi-step model forward. The paper does not explain how this pass is implemented or whether its cost is included in the reported inference time. Figure 9 only compares early-layer and full-layer candidate scoring for steering, rather than providing an end-to-end comparison with FlowEdit. The paper should compare complete runtime, NFE with baselines, including both TAM and candidate selection.
-
The contributions of the individual designs are not sufficiently isolated. Noise Persistence is motivated as a way to strengthen semantic editing by allowing target changes to develop more completely. However, the “+Persistence” configuration also produces a large improvement in background preservation, which appears to be partly caused by the separate source-latent preservation constraint. The current ablation does not isolate persistence from this constraint. Moreover, the mask itself is a central component because it determines where persistence and steering are applied and where the source latent is copied. The paper should separately evaluate persistence-only, preservation-only, steering-only, and their combinations, with estimated masks and no masks. In particular, applying persistence or steering globally would help clarify whether their effectiveness depends mainly on the localized mask.
-
The experimental evaluation is not sufficiently comprehensive. The experiment misses SNR-Edit, a closely related and open-source method that directly improves FlowEdit through noise rectification. Other recent flow-based editing methods, such as FIA-Edit [1] and DRFS [2], should also be compared or discussed more carefully. In addition, the claimed transferability to SD3-Medium is supported only by qualitative examples. Quantitative comparisons against methods that support the SD family, such as SNR-Edit, DNAEdit, and FlowEdit, are necessary. Besides, completely removing the style-transfer category is questionable. FlowEdit is designed as a general editing method, and prior methods such as DNAEdit and FlowAlign evaluate on the complete PIE-Bench. Style transfer would also provide a useful setting for examining persistence and steering when the editing mask covers most or all of the image. Results on the excluded category should be provided. Finally, evaluation on long-prompt and complex editing such as DNA-Bench should also be compared and discussed.
-
The metrics in Table 1 are incomplete. Whole-image CLIP similarity and Structure Distance, both commonly reported on PIE-Bench, should be included.
[1] FIA-Edit: Frequency-Interactive Attention for Efficient and High-Fidelity Inversion-Free Text-Guided Image Editing. AAAI 2026.
[2] Delta Rectified Flow Sampling for Text-to-Image Editing. CVPR 2026.
Questions for authors
-
How is the number of K selected? Any ablation?
-
How well does the attention score corresponds to the final editing quality? A quantitative comparison among highest-score, random, and lowest-score candidate selection would be more convincing than the current qualitative examples.
7 comments
W3 & W4. Expanded Evaluation and Metrics (1/2)
Additional Metrics and SNR-Edit. Thank you for these suggestions. Table 5 below adds CLIP-Whole and Structure Distance, along with SNR-Edit using its official FLUX.1-dev implementation. SNR-Edit’s structure-aware noise rectification helps limit trajectory drift, consistent with its slightly lower Structure Distance. PersistEdit improves background PSNR by 2.37 dB and CLIP-Edited by 0.99 over SNR-Edit.
FIA-Edit on SD3.5 and DRFS on SD3. We added quantitative comparisons with FlowEdit, FlowAlign, and DRFS on SD3-Medium in Appendix G (Table 12), alongside qualitative examples in Fig. 15. DRFS shifts noisy latents toward the target trajectory and achieves strong semantic alignment. PersistEdit comes within 0.14 of DRFS on CLIP-Edited while improving background PSNR by 3.96 dB and achieving lower Structure Distance. On SD3.5-Medium (Table 6 below), PersistEdit and FIA-Edit both improve background PSNR over FlowEdit (+2.85 and +4.20 dB, respectively). PersistEdit also improves CLIP-Edited by 0.18 over FlowEdit, compared with a change of −0.83 for FIA-Edit.
Table 5. Quantitative Comparison on PIE-Bench; * indicates a newly added baseline.
| Method | Structure Distance ×10³ ↓ | HPS ×10² ↑ | AS ↑ | Background PSNR (dB) ↑ | Background LPIPS ×10³ ↓ | Background MSE ×10⁴ ↓ | CLIP-Whole ↑ | CLIP-Edited ↑ | IR ×10² ↑ |
|---|---|---|---|---|---|---|---|---|---|
| P2P | 71.93 | 25.40 | 6.25 | 17.83 | 208.30 | 220.69 | 25.62 | 22.08 | -4.47 |
| P2P (PnP-Inv) | 11.16 | 26.07 | 6.33 | 27.25 | 52.64 | 34.13 | 25.56 | 21.76 | 22.67 |
| PnP | 25.31 | 26.59 | 6.38 | 22.41 | 109.89 | 81.19 | 25.89 | 22.33 | 40.89 |
| MasaCtrl | 23.92 | 24.93 | 6.13 | 22.83 | 81.14 | 79.18 | 25.13 | 21.14 | 5.44 |
| RF-Inv | 51.96 | 27.50 | 6.46 | 19.34 | 211.26 | 167.74 | 25.41 | 21.93 | 40.86 |
| KV-Edit | 24.40 | 27.22 | 6.42 | 27.32 | 56.49 | 43.45 | 25.46 | 22.12 | 58.76 |
| DNAEdit | 17.69 | 27.68 | 6.51 | 25.25 | 85.84 | 47.45 | 25.06 | 21.85 | 55.02 |
| FlowEdit | 24.81 | 28.32 | 6.59 | 22.57 | 97.81 | 83.36 | 25.71 | 22.01 | 56.48 |
| FlowAlign | 13.87 | 27.81 | 6.51 | 25.89 | 53.39 | 42.75 | 25.30 | 21.56 | 38.49 |
| SNR-Edit* | 16.81 | 27.66 | 6.49 | 24.57 | 77.14 | 54.35 | 25.30 | 21.53 | 38.21 |
| Ours | 17.41 | 28.41 | 6.54 | 26.94 | 49.80 | 36.86 | 25.81 | 22.52 | 71.03 |
Table 6. Quantitative Comparison on SD3.5-Medium
| Method | Structure Distance ×10³ ↓ | Background PSNR (dB) ↑ | CLIP-Whole ↑ | CLIP-Edited ↑ | IR ×10² ↑ | HPS ×10² ↑ | AS ↑ |
|---|---|---|---|---|---|---|---|
| FlowEdit | 23.94 | 23.20 | 27.05 | 23.19 | 102.92 | 28.47 | 6.48 |
| FIA-Edit | 9.94 | 27.40 | 26.19 | 22.36 | 75.25 | 27.33 | 6.41 |
| Ours | 19.31 | 26.05 | 27.20 | 23.37 | 112.18 | 28.79 | 6.44 |
W3 & W4. Expanded Evaluation and Metrics (2/2)
Style Transfer on PIE-Bench. We initially excluded style transfer from the aggregate because its masks leave little unedited area for assessing background preservation, and methods generally score higher on CLIP-Edited for this category. Table 7 reports all ten categories. PersistEdit leads on style transfer (26.24) and object replacement, addition, and removal. Non-rigid pose changes remain challenging.
Long-Prompt Editing on DNA-Bench. DNA-Bench uses the same source images and tasks as PIE-Bench with longer, more detailed prompts. We include change_style. PersistEdit gains 1.36 dB in background PSNR over DNAEdit and leads on CLIP-Edited (23.45) and ImageReward (82.71) (Table 8). With the same eight TAM timesteps, pixel/token IoU is 61.24%/68.35% on DNA-Bench vs. 56.36%/64.04% on PIE-Bench. These results demonstrate PersistEdit’s effectiveness for long-prompt editing.
Table7. Category-wise CLIP-Edited comparison on PIE-Bench.
| Method | #0:random | #1:change object | #2:add object | #3:delete object | #4:change attribute content | #5:change attribute pose | #6:change attribute color | #7:change attribute material | #8:change background | #9:change style |
|---|---|---|---|---|---|---|---|---|---|---|
| RF-Inversion | 22.54 | 20.19 | 24.42 | 17.68 | 22.97 | 21.87 | 21.34 | 23.37 | 23.47 | 26.01 |
| KV-Edit | 22.92 | 21.22 | 25.17 | 17.80 | 22.11 | 22.04 | 20.13 | 23.29 | 23.38 | 26.12 |
| DNA-Edit | 22.57 | 20.73 | 24.60 | 18.23 | 22.02 | 22.01 | 20.51 | 23.46 | 22.29 | 25.67 |
| FlowEdit | 22.83 | 20.54 | 24.59 | 18.37 | 22.54 | 22.23 | 20.80 | 23.62 | 22.51 | 25.98 |
| FlowAlign | 22.41 | 20.57 | 23.97 | 17.75 | 22.40 | 21.82 | 20.24 | 23.62 | 21.53 | 24.61 |
| Ours | 23.25 | 21.47 | 25.47 | 18.47 | 22.78 | 21.22 | 21.32 | 23.84 | 23.77 | 26.24 |
Table8. Quantitative Comparison on DNA-Bench.
| Method | Structure Distance ×10³ ↓ | Background PSNR (dB) ↑ | CLIP-Whole ↑ | CLIP-Edited ↑ | IR ×10² ↑ | HPS ×10² ↑ | AS ↑ |
|---|---|---|---|---|---|---|---|
| FlowEdit | 29.07 | 21.61 | 28.13 | 23.27 | 74.98 | 28.67 | 6.67 |
| DNAEdit | 18.73 | 25.39 | 28.34 | 22.99 | 70.93 | 28.13 | 6.48 |
| Ours | 20.43 | 26.75 | 28.54 | 23.45 | 82.71 | 28.83 | 6.56 |
Q1. Candidate Pool Size
We ablated on PIE-Bench (Appendix F.2). CLIP-Edited rises from 22.37 at to 22.54 at , with only 0.02 more at . We use to balance semantic alignment and scoring cost.
Q2. Attention Score and Editing Quality
From lowest-score to random to highest-score selection, CLIP-Edited rises from 21.82 to 22.01 to 22.18, and ImageReward from 51.37 to 56.48 to 64.11 (Table 9 below). Highest-score selection improves alignment over random at similar Structure Distance and background PSNR. With our qualitative examples, this supports attention scores as a proxy for target alignment.
Table 9. Effect of Attention-Based Candidate Selection on Editing Quality
| Candidate Selection | Structure Distance ×10³ ↓ | PSNR (dB) ↑ | CLIP-Whole ↑ | CLIP-Edited ↑ | IR ×10² ↑ | HPS ×10² ↑ | AS ↑ |
|---|---|---|---|---|---|---|---|
| Lowest-score | 24.86 | 22.40 | 25.65 | 21.82 | 51.37 | 28.29 | 6.60 |
| Random | 24.81 | 22.57 | 25.71 | 22.01 | 56.48 | 28.32 | 6.59 |
| Highest-score | 24.83 | 22.56 | 25.78 | 22.18 | 64.11 | 28.53 | 6.61 |
W1. End-to-End Runtime.
We construct the TAM mask before editing in a separate source-conditioned pass. To reduce this cost, we aggregate attention over eight sampled timesteps, which yields mask IoU close to that of 24-step TAM. The binary-mask IoU results on PIE-Bench are shown in Table 1 below and have also been added to Appendix A.2. We have updated Appendix D (Table 5) with end-to-end runtime comparisons on FLUX.1-dev. As shown in Table 2 below, PersistEdit takes 22.19 s per image, including 2.32 s for TAM construction, all Steering candidate evaluations, and decoding. With an existing edit mask, the runtime is 19.87 s per image.
Table 1. Effect of the number of TAM aggregation timesteps on mask IoU on PIE-Bench (FLUX.1-dev).
Table 2: Per-image runtime (s) on FLUX.1-dev.
W2. Component Contributions and Spatial Localization
We have updated Section 4.3 to explicitly separate the contributions of Noise Persistence, Semantic Steering, and Background Constraint (BC), including Persistence-only, Steering-only, and BC-only controls. All variants reported in Section 4.3 use TAM-estimated masks rather than ground-truth masks. Persistence and Steering individually improve semantic alignment, and their combination further improves CLIP-Edited and ImageReward to 22.61 and 78.96, respectively. These results support their complementary roles in promoting more complete target-semantic changes. BC accounts for the main background-preservation gain, increasing PSNR from 22.57 to 26.98 dB.
To further examine the benefits of Noise Persistence, we tested whether it improves semantic alignment at comparable accumulated-update variance. We compared Noise Persistence with an i.i.d. control that samples noise independently across timesteps and approximately matches the variance of accumulated updates within the editing region. Semantic Steering and Background Constraint were disabled in both settings. As shown in Table 3 below, Noise Persistence achieves higher CLIP-Edited (23.25 vs. 22.97) and ImageReward (59.04 vs. 52.72) at comparable Structure Distance and background PSNR. These gains support our proposed mechanism that cross-step noise dependence helps editing tendencies accumulate coherently. Experiment details are provided in Appendix E.4.
We additionally compare global and localized Persistence, with Semantic Steering and BC disabled (Table 4 below). Global Persistence achieves higher semantic alignment scores, but reduces source fidelity: PSNR decreases from 22.56 to 21.24 dB, while Structure Distance increases from 28.07 to 35.04. This demonstrates that semantic gains remain achievable without masks, while localization helps preserve source content. Semantic Steering likewise selects candidate innovations using target-token attention within the editing region; a global scoring objective would remove this spatial constraint. Figure 13 further suggests that early Steering helps establish the target object’s placement and spatial extent. Together, these observations motivate region-aware control, for which TAM provides automatic spatial support.
Table 3. Comparison with a variance-matched i.i.d. control on PIE-Bench.
Table 4: Effect of TAM-based localization on Noise Persistence. Semantic Steering and Background Constraint are disabled.