- Rating
- 3
- Confidence
- 3
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper proposes RACE (Relation-Aware Consistency-Preserving Editing) for reference-image-guided relation editing, aiming to transfer fine-grained object relations from a reference image to a source image while preserving its original appearance and identity. RACE introduces Relation-Aware Token Bottleneck Attention (RTBA) to disentangle relation semantics from reference appearance, and Cross-Modal Semantic Shift Alignment (CSSA) with a Semantic Shift Predictor (SSP) to suppress unintended semantic changes during denoising. The paper also constructs RTD-35K, a dataset containing 35K image triplets across 235 relation categories. Experiments on HICORTBench and RTBench demonstrate competitive performance in relation transfer, source-image preservation, and generation quality.
Critical strengths
- Well-motivated and practically relevant task. The paper addresses reference-image-guided relation editing, a challenging and practical task. It clearly identifies the limitations of existing guidance methods and highlights the challenges of relation–appearance entanglement and unintended semantic shifts.
- Intuitive and well-structured method design. The proposed RACE framework combines RTBA to reduce reference-appearance leakage and CSSA to suppress unintended semantic shifts. The two components are complementary and well-motivated.
- Extensive experiments and promising results. The paper constructs RTD-35K, covering 235 relation categories, and demonstrates competitive performance through quantitative comparisons, ablation studies, and qualitative results on challenging scenarios.
Critical weaknesses
-
Insufficient clarity and fairness in baseline comparisons. The paper uses different types of relation guidance across baselines. While FLUX.1-Kontext and OneHOI are adapted to accept reference images and retrained on RTD-35K, other models, such as OmniGen2, DyRef, and Nano Banana, also support reference-image inputs. However, the paper does not clearly specify their actual input configurations. Moreover, relevant methods such as OrionEdit, FreeEvent, and ReVersion are only compared qualitatively. The paper should clarify the input conditions for each baseline and provide more comprehensive quantitative comparisons under comparable settings.
-
Inconsistent relation transfer performance and potential trade-off introduced by CSSA. Although RACE achieves superior relation consistency on HICORTBench, it underperforms DyRef on RTBench (0.2985 vs. 0.3443), without explaining this discrepancy. Furthermore, the ablation study shows that introducing CSSA decreases relation consistency from 0.3141 to 0.2985, despite improving EI. This suggests a potential trade-off between relation transfer accuracy and source-image preservation. The paper could further analyze this trade-off across different relation categories and CSSA loss weights.
-
Lack of a dedicated quantitative ablation for RTBA. Although the paper provides qualitative ablation results for RTBA, it lacks a dedicated quantitative ablation that isolates its contribution.
Questions for authors
Could the paper include comparisons with more recent and stronger commercial image editing models, such as GPT Image 2? Nano Banana is a relatively outdated baseline, and comparisons with more advanced models would provide a more comprehensive evaluation of RACE's competitiveness.
Related work on balancing semantic editing and source preservation. SynPS (The Devil is in Attention Sharing: Improving Complex Non-rigid Image Editing Faithfulness via Attention Synergy, CVPR 2026) shares a similar motivation with CSSA, as both aim to balance intended semantic modifications and source-image preservation. More specifically, both methods leverage source–target differences during denoising to characterize editing-induced changes, rather than relying solely on final generated images. Although their technical implementations differ, the paper would benefit from citing SynPS and discussing the similarities and differences between the two approaches.
6 comments
Response to Reviewer sSje (Part 2/ 4)
W2: Explaining the CSSA Trade-off and Hyperparameter Sensitivity
We thank the reviewer for this insightful observation regarding the decrease in relation consistency after introducing CSSA. We address this observation by (1) explaining the relation–preservation trade-off and (2) analyzing the sensitivity of CSSA to training steps and loss weight.
1. Explaining the Relation–Preservation Trade-off
Trade-off Mechanism. As discussed in the fourth paragraph of Sec. 1, CSSA aims to preserve the original appearance of the source image by aligning editing-induced image semantic shifts with the intended textual relation shift. However, since textual relation descriptions cannot fully capture fine-grained visual configurations, this semantic constraint may also limit fine-grained relation transfer by suppressing some necessary structural adjustments.
Empirical Evidence. The ablation study in Table 2 further confirms the trade-off introduced by CSSA between relation transfer and source-appearance preservation. Introducing CSSA slightly reduces relation consistency (0.3141→0.2985) while improving EI (0.3708→0.3851). This reflects its intended effect of better preserving source appearance.
2. Analyzing the sensitivity of CSSA to training steps and loss weight
(i) CSSA Loss Weight: As the reviewer pointed out, the CSSA loss weight plays an important role in balancing relation transfer and appearance preservation. We report the effects of the CSSA loss weight in the table below (Table 8, Supplementary Sec. D.2). Specifically, a small weight () provides insufficient semantic alignment to suppress relation-irrelevant shifts. Conversely, an excessively large weight () overemphasizes semantic alignment at the expense of the flow-matching objective, thereby degrading source-image preservation and image quality. In comparison, achieves the best overall balance across the six metrics.
| Parameter | Value | relation consistency ↑ | EI ↑ | HPSv2 ↑ | ImageReward ↑ | PickScore ↑ | VLM-R↑ |
|---|---|---|---|---|---|---|---|
| 0.5 | 0.2672 | 0.3377 | 0.2700 | 0.7223 | 21.7593 | 7.1413 | |
| 1.25 | 0.2739 | 0.3747 | 0.2998 | 1.0457 | 22.0214 | 8.0320 | |
| 2.0 (Ours) | 0.2985 | 0.3851 | 0.2913 | 1.2699 | 22.4189 | 8.5579 | |
| 2.75 | 0.2816 | 0.3453 | 0.2940 | 1.3378 | 21.8418 | 8.4322 | |
| 3.5 | 0.2920 | 0.3574 | 0.2698 | 0.7045 | 21.8534 | 8.1857 |
(ii) CSSA Joint-Training Steps: We also investigate the effect of CSSA joint-training steps, as shown in the table below (Table 8). Increasing the number of joint-training steps from 600 to 1,200 improves EI (0.3821→0.3851). In contrast, extending training to 2,100 steps degrades both relation consistency and EI. We hypothesize that prolonged CSSA training may cause SSP to learn a shortcut. Specifically, SSP may directly predict relation changes rather than model the complete image semantic shift.
| Parameter | Value | relation consistency ↑ | EI ↑ | HPSv2 ↑ | ImageReward ↑ | PickScore ↑ | VLM-R↑ |
|---|---|---|---|---|---|---|---|
| 600 | 0.3182 | 0.3821 | 0.2784 | 0.9413 | 21.9392 | 8.4224 | |
| 900 | 0.2871 | 0.3833 | 0.2787 | 0.9044 | 22.0543 | 8.6857 | |
| 1,200 (Ours) | 0.2985 | 0.3851 | 0.2913 | 1.2699 | 22.4189 | 8.5579 | |
| 1500 | 0.2743 | 0.3718 | 0.3078 | 0.8198 | 22.1234 | 8.3689 | |
| 2,100 | 0.2380 | 0.3045 | 0.2703 | 0.6932 | 21.7103 | 7.7293 |
Response to Reviewer sSje (Part 3 / 4)
W3: Quantitatively and Qualitatively Evaluating RTBA
We thank the reviewer for suggesting a dedicated evaluation of RTBA. We provide both quantitative and qualitative ablations to further demonstrate its contribution to relation transfer and source-content preservation.
1. Quantitative Ablation
Quantitative Results. Following your suggestion, we have added a w/o RTBA variant to Table 2 to isolate the contribution of RTBA to the overall performance of RACE, as shown in the table below.
| Method | Relation Consistency ↑ | EI ↑ | HPSv2 ↑ | ImageReward ↑ | PickScore ↑ | VLM-R ↑ |
|---|---|---|---|---|---|---|
| Baseline | 0.2770 | 0.3308 | 0.2535 | 0.4918 | 21.2388 | 5.7437 |
| w/o RTBA | 0.2715 | 0.3537 | 0.2779 | 0.9246 | 21.8961 | 5.5786 |
| w/o CSSA | 0.3141 | 0.3708 | 0.2748 | 0.9012 | 21.9331 | 8.4369 |
| w/o SSP | 0.2771 | 0.3417 | 0.2710 | 0.7699 | 21.7896 | 4.6917 |
| w/o R-CFG | 0.1782 | 0.2630 | 0.2720 | 0.6773 | 21.7999 | 3.6385 |
| Ours | 0.2985 | 0.3851 | 0.2913 | 1.2699 | 22.4189 | 8.5579 |
Quantitative Results Analysis. As shown in the table, removing RTBA leads to consistent performance degradation across all six metrics. These results demonstrate the contribution of RTBA to relation transfer, source-content preservation, and overall image quality.
Response to Reviewer sSje (Part 1 / 4)
W1 & Q1: Clarifying Comparison Configurations and Expanding Comparisons
We thank the reviewer for raising concerns about the fairness and completeness of our baseline comparisons. We address these concerns through two aspects: (1) clarifying the input configurations of different comparison methods; and (2) expanding quantitative and qualitative comparisons with additional and more recent models.
1. Clarifying Input Configurations of Comparison Methods
Following your suggestion, we have revised Sec. 4.2 to explicitly clarify the input configurations of all comparison methods. Specifically,
OmniGen2,DyRef,Nano Banana,FluxKontext (w/ ref.),Qwen-Image-Edit,HOIEdit,Baseline (w/o train), andBaselineare evaluated with reference-image inputs.2. Expanding Quantitative and Qualitative Comparisons
Expanded Baselines and Evaluation. We expand the quantitative comparisons by including
OrionEdit,FreeEvent,ReVersion, andGPT Image 2. In response to your suggestion, we have also incorporated more recent models, includingNano Banana 2and the latestQwen-Image-2.1. We additionally introduceVLM-R, a complementary VLM-based evaluation metric usingQwen3.8-27Bto assess fine-grained visual relation consistency on a scale of 0–10.Quantitative Comparison. We further compare
RACEwith the newly added methods on HICORTBench. All methods are evaluated with reference-image inputs, and the results are summarized in the table below (Table 10).RACEachieves the best performance across all six metrics.Qualitative Comparison. We additionally include qualitative comparisons with the latest open-source model,
Qwen-Image-2.1, in Fig. 10. WhileQwen-Image-2.1captures the general interaction, it struggles to reproduce the reference-specific fine-grained configurations. In contrast,RACEbetter transfers the reference interactions while preserving source content. Fig. 11 further comparesRACEwith commercial models, includingNano Banana 2andGPT-Image-2. Although these models generate visually plausible results, they often exhibit reference-appearance leakage. For example,Nano Banana 2reproduces the star-shaped detail from the reference image in the third row, whileGPT-Image-2introduces the reference box into its outputs in the first and second rows. In contrast,RACEtransfers the intended interactions without such appearance leakage.Additional Manuscript Details. We have included the additional comparison results and corresponding analyses in Supplementary Sec. D.6.