- Rating
- 3
- Confidence
- 3
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper proposes a training-free inference framework for unified multimodal models. During denoising, it estimates a clean image, asks the same model to produce a state-dependent target description, and backpropagates the description's autoregressive negative log-likelihood into the latent representation. GITO performs multiple local updates, after which candidates are selected using the likelihood of the original user instruction before denoising resumes. The main contribution is the combination of model-native reflection, a differentiable semantic objective, and candidate selection, rather than a new flow-matching or gradient-guidance principle. Experiments cover four flow-based models on GenEval and DPG-Bench, with analyses of semantic targets, intervention windows, and inference cost.
Critical strengths
- The central design is supported by targeted controls, not just a reflection narrative. With selection based on the user instruction, Table 3 reports 79.1 for directly optimizing the original instruction, 78.1 for text-only expansion, 79.4 for static reflection, and 80.4 for dynamic reflection. Using the reflective target for selection instead reduces the score to 78.8. These comparisons support both image-conditioned feedback and the separation between reflection for proposing corrections and the original instruction for selecting candidates. The paper explicitly acknowledges its use of existing gradient-guidance mechanisms and generally positions the contribution appropriately (Section 5.4).
- The cross-model improvements and complementarity are practically meaningful. All four models improve in overall score on both benchmarks. GenEval gains range from 1.0 percentage point for OmniGen2 to 7.1 points for Janus-Flow, while BAGEL's counting score rises from 74.3 to 81.9 (Table 1). BAGEL Think further improves from 79.7 to 81.3, suggesting that trajectory rectification can complement rather than merely replace pre-generation reasoning (Table 2). The evidence spans different baseline strengths, although applicability should remain limited to the evaluated flow-based UMMs.
- The appendix meaningfully examines the repair–preservation trade-off. On paired samples, the paper reports 86 repaired failures and 28 regressions among previously correct outputs, yielding a net gain of 58 and a regression rate of 1.65%, rather than presenting only successful examples (Table 7). Efficiency is also reported in absolute terms: the method takes 43.2 seconds per image versus 24.7 for the baseline, but compares favorably with the evaluated Best-of-3 configuration at 76.9 seconds and a score of 80.1, versus 80.4 for UniRect-CoT (Table 6). This provides useful evidence for the quality–cost trade-off.
Critical weaknesses
- Evidence for proxy reliability is weaker than evidence for final performance. Figure 3 reports only aggregate accuracy when judging whether intermediate images satisfy the instruction, without the proportions of correct and incorrect examples, confusion matrices, or category-level results. It also does not directly test whether the candidate ranking in Equation (11) predicts improvements in the final image (Sections 4.1–4.3). The paper already acknowledges that likelihood is a proxy, and quantifies regressions, so the concern is not an absence of failure discussion. Rather, trajectory-level diagnostics are missing for incorrect reflections and cases where proxy scores improve while final outputs deteriorate. Moreover, BAGEL's position score decreases from 49.0 to 48.5 (Table 1), so the aggregate gain should not be interpreted as improving every compositional capability, much less overall visual quality based on alignment metrics alone.
- Statistical robustness and reproduction details limit the strength of comparative claims. Table 1 specifies four consecutive samples per prompt from a fixed initial seed, but Tables 1–3 do not report repeated-run variation or paired confidence intervals. The overall advantages over TiR and Best-of-3 are only 0.6 and 0.3 percentage points, respectively (Tables 2 and 6). The paper sweeps update counts and intervention windows, but does not explain whether the preferred settings were selected using independent validation prompts; this is a clarification request, not evidence of test-set leakage.
Questions for authors
- Could you provide prompt-paired confidence intervals or results across multiple seed sets for the key differences in Tables 1–3, and clarify whether the hyperparameters in Tables 4–5 were selected on an independent validation set?
- What are the label distributions, balanced accuracy, and error types at each timestep in Figure 3? How strongly does the candidate ranking in Equation (11) correspond to final GenEval correctness?
7 comments
Response to Reviewer Ci3p (Part 2)
W1.3: The paper already acknowledges that likelihood is a proxy, and quantifies regressions, so the concern is not an absence of failure discussion. Rather, trajectory-level diagnostics are missing for incorrect reflections and cases where proxy scores improve while final outputs deteriorate.
A: We conduct a systematic trajectory-level failure analysis and identify three major failure modes, illustrating how errors emerge and propagate throughout the generation process:
(1) Reflection Hallucination: The UMM may be misled by intermediate images, producing inaccurate reflective targets or hallucinated descriptions, which drive subsequent gradient updates in incorrect directions. As shown in Fig. 8 of the revised manuscript, the UMM mistakenly treats the current image as the desired result and describes a spatial relationship that contradicts the prompt. This incorrect reflection provides little corrective guidance, ultimately leading to failure.
(2) Incomplete Rectification: Even with an accurate reflective target, the limited rectification capacity at each timestep may prevent errors from being fully corrected within the predefined update budget. As shown in Fig. 9 of the revised manuscript, although gradient updates turn most of the carrot blue, the limited rectification capacity leaves part of it orange, resulting in an incorrect final image.
(3) GSS Misranking: When correct and incorrect candidates are visually similar, the UMM may struggle to distinguish them, assigning higher scores to incorrect candidates and selecting suboptimal latent states. As shown in Fig. 10 of the revised manuscript, GSS selects an incorrect candidate despite the availability of a correct one, ultimately producing five bowls instead of the requested number.
We have included this trajectory-level failure analysis, together with the corresponding visualizations, in Appendix G of the revised manuscript.
W1.4: Moreover, BAGEL's position score decreases from 49.0 to 48.5 (Table 1), so the aggregate gain should not be interpreted as improving every compositional capability, much less overall visual quality based on alignment metrics alone.
A: We clarify the scope of our performance claims and provide additional multi-seed results. The original results in the paper use a single random seed 42, where BAGEL shows a slight Position decrease (49.0 to 48.5), while the other three backbones improve. To further assess robustness, we conduct experiments with four random seeds(42, 1234, 2026, 3407). As shown in Table 3, UniRect-CoT consistently improves nearly all compositional categories across the four backbones, with BAGEL's average Position score increasing from 48.4 to 49.5. These results provide stronger evidence of improvements across compositional capabilities beyond the aggregate score. Additionally, Figures 4 and 6 in the original manuscript provide qualitative visual comparisons, further illustrating the visual improvements achieved by UniRect-CoT.
Table 3: Multi-seed evaluation across UMM backbones.
| Method | Single Obj. | Two Obj. | Counting | Colors | Position | Color Attr. | Overall |
|---|---|---|---|---|---|---|---|
| JanusFlow | 94.1±1.2 | 44.0±3.1 | 31.7±3.2 | 75.5±2.0 | 31.4±2.9 | 25.7±1.3 | 50.4±1.9 |
| w/ Ours | 95.3±1.1 | 56.8±3.0 | 41.3±1.6 | 81.4±0.3 | 41.1±2.9 | 33.9±1.3 | 58.3±1.2 |
| OpenUni-RecA | 98.9±0.4 | 91.1±0.9 | 46.7±1.4 | 87.4±0.3 | 40.0±1.6 | 67.9±0.2 | 72.0±0.3 |
| w/ Ours | 99.2±0.3 | 93.4±0.9 | 55.7±1.6 | 89.6±0.3 | 45.0±2.4 | 71.1±0.9 | 75.7±0.7 |
| OmniGen2 | 99.6±0.2 | 93.2±0.5 | 70.7±2.0 | 86.8±0.5 | 53.0±0.9 | 66.0±1.6 | 78.2±0.5 |
| w/ Ours | 99.5±0.2 | 95.1±0.6 | 72.1±0.6 | 87.1±0.5 | 53.9±1.7 | 68.6±2.1 | 79.4±0.3 |
| Bagel | 99.1±0.2 | 95.0±0.8 | 75.5±1.8 | 86.3±1.2 | 48.4±1.0 | 62.9±2.1 | 77.9±0.2 |
| w/ Ours | 99.7±0.3 | 96.3±0.9 | 80.7±2.5 | 87.5±0.5 | 49.5±1.7 | 68.4±0.7 | 80.4±0.4 |
Response to Reviewer Ci3p (Part 3)
W2.1: Table 1 specifies four consecutive samples per prompt from a fixed initial seed, but Tables 1–3 do not report repeated-run variation or paired confidence intervals. The overall advantages over TiR and Best-of-3 are only 0.6 and 0.3 percentage points, respectively (Tables 2 and 6).
A: We provide additional multi-seed evaluations to assess statistical robustness, compare performance and efficiency against TiR and Best-of-3, and clarify our experimental settings for reproducibility. We have included the results in Appendix H and J of the revised manuscript.
Statistical Robustness. To assess statistical robustness, we conduct additional experiments for Tables 1-3 in the original paper using four common random seeds (42, 1234, 2026, and 3407), and report the results as mean ± standard deviation. As shown in Table 4-6, UniRect-CoT statistically improves generation performance across different UMM backbones, outperforms the compared training-free test-time methods, and validates the effectiveness of its key design components.
Comparison with Best-of-3 and TiR. As shown in the Table 5, UniRect-CoT achieves 80.4 ± 0.4, outperforming TiR (79.5 ± 0.5) and Best-of-3 (80.1 ± 1.1). It also exhibits lower variance, indicating more stable performance. Moreover, UniRect-CoT requires only 43.2 seconds per image, compared with 116.9s for TiR and 76.9s for Best-of-3, demonstrating a better balance between generation quality and inference efficiency.
Table 4: Multi-seed evaluation across UMM backbones.
| Method | Single Obj. | Two Obj. | Counting | Colors | Position | Color Attr. | Overall |
|---|---|---|---|---|---|---|---|
| JanusFlow | 94.1±1.2 | 44.0±3.1 | 31.7±3.2 | 75.5±2.0 | 31.4±2.9 | 25.7±1.3 | 50.4±1.9 |
| w/ Ours | 95.3±1.1 | 56.8±3.0 | 41.3±1.6 | 81.4±0.3 | 41.1±2.9 | 33.9±1.3 | 58.3±1.2 |
| OpenUni-RecA | 98.9±0.4 | 91.1±0.9 | 46.7±1.4 | 87.4±0.3 | 40.0±1.6 | 67.9±0.2 | 72.0±0.3 |
| w/ Ours | 99.2±0.3 | 93.4±0.9 | 55.7±1.6 | 89.6±0.3 | 45.0±2.4 | 71.1±0.9 | 75.7±0.7 |
| OmniGen2 | 99.6±0.2 | 93.2±0.5 | 70.7±2.0 | 86.8±0.5 | 53.0±0.9 | 66.0±1.6 | 78.2±0.5 |
| w/ Ours | 99.5±0.2 | 95.1±0.6 | 72.1±0.6 | 87.1±0.5 | 53.9±1.7 | 68.6±2.1 | 79.4±0.3 |
| Bagel | 99.1±0.2 | 95.0±0.8 | 75.5±1.8 | 86.3±1.2 | 48.4±1.0 | 62.9±2.1 | 77.9±0.2 |
| w/ Ours | 99.7±0.3 | 96.3±0.9 | 80.7±2.5 | 87.5±0.5 | 49.5±1.7 | 68.4±0.7 | 80.4±0.4 |
Table 5: Multi-seed comparison with test-time methods.
| Method | Single Obj. | Two Obj. | Counting | Colors | Position | Color Attr. | Overall | Latency (s) |
|---|---|---|---|---|---|---|---|---|
| Baseline | 99.1±0.2 | 95.0±0.8 | 75.5±1.8 | 86.3±1.2 | 48.4±1.0 | 62.9±2.1 | 77.9±0.2 | 24.7 |
| Qwen Rewrite | 99.5±0.5 | 94.7±0.8 | 75.2±2.2 | 87.9±2.1 | 48.1±0.9 | 65.3±1.1 | 78.5±0.3 | 28.7 |
| UiG | 99.4±0.4 | 93.8±1.4 | 76.1±1.6 | 86.7±1.1 | 49.8±2.3 | 62.8±1.0 | 78.1±0.1 | 56.4 |
| TiR | 99.1±0.6 | 94.3±1.3 | 80.2±1.6 | 85.0±1.2 | 52.5±2.3 | 66.1±1.7 | 79.5±0.5 | 116.9 |
| Best-of-3 | 99.4±0.4 | 96.3±0.6 | 80.6±2.3 | 87.4±0.8 | 50.9±2.6 | 66.1±3.5 | 80.1±1.1 | 76.9 |
| UniRect-CoT | 99.7±0.3 | 96.3±0.9 | 80.7±2.5 | 87.5±0.5 | 49.5±1.7 | 68.4±0.7 | 80.4±0.4 | 43.2 |
Table 6: Multi-seed ablation of semantic target design and GSS.
| Setting | Update | ITE Target | GSS Target | Single Obj. | Two Obj. | Counting | Colors | Position | Color Attri. | Overall ↑ |
|---|---|---|---|---|---|---|---|---|---|---|
| Baseline | – | – | – | 99.1±0.2 | 95.0±0.8 | 75.5±1.8 | 86.3±1.2 | 48.4±1.0 | 62.9±2.1 | 77.9±0.2 |
| ITE Only | Static | User | – | 99.4±0.3 | 95.0±1.2 | 77.1±2.6 | 86.9±0.7 | 48.1±1.2 | 66.2±0.8 | 78.8±0.3 |
| Static | Exp. | – | 99.6±0.5 | 93.6±0.5 | 76.9±1.6 | 86.3±0.8 | 47.9±2.6 | 64.2±3.2 | 78.1±0.3 | |
| Static | Ref. | – | 99.7±0.4 | 93.8±2.1 | 78.9±2.4 | 87.9±1.5 | 48.7±4.4 | 65.4±2.8 | 79.1±0.2 | |
| Dynamic | Ref. | – | 99.7±0.4 | 94.5±1.6 | 78.5±1.9 | 89.2±1.2 | 48.3±2.6 | 66.9±2.9 | 79.5±0.3 | |
| ITE+GSS | Static | User | User | 99.2±0.4 | 95.0±0.8 | 78.3±1.9 | 87.5±1.0 | 49.6±1.6 | 65.7±1.3 | 79.2±0.2 |
| Static | Exp. | User | 99.5±0.2 | 94.2±0.7 | 77.7±2.0 | 86.8±0.8 | 48.9±2.0 | 63.7±2.8 | 78.4±0.3 | |
| Static | Ref. | User | 99.7±0.3 | 94.8±1.8 | 77.4±1.7 | 87.0±0.7 | 49.8±3.8 | 67.9±1.9 | 79.5±0.5 | |
| Dynamic | Ref. | Ref. | 99.4±0.4 | 94.3±2.1 | 78.4±2.5 | 87.4±1.8 | 49.9±3.3 | 64.6±2.7 | 79.0±0.5 | |
| Dynamic | Ref. | User | 99.7±0.3 | 96.3±0.9 | 80.7±2.5 | 87.5±0.5 | 49.5±1.7 | 68.4±0.7 | 80.4±0.4 |
W2.2: The paper sweeps update counts and intervention windows, but does not explain whether the preferred settings were selected using independent validation prompts; this is a clarification request, not evidence of test-set leakage.
A: We provide the details of the reproduction below. All four hyperparameters are shared across the four UMM backbones without model-specific tuning. Specifically, we use η = 200, δ = 0.001, W = [5, 10], and K = 3 for Janus-Flow, OpenUni-RecA, OmniGen2, and BAGEL. For each backbone, we follow its official inference configuration. This demonstrates the generalizability of our method. We have added these implementation details to Appendix J of the revised manuscript to facilitate reproducibility.
Response to Reviewer Ci3p (Part 1)
We sincerely thank reviewer Ci3p for the thoughtful and constructive feedback, which will help improve our work. Our responses are as follows.
A: We provide additional quantitative and qualitative analyses to further examine the reliability of our proxy guidance and clarify its limitations.
Intermediate Judgment Reliability for Figure 3: We collect samples using GenEval prompts with different random seeds across the four UMMs. To avoid class imbalance, we construct a balanced evaluation set by selecting 500 samples that are correct across all four UMMs and 500 samples that are incorrect across all four UMMs, resulting in 1,000 samples with a 1:1 correct-to-incorrect ratio. At each timestep, we ask each UMM to predict whether the final generated image will satisfy the prompt, using its final GenEval flag as the ground truth at every timestep. We further report the confusion matrices across denoising intervals in Table 1. The accuracy increases from 34.1% in steps [0,5) to 73.3% in [5,10), showing that informative judgments emerge well before generation is complete.
Table 1: Confusion matrix and judgment accuracy across different denoising timesteps.
A: To examine whether Eq. (11) predicts improvements in the final image, we conduct the following experiment. At each timestep within the rectification window, we independently continue denoising from each candidate state to obtain its final image. We then use two signals to evaluate each candidate: (1) its Eq. (11) likelihood score, computed on the estimated clean image at that timestep; and (2) the GenEval correctness of its final image, which serves as the ground-truth label. This allows us to assess whether Eq. (11) reliably ranks intermediate candidates according to their eventual generation correctness. As shown in Tab. 2, candidates that eventually produce correct images consistently receive higher likelihood scores than those producing incorrect images across all evaluated timesteps, with average scores of -3.01 and -3.53, respectively (a mean gap of 0.52). These results demonstrate that Eq. (11) assigns higher scores to candidates leading to correct final images, supporting its reliability for ranking intermediate candidates during rectification.
Table 2: Candidate likelihood scores by final correctness within the rectification window.