- Rating
- 3(+1)
- Confidence
- 2(−1)
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2(+1)
Read review
Paper summary
The paper tries to relate intensity saturation in diffusion-generated images to large coefficients along leading covariance eigenvectors. It measures initial-to-final coefficient linearity, and uses an existing Gaussian CFG analysis to interpret guidance-induced changes. It then proposes magnitude-dependent damping of leading covariance coefficients during sampling.
Critical strengths
-
The paper goes beyond qualitative examples; Table 7 reports mean initial-to-final for the leading eigenvector, and removing that initial-noise component reduces one-sided saturated area by 65.1% (Fig. 5). The distinction between one-sided and two-sided saturation connects observable image behavior to different covariance components.
-
It introduces no additional denoiser evaluations and consistently reduces ECA@10 while increasing HPSv2.1 in Table 2. Appendix R also evaluates matched recall, with improvements across the reported metrics for EDM2-S/CFG over . These results support a practical benefit beyond the fixed-guidance comparisons alone.
Critical weaknesses
-
Section 3 introduces training-data statistics, but Appendix A estimates them from 50,000 model-generated images per class. The final-linearity experiment in Appendix B.6 uses seeds 0–9,999, which overlap that estimation set. This does not make the observed relationship automatic, but leaves its robustness to actual training-data statistics and independent evaluation insufficiently established. Appendix B.2 does use separate seeds, which partially addresses this concern. Separately, Appendix B.4 does not specify how the held-out eigenvector’s transition time is obtained. If it is extracted using Eq. (18), it depends on the final being predicted. This uncertainty matters for interpreting Table 5 as independent endpoint prediction.
-
The Gaussian trajectory and CFG decomposition follow Wang and Vastola (2024) and Li et al. (2025), respectively. The incremental contribution therefore depends substantially on the saturation analysis. However, Table 2 omits frequency-decoupled guidance (Sadat et al., 2025), a closely related approach to controlling oversaturation, and there is no comparison with the same damping rule applied to a fixed low-frequency basis.
-
Section 4.2. An opposite-sign additive shift need not decrease coefficient magnitude because it can overshoot through zero. Writing Eq. (7) as , the additive shift decreases magnitude precisely when . For example, and increase the magnitude from to . Please qualify the corresponding bullet.
-
Appendix H/Fig. 22 versus Table 11 : Appendix H includes and reports one-sided contributions of 47.3–53.1%. Table 11 gives an unguided sample-mean contribution of 41.41%, outside this range, whereas its pooled value of 50.67% falls inside.
[1] Wang and Vastola (2024). The Unreasonable Effectiveness of Gaussian Score Approximation for Diffusion Models and its Applications. TMLR. Relevant locations: Eqs. (13), (15), and (17); Section 4.3 on prediction of coarse image structure. https://arxiv.org/abs/2412.09726
[2] Li, Wang, and Qu (2025). Towards Understanding the Mechanisms of Classifier-Free Guidance. NeurIPS. Relevant locations: Section 3.2, Theorem 1; Section 4.1 on the effects of mean-shift and covariance guidance. https://arxiv.org/abs/2505.19210
[3] Sadat, Vontobel, Salehi, and Weber (2025). Guidance in the Frequency Domain Enables High-Fidelity Sampling at Low CFG Scales. Relevant locations: Section 4, Eqs. (3)–(6); Section 5 on EDM2 and DiT experiments. https://arxiv.org/abs/2506.19713
Questions for authors
-
Please repeat the principal coefficient-linearity and saturation-prediction results using covariance estimated from actual ImageNet images, with evaluation seeds disjoint from any generated-sample calibration set. How do the leading eigenspaces, ?
-
For Tables 5–6, is the held-out transition time predicted from its eigenvalue and the training modes, or measured from its own trajectory? Equation (18) uses the final through . Please specify the transition estimation and factorization fitting procedures, and report endpoint prediction.
-
On EDM2-S, please compare the proposed method with frequency-decoupled guidance and with the same damping rule applied to a fixed, rank-matched low-frequency basis. Use identical seeds and sampling budgets, sweep regulation strength, and report ECA@10, HPSv2.1, FID, .
8 comments
W2. Does frequency-decoupled guidance’s control of oversaturation diminish our contribution?
Our contribution concerns intensity saturation, which differs from the oversaturation addressed by frequency-decoupled guidance (FDG) [1] and related methods [2]. The metric in [2] measures oversaturation through the difference between the maximum and minimum RGB values relative to the maximum, whereas our intensity saturation concerns near-white or near-black pixels. Neither measure reliably captures the phenomenon measured by the other.
Moreover, [1,2] focus on oversaturation induced by strong guidance, whereas intensity saturation occurs even without guidance. The two phenomena also differ in how they appear across samples.
The methods also differ in their sample-wise selectivity. FDG modifies guidance strengths for low- and high-frequency components across samples, whereas our method selectively regulates samples prone to excessive intensity saturation, with weaker corrections to other samples. This sample-wise adjustment aims to reduce excessive intensity saturation while preserving other samples and the performance of the underlying guidance method as much as possible. Our method can also be combined with multiple guidance methods to control intensity saturation while preserving or even improving their image-quality benefits.
Thus, FDG’s control of oversaturation does not diminish our contribution.
W2, Q3. Comparisons with FDG and a fixed, rank-matched low-frequency basis.
Using Table 2’s EDM2-S setup with 50k samples per method, we compared our method with frequency-decoupled guidance. For FDG, we set .
| Method | ECA@10 ↓ | HPSv2.1 ↑ | FID ↓ | FD_DINOv2 ↓ |
|---|---|---|---|---|
| CFG | 8.83 | 24.74 | 6.68 | 94.53 |
| CFG + ours | 6.26 | 24.94 | 6.76 | 92.29 |
| FDG | 3.67 | 23.99 | 3.66 | 95.89 |
| FDG + ours | 2.07 | 24.14 | 3.87 | 95.10 |
Thus, even when FDG is chosen as the underlying guidance method, our correction provides an additional benefit by reducing the intensity saturation that remains. FDG’s ability to reduce intensity saturation therefore does not eliminate the contribution of our complementary correction.
Fixed, rank-matched low-frequency basis. We use the same CFG setup and App. O damping, with variances estimated from the same data. Using normalized Hadamard channel vectors, pairs DCT with the first vector; pairs DCT with all four vectors and DCT with the first. All metrics use 10k matched samples. Each cell reports Ours / DCT.
| ECA@10 ↓ | HPSv2.1 ↑ | FID ↓ | FD-DINOv2 ↓ | |
|---|---|---|---|---|
| 0.5 | 4.50 / 9.27 | 24.80 / 24.75 | 8.59 / 8.69 | 115.04 / 115.37 |
| 1 | 5.35 / 9.23 | 24.78 / 24.76 | 8.60 / 8.67 | 114.98 / 115.10 |
| 2 | 6.80 / 9.00 | 24.78 / 24.76 | 8.68 / 8.66 | 115.07 / 115.47 |
| ECA@10 ↓ | HPSv2.1 ↑ | FID ↓ | FD-DINOv2 ↓ | |
|---|---|---|---|---|
| 0.5 | 3.17 / 6.29 | 25.07 / 24.99 | 8.86 / 8.68 | 114.03 / 113.16 |
| 1 | 4.30 / 6.79 | 25.02 / 24.96 | 8.77 / 8.67 | 113.01 / 112.93 |
| 2 | 6.22 / 7.57 | 24.94 / 24.91 | 8.75 / 8.66 | 113.46 / 113.27 |
Under the same damping rule, at matched rank and , ours consistently achieves lower ECA@10 than the fixed low-frequency basis. While damping low frequencies can also reduce intensity saturation, frequency alone offers no clear rule for selecting an effective basis for a given diffusion model, requiring repeated sampling and evaluation for different candidate bases. In contrast, our approach does not require this repeated evaluation, as it directly derives an appropriate basis for selectively reducing intensity saturation from the model’s sample covariance, as described in the paper.
W2. Fig. 22/Table 11.
Table 11 reports both sample-mean and pooled contributions (App. I, Eqs. (38)–(39)), whereas Fig. 22(b) shows only pooled results. Both identify as the largest contributor among the leading eigenvectors for one-sided saturation. The comparison below also includes the results for Fig. 22(b) using sample-mean.
| Aggregation | Fig. 22 | Table 11 |
|---|---|---|
| Sample mean | 30.9–41.4% | 41.41% |
| Pooled | 47.3–53.1% | 50.67% |
At , the sample mean of 41.4% matches Table 11's 41.41%. We will clarify this and add the sample mean results.
[1] Sadat et al. (2025). Guidance in the Frequency Domain Enables High-Fidelity Sampling at Low CFG Scales.
[2] Sadat et al. (2025). Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models.
Thank you for the response and the additional experiments. Some of my earlier concerns (in Questions and Weakness section) have been addressed, and I appreciate the authors' efforts.
However, I am still not satisfied with the author's response on W2 (incremental contribution + dependence on the saturation analysis). Therefore, I would like to keep my original rating for now.
Thank you for the detailed review.
Our additional experiments sufficiently establish the robustness of the Section 3 results to actual training-data statistics and independent evaluation. We repeated the principal coefficient-linearity and saturation-prediction experiments using covariance from real ImageNet images and evaluation seeds (50,000–59,999) disjoint from all calibration seeds.
The generated-statistics baseline reproduces the paper’s results. With real statistics, coefficient linearity remains high and saturation remains partially predictable, as shown below. The top-5 eigenspaces also nearly coincide (Ec[Ei[∥Qreal,c⊤qgen,c,i∥22]]=0.97).
It is not necessary to interpret Table 5 as independent endpoint prediction, because its purpose is to validate whether ℓc and kc are shared across eigenvectors within each class, conditional on their observed transition times. Additional experiments show that the results remain highly consistent whether or not transition-time estimation uses the held-out eigenvector’s final R2.
Eq. (25) describes each eigenvector’s decorrelation dynamics using the same two class-specific functions, ℓc and kc, with an eigenvector-specific transition time. In Table 5, we fit these functions using nine eigenvectors, then predict the remaining eigenvector’s trajectory and endpoint given its observed transition time. Thus, held-out refers to excluding the target eigenvector from fitting ℓc and kc.
The original transition time was measured from the target’s own trajectory as H50, the interpolated 50% crossing of endpoint-normalized cumulative-max H. This calculation uses endpoint information, which is part of the conditioning in this experiment. Our description should have explicitly stated how transition time was obtained and what was held out.
To additionally address the concern about endpoint-dependent timing, we repeated the evaluation using eigenvalue-based transition times, without using the target’s trajectory to estimate its timing. The procedures are described below.
The results are shown below, with changes relative to the original results (new − original).
We will clarify the purpose, transition-time calculation, and conditioning in Table 5’s caption and App. B.4, and provide fitting details in App. B.4–B.5.
An opposite-sign additive shift decreases coefficient magnitude in almost all of evaluated opposite-sign cases with reduced intensity saturation under CFG, at each evaluated CFG weight.
The statement “decreasing it for those in the opposite direction” was intended as a qualitative explanation of how mean shift contributes to the observed reductions in intensity saturation, rather than a guarantee for every initial projection and CFG weight.
To assess this interpretation, we evaluated each CFG weight separately using the results in Figures 6 and 27. Among samples whose contribution to ECA@10 decreases and whose additive shift has the opposite sign to the variance-scaled coefficient, coefficient magnitude decreases in 98.8%, 98.0%, 97.6%, and 97.0% of cases, respectively.
Although an opposite-sign additive shift can increase coefficient magnitude through overshoot, this rarely occurs in the evaluated cases where saturation decreases. We will revise the wording to make clearer that the bullet provides a qualitative explanation of the mean shift mechanism.