- Rating
- 3(+1)
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes SnapCache, a training-free step-skipping framework for flow-matching inference. It reuses cached velocities over skipped sampling steps, estimates the accumulated error from the residual between cached and freshly computed velocities, and applies a minimum-norm correction called Optimal Brain Trajectory Compensation (OBTC). A causal scheduler selects the longest skip interval whose predicted error satisfies a solver-based error budget. Experiments are conducted on FLUX.1-dev, Krea-2-Raw, Wan2.2, and LingBot-VA.
Critical strengths
- The paper formulates step skipping as a solver-level error compensation problem rather than merely reusing stale velocities. The separation between solver-output perturbations and solver-specific propagation sensitivities provides a coherent engineering framework for implementing compensation in Euler and UniPC samplers.
- The evaluation spans image generation, video generation, and robot action generation, and includes ablations of compensation and scheduling. The cross-modal scope is useful for assessing whether the proposed controller transfers beyond a single image-generation backbone.
Critical weaknesses
- Unsupported linear-error assumption. SnapCache assumes that the hidden velocity error during a skipped interval changes linearly from zero to the residual observed at the next retained step. The paper neither explains why this assumption should hold for learned velocity fields nor validates it empirically across models and skip lengths.
- Limited technical novelty. The proposed method mainly combines cached-velocity reuse, residual correction, solver-sensitivity propagation, and adaptive skip scheduling. The paper does not clearly identify a new underlying principle beyond this combination, and the OBTC formulation currently appears closer to an engineering heuristic than a substantially new technical contribution.
- Incomplete experimental comparison. The paper does not compare SnapCache with several strong and relevant training-free acceleration methods, including VDE [1], LeMiCa [2], and MeanCache [3]. These comparisons are needed to assess the practical advantage of SnapCache.
Questions for authors
- What empirical or theoretical evidence supports the linear residual assumption? For several models and skip lengths, can the authors compare the actual hidden velocity-error trajectory with the proposed linear approximation and report the approximation error?
- Can the authors add comparisons with VDE, LeMiCa, and MeanCache on applicable common tasks and models using the same experimental protocol? Please also clarify how the gate tolerances are selected for each model and whether model-specific tuning is required.
[1] VDE: Training-Free Accelerating Rectified Flow Model via Velocity Decomposition and Estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 37918–37928. [2] Gao, Huanlin, Ping Chen, Fuyuan Shi, Chao Tan, Zhaoxiang Liu, Fang Zhao, Kai Wang, and Shiguo Lian. 2025. LeMiCa: Lexicographic Minimax Path Caching for Efficient Diffusion-Based Video Generation. In Advances in Neural Information Processing Systems, volume 38 (Main Conference), 132388–132415. [3] Gao, Huanlin, Ping Chen, Fuyuan Shi, Ruijia Wu, Li YanTao, Qiang Hui, Youyuren, Ting Lu, Chao Tan, Shaoan Zhao, Zhaoxiang Liu, Fang Zhao, Kai Wang, and Shiguo Lian. 2026. MeanCache: From Instantaneous to Average Velocity for Accelerating Flow Matching Inference. In International Conference on Learning Representations (ICLR), volume 2026, 60466–60485.
6 comments
Technical contribution and comparison with prior methods
Thank you for asking us to clarify SnapCache's technical contribution and compare it with closely related methods. SnapCache uses the residual revealed at the next scheduled model evaluation to correct error accumulated during the preceding skip interval, without an extra network evaluation. Eq. (4) maps this residual through precomputable solver sensitivities to estimate the direct state error. Eq. (6) selects the minimum-norm correction whose first-order effect on the solver matches that estimate. Under the stated rank assumption, this correction is unique in the row space of (Appendix C). Appendix B specifies how the correction updates both the model state and UniPC's active history.
In concrete terms, prior training-free caching methods focus on the skipping decision itself and on predicting the velocity over the skipped interval: TeaCache, MagCache, and LeMiCa determine when to reuse cached outputs, while VDE and MeanCache estimate the skipped velocity from previously available information. ERTACache additionally rectifies cache-induced error, but its rectification is calibrated offline and applied at reuse time. In all of these methods, any correction is determined before or during the skipped interval. SnapCache addresses a complementary problem: it uses the residual revealed at the next scheduled refresh to correct, without offline calibration, the error that has already accumulated over the skipped interval, which allows the correction to be formulated as a minimum-norm compensation problem. We note that other reviewers also highlighted this aspect: Reviewer nCrB described the formulation as a "principled minimum-norm correction rather than heuristic caching," and Reviewer HPEQ identified "retrospective correction of accumulated sampling error" as its strongest idea. We have revised the introduction and Section 3 to state this contribution more explicitly.
We compared SnapCache with VDE, LeMiCa, and MeanCache on the same 553 FLUX prompts. We used the public FLUX implementations of VDE and LeMiCa with matched model, seeds, resolution, guidance, and sampling grid. We ported MeanCache's released JVP computation to FLUX and used its released schedule (Appendix H). NFE counts full-stack evaluations. LeMiCa performs 10 full-stack evaluations across 50 partial or full forward passes.
At comparable full-evaluation budgets, SnapCache improves all four reference-fidelity metrics against each baseline. Against VDE, LPIPS decreases from 0.0924 to 0.0562 at 21.7161 versus 22 evaluations. Against LeMiCa and MeanCache, LPIPS decreases from 0.2944 and 0.3015, respectively, to 0.2591 at 9.2007 versus 10 evaluations.
Table R6. FLUX comparison with VDE.
| Method | NFE | LPIPS ↓ | PSNR ↑ | SSIM ↑ | Ref-FID ↓ |
|---|---|---|---|---|---|
| VDE | 22.0000 | 0.0924 ± 0.0053 | 26.222 ± 0.397 | 0.8924 ± 0.0056 | 14.59 |
| SnapCache () | 21.7161 | 0.0562 ± 0.0047 | 29.571 ± 0.495 | 0.9336 ± 0.0049 | 8.68 |
Table R7. FLUX comparison with LeMiCa and MeanCache.
| Method | NFE | LPIPS ↓ | PSNR ↑ | SSIM ↑ | Ref-FID ↓ |
|---|---|---|---|---|---|
| LeMiCa | 10.0000 | 0.2944 ± 0.0103 | 19.371 ± 0.273 | 0.7436 ± 0.0093 | 32.82 |
| MeanCache () | 10.0000 | 0.3015 ± 0.0102 | 19.418 ± 0.270 | 0.7401 ± 0.0092 | 35.35 |
| SnapCache () | 9.2007 | 0.2591 ± 0.0102 | 20.510 ± 0.338 | 0.7636 ± 0.0092 | 29.78 |
| SnapCache () | 11.2875 | 0.1995 ± 0.0086 | 22.713 ± 0.372 | 0.8087 ± 0.0082 | 24.53 |
LPIPS, PSNR, and SSIM are reported as mean ± 95% CI half-width over 553 prompts. Ref-FID is computed over the full set. Bold values in Table R7 highlight the comparison near 10 NFE.
Table R4 presents a separate experiment on a single prompt. For each method, we report the mean latency per NFE across three timed runs following 50 warmup runs. All four baselines were within 1% of SnapCache.
Table R4. FLUX latency per full model evaluation on A100.
| Method | Latency per NFE (ms) ↓ | Relative to SnapCache |
|---|---|---|
| MagCache | 559.01 | +0.08% |
| SnapCache | 558.56 | Reference |
| VDE | 564.05 | +0.98% |
| LeMiCa | 563.95 | +0.96% |
| MeanCache | 558.31 | −0.04% |
Tolerance selection and model-specific tuning
Thank you for asking how to choose the tolerance and whether SnapCache requires model-specific tuning. SnapCache requires no model-specific calibration or retraining. The user selects to trade off fidelity against the number of model evaluations. Appendix E explains how we selected FLUX operating points near the target compression ratios, and Appendix G lists the settings used in our experiments.
As discussed in our response to Reviewer nCrB, one example of how SnapCache could be deployed in practice is to expose as a sampler parameter in ComfyUI. Users could inspect intermediate results with ComfyUI's latent previews or assess final images with a reference-free metric such as ImageReward. They could then adjust between runs based on this feedback and their NFE budget.
Why linear interpolation, and its measured accuracy
Thank you for highlighting the need to justify the linear reconstruction and validate its empirical accuracy. The ramp is the first-order interpolation determined by the two available anchors in Eq. (3), δ(ts)=0 and δ(tr)=ν. It uses physical time, ρ(t)=(t−ts)/(tr−ts), and Eq. (5) accounts for the reconstruction remainder under a twice-differentiable residual trajectory. The bound quantifies the curvature remainder of this first-order reconstruction.
The added diagnostics measure its empirical accuracy across FLUX and Wan. On 553 FLUX prompts, residual-interpolation relative L2 error is 0.5348/0.6160 and cosine similarity is 0.8379/0.7694 at κ=2 and 8, respectively. At κ=8, this residual metric is 0.5216 for one skipped step (229 prompts) and 0.6550 for nine (360 prompts). The solver-propagated direct-error diagnostic shown in the FLUX figure gives relative L2 errors of 0.4548/0.4942 at κ=2 and 8. On 100 Wan prompts, branchwise residual-interpolation relative L2 error is 0.4872/0.4862 at κ=2.0, 0.4839/0.4843 at κ=2.5, and 0.5086/0.5130 at κ=7. Cosine similarity ranges from 0.8821 to 0.8899. The Wan diagnostic interpolates in physical time and covers skip lengths 1–9.
The plotted FLUX direct-error and Wan residual-interpolation relative errors increase moderately without a sharp jump across scheduler-selected skip intervals within each κ setting.