- Rating
- 3(+1)
- Confidence
- 1(−2)
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes DARC, a training-free framework for accelerating world models (video generation) through steps caching. DARC combines state-conditioned output extrapolation, online calibration, and an accumulated reanchoring strategy. Experiments across four world-model families demonstrate substantial inference speedups while maintaining competitive generation quality. Speedup is measured on five GPU platforms, further enhancing their result.
Critical strengths
-
Its observation that caching errors affect subsequent generation states, making both prediction accuracy and refresh timing important, is well motivated.
-
The paper includes component-wise ablations for baseline comparisons, sensitivity studies, and energy measurements.
Critical weaknesses
-
The speed-quality advantage is less conclusive under matched computational budgets. More experiments are needed to justify this.
-
DARC uses benchmark-specific values for hyperparameters, whereas competing methods are primarily evaluated using their default parameters. This might lead to unfair comparisons with baselines.
Questions for authors
Please refer to the weakness section.
5 comments
Response to Reviewer Z25m (2/2)
W2. Benchmark-specific hyperparameters.
We thank the reviewer for raising this point and would like to separate the two settings clearly. In the submitted Tables 1 and 2, DARC uses the budget ε and span cap m of Table 4, chosen per setting from small sweeps, while all other constants are shared. For all rebuttal experiments, we fixed one configuration for every model and benchmark (span cap m = 10, unit weights, = 1) and set only ε per benchmark, so that DARC runs at its Table 1 speed. Every baseline in Table R1 receives the same treatment: its threshold is set per benchmark to reach that speed, and where the threshold alone could not reach the target speed, one secondary knob was also changed (revised Appendix B). DARC therefore needs no more per-benchmark tuning than any baseline. Under this configuration, DARC remains the closest to the unaccelerated model on all seven benchmarks (Table R1). Table R2 further shows that the span cap and need no tuning, and that the weights only shift the operating point in the same direction as ε. The per-benchmark choice therefore reduces to a single number that sets the speed target, as a threshold does for any cache.
Table R2. Robustness of DARC to its hyperparameters, one parameter changed at a time. All rebuttal experiments use a single configuration shared across model families (m = 10, all weights 1, = 1); only the budget ε differs between benchmarks, and it plays the same role as the threshold of a baseline cache by selecting the operating point on the speed-quality trade-off. Changing the span cap or the weights over a wide range shifts the operating point smoothly and never causes a failure, so DARC's results do not depend on fine-tuning these parameters. ε is given as a multiple of each model's default. Calls: share of native model calls. Speed: relative to the unaccelerated model. PSNR: paired to the unaccelerated model (dB). Each cell: Calls ↓ / Speed ↑ / PSNR ↑.
| Parameter | Value | SANA-WM (VBench-I2V) | MiniWorld (DROID) | Astra (Sekai) |
|---|---|---|---|---|
| ε | ×1/8 | 68% / 1.48× / 36.31 | 65% / 1.54× / 44.18 | 74% / 1.31× / 42.75 |
| ×1/4 | 47% / 2.03× / 34.65 | 62% / 1.63× / 43.03 | 57% / 1.60× / 35.38 | |
| ×1 (default) | 24% / 3.30× / 29.73 | 43% / 2.25× / 39.37 | 33% / 2.43× / 33.97 | |
| ×4 | 16% / 4.33× / 27.07 | 37% / 2.47× / 39.17 | 19% / 3.46× / 28.35 | |
| ×16 | 15% / 4.39× / 26.21 | 35% / 2.56× / 38.69 | 16% / 3.85× / 26.54 | |
| m | 3 | 39% / 2.31× / 30.45 | 60% / 1.64× / 42.83 | 41% / 2.09× / 35.33 |
| 5 | 28% / 2.95× / 30.07 | 43% / 2.25× / 39.37 | 34% / 2.42× / 34.13 | |
| 10 (default) | 24% / 3.30× / 29.73 | 43% / 2.25× / 39.37 | 33% / 2.43× / 33.97 | |
| 20 | 24% / 3.32× / 29.75 | 43% / 2.25× / 39.37 | 33% / 2.43× / 33.97 | |
| none | 24% / 3.39× / 29.75 | 43% / 2.25× / 39.37 | 33% / 2.42× / 33.97 | |
| 0 | 24% / 3.44× / 29.73 | 43% / 2.25× / 39.37 | 29% / 2.62× / 33.16 | |
| 0.25 | 24% / 3.42× / 29.73 | 43% / 2.25× / 39.37 | 29% / 2.66× / 33.15 | |
| 1 (default) | 24% / 3.30× / 29.73 | 43% / 2.25× / 39.37 | 33% / 2.43× / 33.97 | |
| 4 | 24% / 3.30× / 29.73 | 44% / 2.23× / 39.44 | 48% / 1.84× / 34.87 | |
| 16 | 24% / 3.42× / 30.61 | 57% / 1.76× / 42.57 | 60% / 1.56× / 35.12 | |
| 0 | 21% / 3.52× / 28.95 | 42% / 2.28× / 39.49 | 32% / 2.53× / 32.39 | |
| 0.25 | 22% / 3.47× / 28.94 | 43% / 2.27× / 39.51 | 33% / 2.48× / 31.89 | |
| 1 (default) | 24% / 3.30× / 29.73 | 43% / 2.25× / 39.37 | 33% / 2.43× / 33.97 | |
| 4 | 30% / 2.86× / 31.76 | 51% / 1.93× / 40.59 | 37% / 2.24× / 33.99 | |
| 16 | 46% / 2.12× / 34.87 | 62% / 1.62× / 43.52 | 47% / 1.87× / 35.10 | |
| 0 | 23% / 3.38× / 29.69 | 43% / 2.26× / 39.42 | 32% / 2.48× / 33.46 | |
| 0.25 | 23% / 3.39× / 29.70 | 43% / 2.26× / 39.42 | 33% / 2.47× / 33.52 | |
| 1 (default) | 24% / 3.30× / 29.73 | 43% / 2.25× / 39.37 | 33% / 2.43× / 33.97 | |
| 4 | 25% / 3.27× / 30.79 | 44% / 2.22× / 39.56 | 36% / 2.29× / 34.59 | |
| 16 | 29% / 2.97× / 30.38 | 53% / 1.88× / 40.47 | 43% / 2.03× / 34.65 | |
| 0.25 | 24% / 3.36× / 29.73 | 43% / 2.25× / 39.37 | 33% / 2.43× / 33.97 | |
| 0.5 | 24% / 3.35× / 29.73 | 43% / 2.25× / 39.37 | 33% / 2.42× / 33.97 | |
| 1 (default) | 24% / 3.30× / 29.73 | 43% / 2.25× / 39.37 | 33% / 2.43× / 33.97 | |
| none | 24% / 3.48× / 29.73 | 43% / 2.25× / 39.37 | 33% / 2.43× / 33.97 |
Follow-up
Dear Reviewer,
We really thank you again for your detailed review of our paper. We know this is a busy period. We have posted a detailed response addressing each of your comments and the requested experiments, and we would be grateful if you could please take a look when convenient. If you feel your concerns have been addressed, we would sincerely appreciate your considering an updated evaluation. If anything remains unconvincing, we would be glad to discuss it together. We are following this thread closely and will respond promptly. We would greatly welcome any additional feedback you might have.
Response to Reviewer Z25m (1/2)
We thank Reviewer Z25m for the review. We are glad the reviewer found our core motivation convincing, and we appreciate the recognition of our component-wise ablations, sensitivity studies and energy measurements, and of evaluating speed across five GPU platforms. The reviewer's two concerns, about matched compute and about how hyperparameters were set, are addressed below with additional experiments.
W1. Speed-quality advantage under matched computational budgets.
We thank the reviewer for this comment and agree that a comparison at matched speed is essential. In the paper, the speed-matched analysis of Appendix D was an ablation on a small subset of SANA-Bench. To address this concern thoroughly, we re-ran all baselines and DARC under a common protocol on the full benchmarks of all four model families (Table R1). For each benchmark, every baseline, including the recent RACER, is tuned to run at DARC's speed: its speed knob is set on calibration samples disjoint from the evaluation set, by measured speed alone and never by quality. Quality is measured as paired PSNR, SSIM and LPIPS against the unaccelerated model's output for the same sample and seed, which isolates the error introduced by caching.
Under this protocol, DARC achieves the best PSNR, SSIM and LPIPS on all seven benchmarks. Since every method runs at the same speed, these gains come from how DARC chooses which calls to skip and how it predicts their outputs, not from a different operating point. Two further experiments, reported in our responses to the other reviewers, support this. At WorldDynCache's operating point, with the same share of model calls and the same speedup, DARC is closer to the unaccelerated model on Aether-5B (response to Reviewer BmSZ, W6, Table R5). In long autoregressive rollouts of up to 1,021 frames, it remains the closest to the unaccelerated model at every stage of the video (response to Reviewer CEXN, Q2 and Q6, Tables R4 and R5). We have added Table R1 to the revised paper and based the claims of Sec. 4.1 on this matched-speed comparison.
Table R1. Comparison at matched speed, laid out as Table 1. Every baseline is tuned separately for each benchmark to run at DARC's speed (target next to each benchmark name). Its settings are chosen by measured speed alone, on calibration samples disjoint from the evaluation set (revised Appendix B). Unlike Table 1, which compares PSNR/SSIM/LPIPS against the real video, this table compares them against the unaccelerated model's video for the same sample and seed, which isolates the error introduced by the cache. Bold: best; italics: second best among accelerated methods.