See Further When Clear: Curriculum Consistency Model

Boxiao Liu; Guanglu Song; Haihang You; Xingzhong Hou; Yi Zhang; Yu Liu; Yunpeng Liu

REVIEW 4 major objections 6 minor 1 cited by

See Further When Clear: Curriculum Consistency Model

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that consistency distillation underperforms because learning difficulty is uneven across noise levels, and that pacing the teacher's forward jump with a PSNR-based curriculum fixes it, yielding one-step generation that…

desk verdict A plausible adaptive-schedule idea for consistency distillation, but the headline FID gains are not yet shown to come from the schedule rather than the GAN loss. read the letter →

arxiv 2412.06295 v1 pith:DB44CZEL submitted 2024-12-09 cs.CV

Yunpeng Liu , Boxiao Liu , Yi Zhang , Xingzhong Hou , Guanglu Song , Yu Liu , Haihang You This is my paper · ORCID

classification cs.CV

keywords consistencydistillationcurriculumlearningdiffusionmodelsflowmatchingfew-stepsamplingPSNRknowledgediscrepancytext-to-imagegeneration

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Consistency distillation trains a student to jump from a noisy input at timestep t to the teacher's output at a later timestep u, but this paper shows the difficulty of that jump is wildly uneven: near pure noise the student and teacher disagree a lot, while near the clean image they agree so much that the student learns little. The authors claim this imbalance, not just discretization error, is why consistency models underperform, and they fix it with a PSNR-based curriculum metric (KDC) that measures the student–teacher gap at each timestep. Whenever the gap is too small, the teacher iterates forward in small steps, extending u until the KDC clears a fixed threshold — "see further when clear." If the claim holds, a single function evaluation can match or beat multi-step baselines: FID 1.64 on CIFAR-10 and 2.18 on ImageNet 64x64, with the same recipe transferring to SDXL and SD3 text-to-image models.

What carries the argument

The load-bearing object is the Knowledge Discrepancy of the Curriculum (KDC), defined as $\mathrm{KDC}_u^t = 100 - \mathrm{PSNR}(x_{\mathrm{est}}, x_{\mathrm{target}})$, where $x_{\mathrm{est}}$ is the student's one-step prediction and $x_{\mathrm{target}}$ is the teacher's target after advancing from $t$ to $u$. CCM wraps KDC in a repeat-until loop (Algo. 1): advance the teacher by a small base step $s$, recompute the target, and keep advancing only while the discrepancy stays below the threshold $T_{\mathrm{KDC}}$; when it clears the threshold, use that target in the consistency loss $\mathcal{L}_{\mathrm{CCM}}$. This adaptive-loop machinery converts the curriculum-difficulty problem into a scheduling problem, and the multi-step iteration is what keeps the teacher's target accurate even when the selected interval $u-t$ is large.

What would settle it

Take a trained CCM, record the pairs $(t, u)$ chosen by the KDC loop, and compare the teacher's multi-step solution $x_u$ against a high-accuracy reference solver with hundreds of steps at the same $(t, u)$. If the error grows sharply as $u-t$ increases, the large-step targets that CCM relies on are not accurate, and the reported gains would come from something other than faithful teacher targets. A second check: corrupt the teacher at $u$ above a cutoff (e.g., freeze its weights or add noise) and see whether the KDC loop still selects large steps and whether FID holds; if it does, the KDC signal is not tracking target quality.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the knowledge discrepancy of a consistency-distillation curriculum — quantified as $\mathrm{KDC} = 100 - \mathrm{PSNR}(x_{\mathrm{est}}, x_{\mathrm{target}})$ — falls monotonically as the timestep moves from noise to image, and that this falling difficulty is harmful. CCM therefore replaces the fixed or monotonically shrinking distillation step $l = u - t$ used by prior consistency models with an adaptive schedule: starting from $t$, the teacher advances in steps of size $s$, recomputing the target each time, until KDC exceeds a threshold $T_{\mathrm{KDC}}$. The resulting step is small near pure noise and grows when the image is clear, so the student always trains on a curriculum of comparable difficulty. The paper reports that this balanced curriculum, combined with a GAN loss, yields one-step FID 1.64 on CIFAR-10 and 2.18 on ImageNet 64x64, and that on SDXL and SD3 it improves FID, CLIP score, and all six T2I-CompBench categories relative to LCM and PCM.

Load-bearing premise

The whole scheme assumes the frozen teacher's forward ODE solves stay accurate over the sometimes-large intervals $u-t$ that the KDC loop selects; if the teacher drifts on those long jumps, the student is trained against a degraded target while the KDC threshold still reads "difficult enough."

Editorial extensions

If this is right

One-step sampling on CIFAR-10 reaches FID 1.64, and on ImageNet 64x64 FID 2.18, competitive with teachers that use dozens or hundreds of function evaluations.
The adaptive schedule transfers to large text-to-image models: on SDXL and SD3, CCM reports lower FID and higher CLIP scores than LCM and PCM at 4 steps.
T2I-CompBench scores improve across all six compositional categories for both SD3 and SDXL, indicating better attribute binding, object relationship, and complex reasoning.
Training converges faster: CCM reaches the same FID as vanilla consistency distillation in about 1.3x fewer iterations.
The method is robust to its main hyperparameters: $T_{\mathrm{KDC}}$ in the range 60–70 and step size $s$ around 0.03 both give stable gains over CD.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same curriculum-balancing idea should apply to any self-distillation scheme where a student learns from a frozen or EMA teacher along an ODE — for instance, few-step video or audio generation — provided the discrepancy metric is chosen for that modality.
Because CCM's selected intervals grow exactly when the image is clear, the KDC signal could double as a per-sample difficulty weight for sampling $t$ or for annealing $T_{\mathrm{KDC}}$, which the paper leaves open.
A testable extension would couple the KDC threshold to per-prompt statistics (e.g., guidance scale or prompt complexity), since text-to-image results show the largest gains on compositional prompts.
If the multi-step teacher iterations are the true source of gains, then a version that precomputes the adaptive $u$ schedule once per training stage and then trains with one-step targets should recover most of the benefit at lower compute.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The core idea here is simple and worth taking seriously: measure the student–teacher gap in PSNR terms, and when the gap is small, let the teacher iterate further so the student always faces a roughly constant difficulty. That framing is new relative to iCT and ECM, which use fixed decaying schedules, and the extension to flow matching and to SDXL/SD3 is genuinely useful. The paper also does some things right: it shows KDC behavior across datasets and models, includes ablations for static versus dynamic schedules, tests single-step versus multi-step teacher targets, and checks sensitivity to its main threshold. Section 3.3 even acknowledges the teacher-drift risk. That is more than many distillation papers bother to do.

The soft spot, though, is structural. Every headline result—CIFAR-10 FID 1.64, ImageNet 64 FID 2.18, the SDXL/SD3 tables—comes from the full pipeline with the GAN loss. The only clean comparison of the adaptive schedule itself is Table 5, which is explicitly without adversarial losses, and there the gap over the best static schedule is 0.57 FID. There is no static-schedule-plus-GAN control anywhere. So we cannot tell whether the adaptive schedule, the GAN loss, or their interaction produces the headline numbers. The stress-test note is right: the paper sets up the exact control needed and never runs it. This is not a fatal flaw—the adaptive idea may well survive the test—but it is the load-bearing attribution claim, and it is currently unverified.

Other issues are real but secondary: no error bars or seeds, so even the 0.57 FID gap in Table 5 is not established as significant; T_KDC is tuned on the reported FID (though the 60–70 band looks robust); the user study is one sentence with no protocol; and the large-scale setup omits details like latent-space PSNR computation, LoRA rank, and the adversarial weight. These are the kinds of things a referee would ask for, not reasons to dismiss the work.

Overall, this is a conditional accept in my mind. The paper deserves a serious referee because the idea is novel, the experiments span small and large models, and the ablation design is on the right track. I would ask for the missing static+GAN control, error bars, and a fuller experimental appendix before trusting the headline. For someone working on consistency distillation or few-step generation, this is a useful read despite the unresolved attribution gap.

Referee Report

4 major / 6 minor

Summary. The manuscript proposes the Curriculum Consistency Model (CCM), a modification of consistency distillation in which the distillation interval is chosen adaptively rather than fixed. At a sampled timestep t, the student prediction is compared with teacher rollouts advanced in small steps of size s; the rollout stops when the PSNR-based Knowledge Discrepancy of the Curriculum (KDC) exceeds a fixed threshold T_KDC, and the student is trained against the resulting target together with an adversarial loss. Experiments report one-step FID scores of 1.64 on CIFAR-10 and 2.18 on ImageNet 64x64, and improvements in CLIP score, FID, and T2I-CompBench over LCM and PCM on SDXL and SD3. Ablations on CIFAR-10 show that the dynamic schedule outperforms static schedules when no GAN loss is used.

Significance. If the reported results hold, the paper would establish a simple and general recipe for stabilizing consistency-distillation difficulty across timesteps, with a rare demonstration on both diffusion-based and flow-matching text-to-image models. The core mechanism is concrete, computationally cheap, and the CIFAR-10 ablations in Table 5 do show a consistent ranking of dynamic over static schedules under the distillation-only objective. The extension to SDXL and SD3 gives the work practical relevance. However, the absence of GAN-matched controls for the schedule, the lack of error bars, and incomplete training details currently prevent the evidence from supporting the headline attribution at the level claimed.

major comments (4)

[§4.3, Table 5 vs. Tables 1–3, Eq. (9)] The causal contribution of the KDC schedule to the headline numbers is not established. All headline results in Tables 1–4 are reported for the full objective including adversarial losses, while the only direct comparison between static and dynamic schedules is explicitly run "without adversarial losses" in Table 5. In that table, CCM's advantage over the best static strategy is 9.32 vs. 9.89 FID, and no seeds or error bars are reported, so the gap may not be statistically meaningful. A static schedule combined with the same GAN objective could match or beat CCM+GAN, in which case the headline FIDs would be attributable to the adversarial component rather than to the curriculum mechanism. The authors should run the best static strategy (l=0.09, n=3, s=0.03) and at least one other static baseline with the identical GAN loss and training budget, and report FID over multiple seeds.
[§3.3, Eq. (6), Algorithm 1, Figure 9] The adaptive schedule is endogenous to the student being trained, and the threshold T_KDC is a tuned hyperparameter. KDC is computed between the current student fθ and the target model fθ−, so the stopping condition depends on the very model being trained. The threshold T_KDC=60 is used for all datasets and model scales, but the sensitivity analysis in Figure 9 is only shown for CIFAR-10. To support the claim that a single threshold is appropriate, the authors should provide threshold sensitivity or an adaptation mechanism for ImageNet 64x64 and for the SDXL/SD3 experiments, and should state explicitly how T_KDC was selected.
[§3.3 and Algorithm 1] The stopping criterion does not protect against teacher drift. The paper itself notes that the teacher model may make inaccurate predictions due to large timestep sizes in the same section, and because the loop only checks KDC between the student and the teacher-produced target, a low-quality target can still satisfy the threshold and the algorithm will appear to work while training against noisy labels. Please quantify target quality, for example by reporting PSNR between multi-step teacher rollouts and reference trajectory estimates, or by comparing CCM against a variant that caps the distillation interval l or uses a fixed high-quality target schedule.
[§4.1 and Appendix A] The large-scale experiments omit several details needed to reproduce or interpret them: the PSNR computation for latent-space models (latent channels, dynamic range, whether computed in latent space or after decoding), the LoRA rank and insertion points, the value and scheduling of λGAN, the ODE solver and discretization for the teacher rollouts, and the exact COCO-2017 5K split used for evaluation. The user study in Figure 5 is also described only as "subjects were shown generated images and asked for preference," with no number of subjects, protocol, or significance test. These omissions make the text-to-image generalization claims difficult to verify.

minor comments (6)

[§3.2] The phrase "to access knowledge discrepancy" should read "to assess knowledge discrepancy."
[Eq. (6)] The definition of KDC uses (2^n − 1)^2 with n representing bit depth, but the text does not state the range of xest and xtarget; this is especially important for latent-space models, where values are not 8-bit image intensities.
[Algorithm 1] Inside the loop, t and xt are reassigned (lines 10–11), so the subscript t in KDCu_t and in xest becomes ambiguous. Use a separate variable such as t0 for the initial time and keep xest fixed.
[Table 2] The CTM entry appears as "11.92"; clarify whether this is NFE=1 with FID=1.92 or another configuration, and add a column separator for readability.
[Table 5] The caption writes l = Pn i=1 si, but for dynamic rows such as l=0.1t and l=1−t, l is a function of t rather than a fixed quantity; state explicitly that these are functional forms.
[§3.3 and Figure 9] The text says KDC stability across training iterations is shown in Section 3, but Figure 9 reports only final FID as a function of T_KDC; either add the promised KDC-versus-iteration plot or point to the correct figure.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: KDC is an endogenous control signal, and the headline FID/CLIP claims rest on external benchmarks.

full rationale

The central empirical claims—one-step FID 1.64 on CIFAR-10, 2.18 on ImageNet 64x64, and improved CLIP/T2I-CompBench scores on SD3/SDXL—are measured against external benchmarks (FID, CLIP, T2I-CompBench, user study) using fixed pretrained teachers (OT-CFM, SD3, SDXL). These numbers are not equal by construction to the KDC definition or to the KDC-driven schedule. Equation 6 defines KDC as 100 minus the PSNR between the student output f_theta(x_t,t,1) and the teacher-derived target f_theta-(Solver(...),u,1), and Algorithm 1 uses this quantity as a control signal to advance u until KDC exceeds T_KDC. This is an endogenous, self-paced training heuristic: the student's current output influences which target is used, but the final evaluation is an independent measurement of the trained student on held-out data. The threshold T_KDC=60 is examined in the ablation in Figure 9, and the paper reports that the method is not very sensitive to it over the range 60-70; selecting a hyperparameter by ablation and then reporting the resulting benchmark score is standard practice, not a fitted quantity renamed as a prediction. The controlled ablation in Table 5 compares dynamic and static schedules under the same external FID metric without adversarial losses, which provides independent evidence that the adaptive schedule helps even apart from the headline GAN-equipped results. The absence of a static-schedule-plus-GAN control is a real attribution gap for the headline 1.64 number, but it is an experimental omission, not a circular derivation. Self-citations to the Phased Consistency Model and torchcfm appear as implementation baselines and comparison methods, but the core claim does not reduce to those citations, and no uniqueness theorem or prior-work ansatz is invoked to force the KDC choice. Overall, no load-bearing step in the paper is equivalent to its own input by construction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method is fundamentally empirical: the KDC metric, the threshold, and the step size are chosen by hand or by ablation, and the teacher-accuracy assumption is stated but not proven. No new geometric, statistical, or physical quantities are invented, so the entity ledger is empty.

free parameters (3)

T_KDC = 60
Stopping threshold in Algorithm 1; selected via Figure 9 ablation on CIFAR-10, then used for all datasets and model scales.
s (base timestep size) = 0.03
Step size for teacher multi-step iteration; selected via Table 6 ablation on CIFAR-10. The paper states s=0.03 slightly outperforms other choices.
lambda_GAN = adaptive, unspecified
Adversarial loss weight; the paper says the weighting is adaptive with details in Kim et al. (2023), but no concrete value or schedule is given.

assumptions (4)

domain assumption The PF-ODE trajectory is smooth enough that Euler solves with step size s yield approximately correct targets.
Invoked by Algorithm 1 when calling Solver(xt, t, u; phi) repeatedly; if the teacher drifts, the enlarged distillation interval propagates error into the student's targets.
ad hoc to paper PSNR between student and teacher outputs is a reliable, stable measure of learning difficulty across timesteps, datasets, and model architectures.
Propounded in Sec 3.2 and defended by Figure 2, but no theoretical justification is given; the entire adaptive schedule relies on this metric.
ad hoc to paper A single fixed threshold T_KDC=60 is appropriate for all stages of training and all model scales.
Algorithm 1 uses a fixed threshold; Figure 9 shows sensitivity only on CIFAR-10 without adversarial loss, and the same threshold is then applied to ImageNet and large text-to-image models.
standard math Standard results of consistency distillation, including local consistency (Eq. 2) and the stop-gradient on the teacher, are valid.
Taken from Song et al. (2023), Kim et al. (2023), and Lu & Song (2024); not re-derived in this paper.

how reviews work

0 comments

Cite this review

Pith. "Pith review of See Further When Clear: Curriculum Consistency Model." pith.science (2026). https://pith.science/paper/DB44CZEL

@misc{pith2026241206295,
  author       = {Pith},
  title        = {Pith review of: See Further When Clear: Curriculum Consistency Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DB44CZEL}},
  note         = {Machine review of arXiv:2412.06295}
}

read the original abstract

Significant advances have been made in the sampling efficiency of diffusion models and flow matching models, driven by Consistency Distillation (CD), which trains a student model to mimic the output of a teacher model at a later timestep. However, we found that the learning complexity of the student model varies significantly across different timesteps, leading to suboptimal performance in CD.To address this issue, we propose the Curriculum Consistency Model (CCM), which stabilizes and balances the learning complexity across timesteps. Specifically, we regard the distillation process at each timestep as a curriculum and introduce a metric based on Peak Signal-to-Noise Ratio (PSNR) to quantify the learning complexity of this curriculum, then ensure that the curriculum maintains consistent learning complexity across different timesteps by having the teacher model iterate more steps when the noise intensity is low. Our method achieves competitive single-step sampling Fr\'echet Inception Distance (FID) scores of 1.64 on CIFAR-10 and 2.18 on ImageNet 64x64.Moreover, we have extended our method to large-scale text-to-image models and confirmed that it generalizes well to both diffusion models (Stable Diffusion XL) and flow matching models (Stable Diffusion 3). The generated samples demonstrate improved image-text alignment and semantic structure, since CCM enlarges the distillation step at large timesteps and reduces the accumulated error.

Figures

Figures reproduced from arXiv: 2412.06295 by the authors.

**Figure 2.** Knowledge Discrepancy Investigation: Analysis of the KDC over (t, u) across different timesteps on various datasets for both flow matching models and diffusion models. We conducted measurements on both diffusion models (SD 1.5 Rombach et al. (2022), SDXL Podell et al. (2024)) and flow matching models (SD3 Esser et al. (2024), OTCFM Tong et al. (2023)) and select 3 classic datasets (CIFAR-10, ImageNet, and CC3M) co… view at source ↗

**Figure 3.** The relationship of KDC with different distillation steps l. We further explored the effect of distillation step l = u−t in CM, and the results are presented in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (8 more)

**Figure 4.** Figure 4: Distillation step vs. training iterations in CCM and iCT. Algorithm 1 KDC-Adjusted Target Computation 1: Input: noisy input xt, timestep size s, threshold TKDC, teacher model ϕ, target model fθ− , student model fθ 2: Output: KDC-Adjusted target x KDC target 3: Sample …

**Figure 5.** Figure 5: User study. Subjects were shown generated images and asked for preference [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

**Figure 6.** Figure 6: Samples generated by OT-CFM and CCM on CIFAR-10 and ImageNet 64x64. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

**Figure 7.** Figure 7: Semantic comparison of images generated by LCM (up) and CCM (down). CCM shows better image-text alignment and generates images that better fit the text. A glass of orange juice to the right of a plate with buttered toast on it. A drawing of a stork playing a violin. …

**Figure 9.** Figure 9: Comparisons of different TKDC. The choice of TKDC. Different TKDC determine the dynamically selected number of iterative steps during the training process, which is a hyperparameter in the methods presented in this paper. We conducted experiments with different values…

**Figure 10.** Figure 10: Samples generated by CCM (NFE=1) trained on CIFAR-10. [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]

**Figure 11.** Figure 11: Samples generated by CCM (NFE=1) trained on ImageNet 64x64 according to random [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]

**Figure 12.** Figure 12: Samples generated by CCM (NFE=1) trained on ImageNet 64x64 according to specified [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Consistency Deep Equilibrium Models
cs.LG 2026-02 conditional novelty 6.0 of 10

C-DEQ trains a consistency model to map intermediate solver states directly to the DEQ equilibrium, enabling accurate one-to-few-step inference for deep equilibrium models.

Reference graph

Works this paper leans on

17 extracted references · 3 canonical work pages · cited by 1 Pith paper

[3]

Efficient diffusion training via min-snr weighting strategy

Tiankai Hang, Shuyang Gu, Chen Li, Jianmin Bao, Dong Chen, Han Hu, Xin Geng, and Baining Guo. Efficient diffusion training via min-snr weighting strategy. arXiv preprint arXiv:2303.09556,

arXiv
[5]

Microsoft coco: Common objects in context

Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13, pp. 740–755. Springer,

2014
[7]

Rectified flow: A marginal preserving approach to optimal transport

Qiang Liu. Rectified flow: A marginal preserving approach to optimal transport. arXiv preprint arXiv:2209.14577,

arXiv
[9]

Simplifying, stabilizing and scaling continuous-time consistency models

Cheng Lu and Yang Song. Simplifying, stabilizing and scaling continuous-time consistency models. arXiv preprint arXiv:2410.11081,

arXiv
[10]

Cm-gan: Stabilizing gan training with consistency models

Haoye Lu, Yiwei Lu, Dihong Jiang, Spencer Ryan Szabados, Sun Sun, and Yaoliang Yu. Cm-gan: Stabilizing gan training with consistency models. In ICML 2023 Workshop,

work page 2023
[11]

Latent consistency models: Synthe- sizing high-resolution images with few-step inference

Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthe- sizing high-resolution images with few-step inference. arXiv preprint arXiv:2310.04378,

arXiv
[12]

Multisample flow matching: Straightening flows with minibatch cou- plings

Aram-Alexandre Pooladian, Heli Ben-Hamu, Carles Domingo-Enrich, Brandon Amos, Yaron Lip- man, and Ricky TQ Chen. Multisample flow matching: Straightening flows with minibatch cou- plings. arXiv preprint arXiv:2304.14772,

arXiv
[14]

Stylegan-xl: Scaling stylegan to large diverse datasets

Axel Sauer, Katja Schwarz, and Andreas Geiger. Stylegan-xl: Scaling stylegan to large diverse datasets. In ACM SIGGRAPH 2022 conference proceedings, pp. 1–10,

2022

Show all 17 references

[16]

A E XPERIMENTAL HYPERPARAMETERS We minimally change the OT-CFM’s Tong et al

12 Preprint. A E XPERIMENTAL HYPERPARAMETERS We minimally change the OT-CFM’s Tong et al. (2023) design to comply the previous implementa- tion, and important modifications are listed in Table

2023
[17]

Table 7: Experimental details on hyperparameters. Hyperparameter CIFAR-10 32x32 ImageNet 64x64 CC3M 1024x1024 Training type unconditional conditional conditional Learning rate 2e-4 1e-5 5e-6 Discriminator learning rate 0.002 0.002 1e-5 target EMA decay rate µ 0.9 0.9 - student...

2024
[2009]

Sdxl-lightning: Progressive adversarial diffusion distillation

Shanchuan Lin, Anran Wang, and Xiao Yang. Sdxl-lightning: Progressive adversarial diffusion distillation. arXiv preprint arXiv:2402.13929,

arXiv
[2018]

Imagenet: A large-scale hi- erarchical image database

Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hi- erarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee,

2009
[2020]

Improved techniques for training consistency models

Yang Song and Prafulla Dhariwal. Improved techniques for training consistency models. arXiv preprint arXiv:2310.14189,

arXiv
[2021]

Hyper-sd: Trajectory segmented consistency model for efficient image synthesis

Yuxi Ren, Xin Xia, Yanzuo Lu, Jiacheng Zhang, Jie Wu, Pan Xie, Xing Wang, and Xuefeng Xiao. Hyper-sd: Trajectory segmented consistency model for efficient image synthesis. arXiv preprint arXiv:2404.13686,

arXiv
[2022]

Xingchao Liu, Chengyue Gong, and Qiang Liu

11 Preprint. Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003,

arXiv
[2023]

Scott: Accelerating diffusion models with stochastic consistency distillation

Hongjian Liu, Qingsong Xie, Zhijie Deng, Chen Chen, Shixiang Tang, Fueyang Fu, Zheng-jun Zha, and Haonan Lu. Scott: Accelerating diffusion models with stochastic consistency distillation. arXiv preprint arXiv:2403.01505,

arXiv
[2024]

Consistency models made easy

Zhengyang Geng, Ashwini Pokle, William Luo, Justin Lin, and J Zico Kolter. Consistency models made easy. arXiv preprint arXiv:2406.14548,

arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [3]

Efficient diffusion training via min-snr weighting strategy

Tiankai Hang, Shuyang Gu, Chen Li, Jianmin Bao, Dong Chen, Han Hu, Xin Geng, and Baining Guo. Efficient diffusion training via min-snr weighting strategy. arXiv preprint arXiv:2303.09556,

arXiv

[2] [5]

Microsoft coco: Common objects in context

Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13, pp. 740–755. Springer,

2014

[3] [7]

Rectified flow: A marginal preserving approach to optimal transport

Qiang Liu. Rectified flow: A marginal preserving approach to optimal transport. arXiv preprint arXiv:2209.14577,

arXiv

[4] [9]

Simplifying, stabilizing and scaling continuous-time consistency models

Cheng Lu and Yang Song. Simplifying, stabilizing and scaling continuous-time consistency models. arXiv preprint arXiv:2410.11081,

arXiv

[5] [10]

Cm-gan: Stabilizing gan training with consistency models

Haoye Lu, Yiwei Lu, Dihong Jiang, Spencer Ryan Szabados, Sun Sun, and Yaoliang Yu. Cm-gan: Stabilizing gan training with consistency models. In ICML 2023 Workshop,

work page 2023

[6] [11]

Latent consistency models: Synthe- sizing high-resolution images with few-step inference

Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthe- sizing high-resolution images with few-step inference. arXiv preprint arXiv:2310.04378,

arXiv

[7] [12]

Multisample flow matching: Straightening flows with minibatch cou- plings

Aram-Alexandre Pooladian, Heli Ben-Hamu, Carles Domingo-Enrich, Brandon Amos, Yaron Lip- man, and Ricky TQ Chen. Multisample flow matching: Straightening flows with minibatch cou- plings. arXiv preprint arXiv:2304.14772,

arXiv

[8] [14]

Stylegan-xl: Scaling stylegan to large diverse datasets

Axel Sauer, Katja Schwarz, and Andreas Geiger. Stylegan-xl: Scaling stylegan to large diverse datasets. In ACM SIGGRAPH 2022 conference proceedings, pp. 1–10,

2022