[ICLR 2026] MeanCache: From Instantaneous to Average Velocity for Accelerating Flow Matching Inference

February 5, 2026 · View on GitHub

Huanlin Gao^1,2, Ping Chen^1,2, Fuyuan Shi^1,2, Ruijia Wu², Yantao Li^1,2,3, Qiang Hui^1,2,
Yuren You², Ting Lu^1,2, Chao Tan^1,2, Shaoan Zhao^1,2, Zhaoxiang Liu^1,2
Fang Zhao^1,2*, Kai Wang^1,2, Shiguo Lian^1,2*

¹Data Science & Artificial Intelligence Research Institute, China Unicom,
²Unicom Data Intelligence, China Unicom,
³National Key Laboratory for Novel Software Technology, Nanjing University

(* Corresponding author.)

🎬 Demo Video

https://github.com/user-attachments/assets/6deadbcf-0a7f-4ecc-96fa-645ca86bba7f

In Flow Matching inference, existing caching methods primarily rely on reusing Instantaneous Velocity or its feature-level proxies. However, we observe that instantaneous velocity often exhibits sharp fluctuations across timesteps. This leads to severe trajectory deviations and cumulative errors, especially as the cache interval increases. Inspired by MeanFlow, we propose MeanCache. Compared to unstable instantaneous velocity, Average Velocity is significantly smoother and more robust over time. By shifting the caching perspective from a single "point" to an "interval," MeanCache effectively mitigates trajectory drift under high acceleration ratios.

Latest News

[2026/02/05] Community Contribution: ComfyUI-MeanCache-Z is now available! thanks to @facok!
[2025/02/04] Support Z-Image and released the MeanCache vs. LeMiCa comparative study.
[2025/02/02] Support Qwen-Image and Inference Code Released !

MeanCache vs. LeMiCa

This benchmark evaluates the performance of MeanCache against LeMiCa using the Qwen-Image-2512 model as the base.

🚀 Efficiency

Baseline Latency (Original Qwen-Image-2512): 32.8s

Constraint	Method	Latency	Speedup	Time Reduction
$B=25$	LeMiCa	18.83 s	1.74x	-
	MeanCache	17.13 s	1.91x	9.0%
$B=17$	LeMiCa	14.35 s	2.29x	-
	MeanCache	11.67 s	2.81x	18.7%
$B=10$	LeMiCa	10.41 s	3.15x	-
	MeanCache	6.95 s	4.72x	33.2%

🎨 Quality

Constraint	Method	PSNR (↑)	SSIM (↑)	LPIPS (↓)
$B=25$	LeMiCa	29.20	0.945	0.065
	MeanCache	29.46	0.944	0.057
$B=17$	LeMiCa	24.31	0.835	0.176
	MeanCache	26.49	0.907	0.104
$B=10$	LeMiCa	17.80	0.637	0.368
	MeanCache	19.44	0.767	0.237

Demo

Z-Image

Z-Image-base	MeanCache(B=25)	MeanCache(B=20)	MeanCache(B=15)	MeanCache(B=13)
18.07 s	9.15 s	7.36 s	5.58 s	4.85 s

Qwen-Image-2512

Method	Qwen-Image-2512	MeanCache(B=25)	MeanCache(B=17)	MeanCache(B=10)
Latency	32.8 s	17.13 s	11.67 s	6.95 s
T2I

Qwen-Image

Method	Qwen-Image	MeanCache(B=25)	MeanCache(B=17)	MeanCache(B=10)
Latency	33.13 s	17.04 s	11.63 s	6.92 s
T2I

License

The majority of this project is released under the Apache 2.0 license as found in the LICENSE file.

📖 Citation

If you find MeanCache useful in your research or applications, please consider giving us a star ⭐ and citing it by the following BibTeX entry:

@inproceedings{gao2025meancache,
  title     = {MeanCache: From Instantaneous to Average Velocity for Accelerating Flow Matching Inference},
  author    = {Huanlin Gao and Ping Chen and Fuyuan Shi and Ruijia Wu and Yantao Li and Qiang Hui and Yuren You and Ting Lu and Chao Tan and Shaoan Zhao and Zhaoxiang Liu and Fang Zhao and Kai Wang and Shiguo Lian},
  journal   = {International Conference on Learning Representations (ICLR)},
  year      = {2026},
  url       = {https://arxiv.org/abs/2601.19961}
}