LeMiCa already supports accelerated inference for Z-Image and provides three optional acceleration paths that balance image quality and speed.

Z-Image	LeMiCa-slow	LeMiCa-medium	LeMiCa-fast
2.55 s	2.19 s	1.94 s	1.78 s

Note: The above numbers are example latency measurements for a single H800 GPU with a 1024×1024 resolution. Actual performance may vary depending on hardware and configuration.

Please refer to the original Z-Image project for base installation instructions.

# vanilla Z-Image (no caching / acceleration)
python inference_zimage.py

# LeMiCa acceleration modes
python inference_zimage.py --cache slow
python inference_zimage.py --cache medium
python inference_zimage.py --cache fast

# use an explicit numeric cache step
python inference_zimage.py --cache 8

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

@inproceedings{gao2025lemica,
  title     = {LeMiCa: Lexicographic Minimax Path Caching for Efficient Diffusion-Based Video Generation},
  author    = {Huanlin Gao and Ping Chen and Fuyuan Shi and Chao Tan and Zhaoxiang Liu and Fang Zhao and Kai Wang and Shiguo Lian},
  journal   = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2025},
  url       = {https://arxiv.org/abs/2511.00090}
}

We would like to thank the contributors to the Z-Image and Diffusers.