This repository contains Nunchaku-quantized versions of Z-Image-Turbo, a high-performance image generation model. It is optimized for efficient inference while maintaining minimal loss in performance.

No recent news. Stay tuned for updates!

• Developed by: Nunchaku Team (thank @devgdovg)
• Model type: image-to-image
• License: apache-2.0
• Quantized from model: Z-Image-Turbo

Data Type: INT4 for non-Blackwell GPUs (pre-50-series), NVFP4 for Blackwell GPUs (50-series). Rank:

• r32 for faster inference,
• r128 for better quality but slower inference,
• r256 for highest quality (slowest inference).

Standard inference speed models for general use

Data Type	Rank	Model Name	Comment
INT4	r32	svdq-int4_r32-z-image-turbo.safetensors
	r128	svdq-int4_r128-z-image-turbo.safetensors
	r256	svdq-int4_r256-z-image-turbo.safetensors
NVFP4	r32	svdq-fp4_r32-z-image-turbo.safetensors
	r128	svdq-fp4_r128-z-image-turbo.safetensors

• Inference Engine: nunchaku
• Quantization Library: deepcompressor
• Paper: SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models
• Demo: demo.nunchaku.tech

• Diffusers Usage: See z-image-turbo.py. Check this tutorial for more advanced usage.
• ComfyUI Usage: See nunchaku-z-image-turbo.json.

@inproceedings{
  li2024svdquant,
  title={SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models},
  author={Li*, Muyang and Lin*, Yujun and Zhang*, Zhekai and Cai, Tianle and Li, Xiuyu and Guo, Junxian and Xie, Enze and Meng, Chenlin and Zhu, Jun-Yan and Han, Song},
  booktitle={The Thirteenth International Conference on Learning Representations},
  year={2025}
}