Z-Image-Turbo-MLX

Unofficial MLX-format weights converted from Tongyi-MAI/Z-Image-Turbo for native Apple Silicon inference.

This is NOT the official Z-Image repository. For the original model and official resources, please visit:

• Official model: Tongyi-MAI/Z-Image-Turbo
• Official GitHub: github.com/Tongyi-MAI/Z-Image

About

Z-Image-Turbo is a powerful and efficient 6B-parameter image generation model developed by the Tongyi-MAI team (Alibaba). It achieves sub-second inference on enterprise GPUs with only 8 NFEs, excelling at photorealistic generation, bilingual text rendering, and instruction adherence.

This repository contains the same weights converted to MLX format (Apple's ML framework) so they can be loaded natively on Apple Silicon (M1/M2/M3/M4) without PyTorch or CUDA dependencies.

Conversion Details

Item	Detail
Source model	Tongyi-MAI/Z-Image-Turbo
Target format	MLX (safetensors with MLX-compatible layout)
Precision	bfloat16
License	Apache-2.0 (inherited from the original model)
Converted by	illusion615

Intended Use

These weights are designed for use with a local project's zimage_mlx_service backend, enabling local image generation on Mac with Apple Silicon.

Citation

If you use these weights, please cite the original Z-Image work:

@article{team2025zimage,
  title={Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer},
  author={Z-Image Team},
  journal={arXiv preprint arXiv:2511.22699},
  year={2025}
}

Acknowledgments

All credit for the Z-Image model architecture, training, and research goes to the Tongyi-MAI team. This repository only provides a format conversion for Apple Silicon compatibility.