Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer
Paper • 2511.22699 • Published • 245
Z-Image Supervised FineTune (FP32)
This repository provides an FP32 (32-bit floating point) version of the Z-Image Supervised FineTune model.
The FP32 variant is intended for research, analysis, and compatibility testing, and preserves the model in full numerical precision compared to reduced-precision releases (e.g. BF16).
Model Overview
- Model name: Z-Image Supervised FineTune (FP32)
- Precision: FP32
- Task: Image generation
- Framework: PyTorch
- Weights format:
.safetensors
This release is functionally equivalent to the original Z-Image Supervised FineTune architecture, differing only in numerical precision.
FP32 inference on consumer GPUs may be slower than BF16.
🔗 Credits & Source Code
@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} }
-- Disclaimer: These files are provided "as is" for experimental builds. Please check the original repositories for licensing information regarding commercial use.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support