Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,75 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
pipeline_tag: text-to-image
|
| 6 |
+
library_name: diffusers
|
| 7 |
+
tags:
|
| 8 |
+
- comfyui
|
| 9 |
+
- comfy
|
| 10 |
+
- z-img
|
| 11 |
+
- fp8
|
| 12 |
+
- quantized
|
| 13 |
---
|
| 14 |
+
|
| 15 |
+
# 🔢 FP8 Quantized Version - ComfyUI Compatible
|
| 16 |
+
|
| 17 |
+
This is the **fp8_e4m3fn** and **fp8_e5m2** quantized version of the Z-Image model, optimized for ComfyUI workflows. These quantized formats significantly reduce VRAM requirements while maintaining high image quality, making the model more accessible for consumer-grade GPUs.
|
| 18 |
+
|
| 19 |
+
**Quantization Formats:**
|
| 20 |
+
- `fp8_e4m3fn`: 4-bit exponent, 3-bit mantissa format
|
| 21 |
+
- `fp8_e5m2`: 5-bit exponent, 2-bit mantissa format
|
| 22 |
+
|
| 23 |
+
**Benefits:**
|
| 24 |
+
- Reduced memory footprint (~50% VRAM savings)
|
| 25 |
+
- Faster inference times
|
| 26 |
+
- Full ComfyUI compatibility
|
| 27 |
+
- Minimal quality degradation
|
| 28 |
+
|
| 29 |
+
---
|
| 30 |
+
|
| 31 |
+
<h1 align="center">⚡️- Image<br><sub><sup>An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer</sup></sub></h1>
|
| 32 |
+
<div align="center">
|
| 33 |
+
|
| 34 |
+
[](https://tongyi-mai.github.io/Z-Image-blog/) 
|
| 35 |
+
[](https://github.com/Tongyi-MAI/Z-Image) 
|
| 36 |
+
[](https://huggingface.co/Tongyi-MAI/Z-Image) 
|
| 37 |
+
[](https://www.modelscope.cn/models/Tongyi-MAI/Z-Image) 
|
| 38 |
+
[](https://www.modelscope.cn/aigc/imageGeneration?tab=advanced&versionId=569345&modelType=Checkpoint&sdVersion=Z_IMAGE&modelUrl=modelscope%3A%2F%2FTongyi-MAI%2FZ-Image%3Frevision%3Dmaster) 
|
| 39 |
+
<a href="https://arxiv.org/abs/2511.22699" target="_blank"><img src="https://img.shields.io/badge/Report-b5212f.svg?logo=arxiv" height="21px"></a>
|
| 40 |
+
|
| 41 |
+
</div>
|
| 42 |
+
|
| 43 |
+
## 🎨 Z-Image
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
|
| 50 |
+
|
| 51 |
+
**Z-Image** is the foundation model of the ⚡️- Image family, engineered for good quality, robust generative diversity, broad stylistic coverage, and precise prompt adherence.
|
| 52 |
+
|
| 53 |
+
While Z-Image-Turbo is built for speed, Z-Image is a full-capacity, undistilled transformer designed to be the backbone for creators, researchers, and developers who require the highest level of creative freedom.
|
| 54 |
+
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
### 🌟 Key Features
|
| 58 |
+
|
| 59 |
+
- **Undistilled Foundation**: As a non-distilled base model, Z-Image preserves the complete training signal. It supports full Classifier-Free Guidance (CFG), providing the precision required for complex prompt engineering and professional workflows.
|
| 60 |
+
- **Aesthetic Versatility**: Z-Image masters a vast spectrum of visual languages—from hyper-realistic photography and cinematic digital art to intricate anime and stylized illustrations. It is the ideal engine for scenarios requiring rich, multi-dimensional expression.
|
| 61 |
+
- **Enhanced Output Diversity**: Built for exploration, Z-Image delivers significantly higher variability in composition, facial identity, and lighting across different seeds, ensuring that multi-person scenes remain distinct and dynamic.
|
| 62 |
+
- **Built for Development**: The ideal starting point for the community. Its non-distilled nature makes it a good base for LoRA training, structural conditioning (ControlNet) and semantic conditioning.
|
| 63 |
+
- **Robust Negative Control**: Responds with high fidelity to negative prompting, allowing users to reliably suppress artifacts and adjust compositions.
|
| 64 |
+
|
| 65 |
+
### 🆚 Z-Image vs Z-Image-Turbo
|
| 66 |
+
|
| 67 |
+
| Aspect | Z-Image | Z-Image-Turbo |
|
| 68 |
+
|------|------|------|
|
| 69 |
+
| CFG | ✅ | ❌ |
|
| 70 |
+
| Steps | 28~50 | 8 |
|
| 71 |
+
| Fintunablity | ✅ | ❌ |
|
| 72 |
+
| Negative Prompting | ✅ | ❌ |
|
| 73 |
+
| Diversity | High | Low |
|
| 74 |
+
| Visual Quality | High | Very High |
|
| 75 |
+
| RL | ❌ | ✅ |
|