---
base_model: Qwen/Qwen-Image
base_model_relation: quantized
datasets:
- mit-han-lab/svdquant-datasets
language:
- en
library_name: diffusers
license: apache-2.0
pipeline_tag: text-to-image
tags:
- text-to-image
- SVDQuant
- Qwen-Image
- Diffusion
- Quantization
- ICLR2025

---
<p align="center" style="border-radius: 10px">
  <img src="https://huggingface.co/datasets/nunchaku-tech/cdn/resolve/main/nunchaku/assets/nunchaku.svg" width="30%" alt="Nunchaku Logo"/>
</p>

# Model Card for nunchaku-qwen-image

![comfyui](https://huggingface.co/datasets/nunchaku-tech/cdn/resolve/main/ComfyUI-nunchaku/workflows/nunchaku-qwen-image.png)![visual](https://huggingface.co/datasets/nunchaku-tech/cdn/resolve/main/nunchaku/assets/qwen-image.jpg)
This repository contains Nunchaku-quantized versions of [Qwen-Image](https://huggingface.co/Qwen/Qwen-Image), designed to generate high-quality images from text prompts, advances in complex text rendering. It is optimized for efficient inference while maintaining minimal loss in performance.

## Model Details

### Model Description

- **Developed by:** Nunchaku Team
- **Model type:** text-to-image
- **License:** apache-2.0
- **Quantized from model:** [Qwen-Image](https://huggingface.co/Qwen/Qwen-Image)

### Model Files

- [`svdq-int4_r32-qwen-image.safetensors`](./svdq-int4_r32-qwen-image.safetensors): SVDQuant quantized INT4 Qwen-Image model with rank 32. For users with non-Blackwell GPUs (pre-50-series).
- [`svdq-int4_r128-qwen-image.safetensors`](./svdq-int4_r128-qwen-image.safetensors): SVDQuant quantized INT4 Qwen-Image model with rank 128. For users with non-Blackwell GPUs (pre-50-series). It offers better quality than the rank 32 model, but it is slower.
- [`svdq-fp4_r32-qwen-image.safetensors`](./svdq-fp4_r32-qwen-image.safetensors): SVDQuant quantized NVFP4 Qwen-Image model with rank 32. For users with Blackwell GPUs (50-series).
- [`svdq-fp4_r128-qwen-image.safetensors`](./svdq-fp4_r128-qwen-image.safetensors): SVDQuant quantized NVFP4 Qwen-Image model with rank 128. For users with Blackwell GPUs (50-series). It offers better quality than the rank 32 model, but it is slower.


### Model Sources

- **Inference Engine:** [nunchaku](https://github.com/nunchaku-tech/nunchaku)
- **Quantization Library:** [deepcompressor](https://github.com/nunchaku-tech/deepcompressor)
- **Paper:** [SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models](http://arxiv.org/abs/2411.05007)
- **Demo:** [svdquant.mit.edu](https://svdquant.mit.edu)

## Usage

- Diffusers Usage: See [qwen-image.py](https://github.com/nunchaku-tech/nunchaku/blob/main/examples/v1/qwen-image.py).
- ComfyUI Usage: See [nunchaku-qwen-image.json](https://nunchaku.tech/docs/ComfyUI-nunchaku/workflows/qwenimage.html#nunchaku-qwen-image-json).

## Performance

![performance](https://huggingface.co/datasets/nunchaku-tech/cdn/resolve/main/nunchaku/assets/efficiency.jpg)

## Citation

```bibtex
@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}
}
```