README.md

---
license: apache-2.0
language:
- en
- zh
library_name: diffusers
pipeline_tag: text-to-image
base_model:
- Qwen/Qwen-Image
---

<h1 align="center">TwinFlow: Realizing One-step Generation on Large Models with Self-adversarial Flows</h1>

<div align="center">

[![Project Page](https://img.shields.io/badge/Project%20Page-133399.svg?logo=homepage)](https://zhenglin-cheng.com/twinflow)&#160;
[![Hugging Face](https://img.shields.io/badge/%F0%9F%A4%97%20Model-TwinFlow--Qwen--Image-yellow)](https://huggingface.co/inclusionAI/TwinFlow)&#160;
[![Github Repo](https://img.shields.io/badge/inclusionAI%2FTwinFlow-black?logo=github)](https://github.com/inclusionAI/TwinFlow)&#160;
<a href="https://arxiv.org/abs/2512.05150" target="_blank"><img src="https://img.shields.io/badge/Paper-b5212f.svg?logo=arxiv" height="21px"></a>


</div>

## News

- We release **TwinFlow-Qwen-Image-v1.0**! And we are also working on **Z-Image-Turbo to make it more faster**!

## TwinFlow

Checkout 2-NFE visualization of TwinFlow-Qwen-Image 👇

![](https://raw.githubusercontent.com/inclusionAI/TwinFlow/refs/heads/main/assets/demo.jpg)


### Overview

We introduce TwinFlow, a framework that realizes high-quality 1-step and few-step generation without the pipeline bloat.

Instead of relying on external discriminators or frozen teachers, TwinFlow creates an internal "twin trajectory". By extending the time interval to $t\in[−1,1]$, we utilize the negative time branch to map noise to "fake" data, creating a self-adversarial signal directly within the model.

Then, the model can rectify itself by minimizing the difference of the velocity fields between real trajectory and fake trajectory, i.e. the $\Delta_\mathrm{v}$. The rectification performs distribution matching as velocity matching, which gradually transforms the model into a 1-step/few-step generator.

![](https://raw.githubusercontent.com/inclusionAI/TwinFlow/refs/heads/main/assets/twinflow.png)

Key Advantages:
- **One-model Simplicity.** We eliminate the need for any auxiliary networks. The model learns to rectify its own flow field, acting as the generator, fake/real score. No extra GPU memory is wasted on frozen teachers or discriminators during training.
- **Scalability on Large Models.** TwinFlow is **easy to scale on 20B full-parameter training** due to the one-model simplicity. In contrast, methods like VSD, SiD, and DMD/DMD2 require maintaining three separate models for distillation, which not only significantly increases memory consumption—often leading OOM, but also introduces substantial complexity when scaling to large-scale training regimes.

### Inference Demo

Install the latest diffusers:

```bash
pip install git+https://github.com/huggingface/diffusers
```

Run inference demo `inference.py`:

```python
python inference.py
```

## Citation

```bibtex
@article{cheng2025twinflow,
  title={TwinFlow: Realizing One-step Generation on Large Models with Self-adversarial Flows},
  author={Cheng, Zhenglin and Sun, Peng and Li, Jianguo and Lin, Tao},
  journal={arXiv preprint arXiv:2512.05150},
  year={2025}
}
```

## Acknowledgement

TwinFlow is built upon [RCGM](https://github.com/LINs-lab/RCGM) and [UCGM](https://github.com/LINs-lab/UCGM), with much support from [InclusionAI](https://github.com/inclusionAI).