---
license: mit
pipeline_tag: unconditional-image-generation
---

# Adversarial Flow Models

This repository contains the official checkpoints for the paper [Adversarial Flow Models](https://huggingface.co/papers/2511.22475).

Adversarial Flow Models is a class of generative models that unifies Adversarial Models and Flow Models. This repository contains the pre-trained ImageNet-256px models described in the paper.

- **GitHub Repository**: [ByteDance-Seed/Adversarial-Flow-Models](https://github.com/ByteDance-Seed/Adversarial-Flow-Models)
- **Paper**: [Adversarial Flow Models](https://huggingface.co/papers/2511.22475)

## Usage

Code and instructions for generation and training are available in the [official GitHub repository](https://github.com/ByteDance-Seed/Adversarial-Flow-Models).

## Repository Content 
* `models/` contains pre-trained ImageNet-256px checkpoints.
* `eval/` contains pre-generated 50k samples for evaluations following ADM npz format.
* `misc/` contains VAE and other checkpoints used in training.

## Citation

```bibtex
@article{lin2025adversarial,
  title={Adversarial Flow Models},
  author={Lin, Shanchuan and Yang, Ceyuan and Lin, Zhijie and Chen, Hao and Fan, Haoqi},
  journal={arXiv preprint arXiv:2511.22475},
  year={2025}
}
```