---
license: mit
pipeline_tag: image-to-image
library_name: diffusers
---

<h1 align="center"> REPA-E: Unlocking VAE for End-to-End Tuning of Latent Diffusion Transformers </h1>

<p align="center">
  [🌐 Project Page](https://end2end-diffusion.github.io) &ensp;
  [📃 Paper](https://arxiv.org/abs/2504.10483) &ensp;
  [🤗 Github](https://github.com/REPA-E/REPA-E)
</p>

![](assets/vis-examples.jpg)

REPA-E enables stable and effective joint training of both the VAE and the diffusion model, significantly accelerating training and improving generation quality. It achieves state-of-the-art FID scores on ImageNet 256×256.  For detailed usage instructions, including environment setup, training, and evaluation, please refer to the [project page](https://end2end-diffusion.github.io) and the [GitHub repository](https://github.com/REPA-E/REPA-E).