REPA-E/e2e-sd3.5-vae
<h1 align="center"> 🚀 REPA-E <em>for</em> T2I </h1>
<p align="center"> <em>End-to-End Tuned VAEs for Supercharging Text-to-Image Diffusion Transformers</em> </p>
<p align="center"> <a href="https://End2End-Diffusion.github.io/repa-e-t2i">🌐 Project Page</a>   <a href="https://huggingface.co/REPA-E/models">🤗 Models</a>   <a href="https://arxiv.org/abs/2504.10483">📃 Paper</a>   <br><br> <!-- <a href="https://paperswithcode.com/sota/image-generation-on-imagenet-256x256?p=repa-e-unlocking-vae-for-end-to-end-tuning-of"><img src="https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/repa-e-unlocking-vae-for-end-to-end-tuning-of/image-generation-on-imagenet-256x256" alt="PWC"></a> --> </p>
<!-- <p align="center"> <a href="https://scholar.google.com.au/citations?user=GQzvqS4AAAAJ" target="blank">Xingjian Leng</a><sup>1,2*</sup>   <b>·</b>   <a href="https://1jsingh.github.io/" target="blank">Jaskirat Singh</a><sup>1</sup>   <b>·</b>   <a href="https://rynmurdock.github.io/" target="blank">Ryan Murdock</a><sup>2</sup>   <b>·</b>   <a href="https://www.ethansmith2000.com/" target="blank">Ethan Smith</a><sup>2</sup>   <b>·</b>   <a href="https://xiaoyang-rebecca.github.io/cv/" target="blank">Rebecca Li</a><sup>2</sup>   <b>·</b>   <a href="https://www.sainingxie.com/" target="blank">Saining Xie</a><sup>3</sup>  <b>·</b>   <a href="https://zheng-lab-anu.github.io/" target="_blank">Liang Zheng</a><sup>1</sup>  </p> -->
<!-- <p align="center"> <sup>1</sup> Australian National University   <sup>2</sup>Canva   <sup>3</sup>New York University   <br> <sub><sup>*</sup>Done during internship at Canva  </sub> </p>
<p align="center"> <a href="https://arxiv.org/abs/2504.10483" target="blank">📄 REPA-E Paper</a>   |   <a href="https://end2end-diffusion.github.io/repa-e-t2i/" target="blank">🌐 Blog Post</a>   |   <a href="https://huggingface.co/REPA-E" target="_blank">🤗 Models</a> </p> -->
🚀 Overall
<p> We present REPA-E for T2I, a family of end-to-end tuned VAEs designed to supercharge text-to-image generation training. These models consistently outperform SD-3.5-VAE across all benchmarks (COCO-30K, DPG-Bench, GenAI-Bench, GenEval, and MJHQ-30K) without requiring any additional representation alignment losses. </p>
<p> For training, we adopt the <a href="https://github.com/End2End-Diffusion/REPA-E" target="blank"><strong>official REPA-E training code</strong></a> to optimize the <a href="https://huggingface.co/stabilityai/stable-diffusion-3.5-large" target="blank">SD-3.5-VAE</a> for <strong>80 epochs</strong> with a batch size of <strong>256</strong> on the <strong>ImageNet-256</strong> dataset. The REPA-E training effectively refines the VAE’s latent-space structure and enables faster convergence in downstream text-to-image latent diffusion model training. </p>
<p> This repository provides <code>diffusers</code>-compatible weights for the <strong>end-to-end trained SD-3.5-VAE</strong>. In addition, we release <strong>end-to-end trained variants</strong> of several other widely used VAEs to facilitate research and integration within text-to-image diffusion frameworks. </p>
⚡️ Quickstart
from diffusers import AutoencoderKL
vae = AutoencoderKL.from_pretrained("REPA-E/e2e-sd3.5-vae").to("cuda")Usevae.encode(...)/vae.decode(...)in your pipeline. (A full example is provided below.)
🧩 End-to-End Trained VAE Releases
📦 Requirements
The following packages are required to load and run the REPA-E VAEs with the diffusers library:
pip install diffusers>=0.33.0
pip install torch>=2.3.1🚀 Example Usage
Below is a minimal example showing how to load and use the REPA-E end-to-end trained SD-3.5-VAE with diffusers:
from io import BytesIO
import requests
from diffusers import AutoencoderKL
import numpy as np
import torch
from PIL import Image
response = requests.get("https://raw.githubusercontent.com/End2End-Diffusion/fuse-dit/main/assets/example.png")
device = "cuda"
image = torch.from_numpy(
np.array(
Image.open(BytesIO(response.content))
)
).permute(2, 0, 1).unsqueeze(0).to(torch.float32) / 127.5 - 1
image = image.to(device)
vae = AutoencoderKL.from_pretrained("REPA-E/e2e-sd3.5-vae").to(device)
with torch.no_grad():
latents = vae.encode(image).latent_dist.sample()
reconstructed = vae.decode(latents).sample