# U-REPA: Aligning Diffusion U-Nets to ViTs

Source: [U-REPA: Aligning Diffusion U-Nets to ViTs](<https://www.sota2.com/research/paper/68fa77911e209131cea73f90>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-09T05:00:05.325Z

Authors: Yuchuan Tian, Hanting Chen, Mengyu Zheng, Yuchen Liang, Chao Xu, Yunhe Wang

Published: 2025-03-24T07:46:00.000Z

## Abstract

Representation Alignment (REPA) that aligns Diffusion Transformer (DiT) hidden-states with ViT visual encoders has proven highly effective in DiT training, demonstrating superior convergence properties, but it has not been validated on the canonical diffusion U-Net architecture that shows faster convergence compared to DiTs. However, adapting REPA to U-Net architectures presents unique challenges: (1) different block functionalities necessitate revised alignment strategies; (2) spatial-dimension inconsistencies emerge from U-Net's spatial downsampling operations; (3) space gaps between U-Net and ViT hinder the effectiveness of tokenwise alignment. To encounter these challenges, we propose \\textbf{U-REPA}, a representation alignment paradigm that bridges U-Net hidden states and ViT features as follows: Firstly, we propose via observation that due to skip connection, the middle stage of U-Net is the best alignment option. Secondly, we propose upsampling of U-Net features after passing them through MLPs. Thirdly, we observe difficulty when performing tokenwise similarity alignment, and further introduces a manifold loss that regularizes the relative similarity between samples. Experiments indicate that the resulting U-REPA could achieve excellent generation quality and greatly accelerates the convergence speed. With CFG guidance interval, U-REPA could reach $FID\<1.5$ in 200 epochs or 1M iterations on ImageNet 256 $\\times$ 256, and needs only half the total epochs to perform better than REPA under sd-vae-ft-ema. Codes: https://github.com/YuchuanTian/U-REPA

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Image Generation on ImageNet 256x256](<https://www.sota2.com/research/sota/image-generation-on-imagenet-256x256.md>): Image Generation; ImageNet 256x256; IS: not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2503.18414>)

[PDF](<https://arxiv.org/pdf/2503.18414.pdf>)

[Implementation](<https://github.com/YuchuanTian/U-REPA>)
