---
title: "new paper: iREPA diffusion models are a renderer of their underlying representations. with this new..."
canonical: "https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/"
date: "2025-12-15"
updated: "2026-11-05"
source: "X：谢赛宁 (@sainingxie)"
source_url: "https://x.com/sainingxie/status/2000709656491286870"
language: "zh"
topics: []
license: "https://new.nat6.net/sources/"
attribution: "Content is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back; original copyright remains with the original publishers and authors."
---

# new paper: iREPA diffusion models are a renderer of their underlying representations. with this new...

- 发布日期 / Published: 2025-12-15
- 更新时间 / Updated: 2026-11-05
- 分类 / Category: 其他
- 来源 / Source: X：谢赛宁 (@sainingxie) — [原文 / Original](https://x.com/sainingxie/status/2000709656491286870)
- AIHOT: [条目页 / Item](https://aihot.news/items/cmnz6dpf402amsl0fxz3kh21f)
- 页面 / Page: https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/
- 结构化数据 / JSON: https://new.nat6.net/api/v1/articles/cmnz6dpf402amsl0fxz3kh21f.json

- 语言 / Languages: [zh](https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/) · [en](https://new.nat6.net/en/news/cmnz6dpf402amsl0fxz3kh21f/) · [de](https://new.nat6.net/de/news/cmnz6dpf402amsl0fxz3kh21f/) · [fr](https://new.nat6.net/fr/news/cmnz6dpf402amsl0fxz3kh21f/) · [ru](https://new.nat6.net/ru/news/cmnz6dpf402amsl0fxz3kh21f/) · [ja](https://new.nat6.net/ja/news/cmnz6dpf402amsl0fxz3kh21f/) · [es](https://new.nat6.net/es/news/cmnz6dpf402amsl0fxz3kh21f/) · [ko](https://new.nat6.net/ko/news/cmnz6dpf402amsl0fxz3kh21f/) · [vi](https://new.nat6.net/vi/news/cmnz6dpf402amsl0fxz3kh21f/) · [th](https://new.nat6.net/th/news/cmnz6dpf402amsl0fxz3kh21f/) · [id](https://new.nat6.net/id/news/cmnz6dpf402amsl0fxz3kh21f/) · [ms](https://new.nat6.net/ms/news/cmnz6dpf402amsl0fxz3kh21f/) · [fil](https://new.nat6.net/fil/news/cmnz6dpf402amsl0fxz3kh21f/) · [hi](https://new.nat6.net/hi/news/cmnz6dpf402amsl0fxz3kh21f/)

## AI 摘要 / AI summary

新论文：iREPA 扩散模型是其底层表征的渲染器。通过这种新设置，我们能更清楚地洞察这些表征的真正含义。Jas 开始了一场自发的探索，过去三个月我们学到了很多 ps. 这也是我们对一种新型线上"饮水机效应"的小实验，我很喜欢看到这种现象。让我们争论、讨论，然后用真正的努力将其转化为正经科学 [引用 @1jaskiratsingh]：‼️ 表征对生成很重要！但事实证明，我们对表征如何帮助生成的理解一直都是错的 ‼️ 我们之前的想法：（我们错了） ❌ 更大的视觉编码器 → 更好的表征 → 更好的生成 ❌ 更好的全局语义 → 更好的表征 → 更好的生成 结果发现： 🤯 在表征对齐方面，小 20 倍以上的视觉编码器可以达到与更大模型相似或更好的性能 🤯 线性探测准确率约 20%（全局语义的衡量指标）的视觉编码器可以胜过准确率 >80% 的编码器 🤯 即使是 SiFT 和 HoG 这类经典特征也能带来与现代大得多的视觉编码器相媲美的提升 ‼️ 🚨 介绍：什么对表征对齐重要？全局信息还是空间结构 🚨 TL;DR： ✅ 更好的全局语义信息 ≠ 更好的生成 ✅ 空间结构（而非全局语义）驱动表征的生成性能 ✅ 我们提出 iREPA：仅需 3 行代码，强调空间结构迁移，并在 REPA、REPA-E、Meanflow、JiT 等方法上持续提高收敛速度 在 @AdobeResearch 的激动人心的项目，与 @xingjian_leng、@zongze_wu、@LiangZheng_06、@rzhang88、@elishechtman 和 @sainingxie 合作 🙏 对我来说这也是一次特别有趣且独特的经历，在项目的每一步我们都在证明自己的偏见是错误的 😆 还要大力感谢 @YouJiacheng、@ShumingHu 和 @gallabytes，他们在 X 上的评论开启了这一方向的探索 🫡 论文：https://arxiv.org/abs/2512.10794 代码：https://github.com/End2End-Diffusion/iREPA 项目页面：https://end2end-diffusion.github.io/irepa 更多细节见线程：[1/n] 🧵

## 推荐理由 / Why it matters

颠覆认知：小20倍视觉编码器也能驱动高质量生成，空间结构才是关键

## 来源与版权 / Source & copyright

内容基于 AIHOT 公开 API 的精选标题与 AI 摘要，已注明来源并回链；原文版权归原媒体与作者所有。

原文版权归原媒体与作者所有；本站不镜像全文。/ Original copyright remains with the original publishers and authors; this site does not mirror full texts.

## 关于橙子中转站 / About Orange Relay

橙子中转站（pro.nat6.net）是一家优质的 AI API 中转站：价格优惠、质量稳定、模型齐全，支持 Claude、GPT、Gemini 等主流模型，欢迎体验。

[访问橙子中转站](https://pro.nat6.net/?utm_source=orange_news&utm_medium=article_footer&utm_campaign=brand&lang=zh)

## JSON-LD

```json
{
  "@context": "https://schema.org",
  "@type": "NewsArticle",
  "@id": "https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/#article",
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/"
  },
  "url": "https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/",
  "headline": "new paper: iREPA diffusion models are a renderer of their underlying representations. with this new...",
  "description": "新论文：iREPA 扩散模型是其底层表征的渲染器。通过这种新设置，我们能更清楚地洞察这些表征的真正含义。Jas 开始了一场自发的探索，过去三个月我们学到了很多 ps. 这也是我们对一种新型线上\"饮水机效应\"的小实验，我很喜欢看到这种现象。让我们争论、讨论，然后用真正的努力将其转化为正经科学 [引用 @1jaskira…",
  "articleSection": "其他",
  "keywords": "AI, X：谢赛宁 (@sainingxie)",
  "datePublished": "2025-12-15T23:28:48Z",
  "dateModified": "2026-11-05T20:56:41Z",
  "inLanguage": "zh-CN",
  "isAccessibleForFree": true,
  "author": {
    "@type": "Organization",
    "name": "橙子新闻编辑部",
    "url": "https://new.nat6.net/about/"
  },
  "publisher": {
    "@type": "NewsMediaOrganization",
    "@id": "https://new.nat6.net/#org",
    "name": "橙子新闻",
    "url": "https://new.nat6.net/",
    "logo": {
      "@type": "ImageObject",
      "url": "https://new.nat6.net/assets/img/logo-512.png",
      "width": 512,
      "height": 512
    }
  },
  "image": {
    "@type": "ImageObject",
    "url": "https://new.nat6.net/assets/img/og-default.png",
    "width": 1200,
    "height": 630
  },
  "isPartOf": {
    "@id": "https://new.nat6.net/#website"
  },
  "isBasedOn": "https://x.com/sainingxie/status/2000709656491286870",
  "citation": [
    "https://aihot.news/items/cmnz6dpf402amsl0fxz3kh21f",
    "https://x.com/sainingxie/status/2000709656491286870"
  ],
  "license": "https://new.nat6.net/sources/",
  "about": [],
  "mentions": [
    {
      "@type": "Thing",
      "name": "X：谢赛宁 (@sainingxie)"
    }
  ],
  "speakable": {
    "@type": "SpeakableSpecification",
    "cssSelector": [
      "h1",
      ".article-summary"
    ]
  },
  "copyrightHolder": {
    "@type": "Organization",
    "@id": "https://new.nat6.net/#org",
    "name": "橙子新闻",
    "url": "https://new.nat6.net/"
  },
  "creditText": "橙子新闻（AI 摘要）；原文：X：谢赛宁 (@sainingxie)",
  "copyrightNotice": "© 2025 橙子新闻。标题与 AI 摘要基于 AIHOT 公开数据，原文版权归原媒体与作者所有；本站不镜像全文。",
  "copyrightYear": 2025,
  "alternativeHeadline": "new paper: iREPA diffusion models are a renderer of their underlying representations. with this new...",
  "abstract": "颠覆认知：小20倍视觉编码器也能驱动高质量生成，空间结构才是关键",
  "subjectOf": [
    {
      "@type": "DigitalDocument",
      "name": "new paper: iREPA diffusion models are a renderer of their underlying representations. with this new... (Markdown)",
      "url": "https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/index.md",
      "encodingFormat": "text/markdown"
    },
    {
      "@type": "DigitalDocument",
      "name": "new paper: iREPA diffusion models are a renderer of their underlying representations. with this new... (JSON)",
      "url": "https://new.nat6.net/api/v1/articles/cmnz6dpf402amsl0fxz3kh21f.json",
      "encodingFormat": "application/json"
    }
  ]
}
```

## 如何引用 / How to cite

- 引用格式 / Citation format (zh): 橙子新闻.《文章标题》. 发布日期. https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/
- Citation format (en): Orange News. "Article title". Published YYYY-MM-DD. https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/
- 回链要求 / Backlink: 引用请保留「橙子新闻」署名并回链 canonical URL；AI 摘要请注明数据来源 AIHOT；原文版权归原媒体与作者所有。
- Attribution & backlinks: Keep the "Orange News" attribution and link back to the canonical URL; cite AIHOT as the data source for AI summaries; original copyright remains with the original publishers and authors.

### 机器入口 / Machine entry points

- Markdown: https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/index.md
- JSON: https://new.nat6.net/api/v1/articles/cmnz6dpf402amsl0fxz3kh21f.json
- API 清单 / API index: https://new.nat6.net/api/v1/site.json
- Markdown 内容协商 / negotiation: 对 https://new.nat6.net/news/cmnz6dpf402amsl0fxz3kh21f/ 发送 `Accept: text/markdown`
