🐦 Twitter Post Details

Viewing enriched Twitter post

@YichuanM

The web was never meant to be flattened into text. Yet most web RAG systems start by parsing HTML --- a complex and lossy process. šŸ”„ Introducing PixelRAG: the first RAG system that retrieves and reads 30M+ web pages as pixels. Instead of extracting text, PixelRAG retrieves screenshots and lets a VLM read them directly. PixelRAG not only preserves visual information, but also outperforms text-based RAG on text-only QA benchmarks by +18.1%. Why? (1) HTML-to-text conversion often discards layout, structure, tables, and other useful signals. (2) We continued pretraining a VLM on web page screenshots and turned it into a surprisingly strong visual retriever. (3) Recent VLMs are remarkably good at understanding web pages, often with better accuracy and token efficiency than text-only pipelines. Takeaway: HTML parsing may be one of the biggest self-inflicted bottlenecks in web RAG. Demo below šŸ‘‡ Code: https://t.co/ssDF0nnVwZ Paper: https://t.co/OIpQ26Vb8H Playground: https://t.co/UdzM7GQmu3

Media 2
Media 3

šŸ“Š Media Metadata

{
  "media": [
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2064756382386266136/media_0.mp4",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2064756382386266136/media_0.mp4",
      "type": "video",
      "filename": "media_0.mp4"
    },
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2064756382386266136/media_1.jpg",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2064756382386266136/media_1.jpg",
      "type": "photo",
      "filename": "media_1.jpg"
    },
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2064756382386266136/media_2.jpg",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2064756382386266136/media_2.jpg",
      "type": "photo",
      "filename": "media_2.jpg"
    }
  ],
  "processed_at": "2026-06-11T22:38:32.705877",
  "pipeline_version": "2.0"
}

šŸ”§ Raw API Response

{
  "type": "tweet",
  "id": "2064756382386266136",
  "url": "https://x.com/YichuanM/status/2064756382386266136",
  "twitterUrl": "https://twitter.com/YichuanM/status/2064756382386266136",
  "text": "The web was never meant to be flattened into text.\n\nYet most web RAG systems start by parsing HTML --- a complex and lossy process.\n\nšŸ”„ Introducing PixelRAG: the first RAG system that retrieves and reads 30M+ web pages as pixels.\n\nInstead of extracting text, PixelRAG retrieves screenshots and lets a VLM read them directly.\n\nPixelRAG not only preserves visual information, but also outperforms text-based RAG on text-only QA benchmarks by +18.1%.\n\nWhy?\n(1) HTML-to-text conversion often discards layout, structure, tables, and other useful signals.\n(2) We continued pretraining a VLM on web page screenshots and turned it into a surprisingly strong visual retriever.\n(3) Recent VLMs are remarkably good at understanding web pages, often with better accuracy and token efficiency than text-only pipelines.\n\nTakeaway: HTML parsing may be one of the biggest self-inflicted bottlenecks in web RAG.\n\nDemo below šŸ‘‡\n\nCode: https://t.co/ssDF0nnVwZ\nPaper: https://t.co/OIpQ26Vb8H\nPlayground: https://t.co/UdzM7GQmu3",
  "source": "Twitter for iPhone",
  "retweetCount": 96,
  "replyCount": 18,
  "likeCount": 486,
  "quoteCount": 4,
  "viewCount": 46441,
  "createdAt": "Wed Jun 10 17:07:37 +0000 2026",
  "lang": "en",
  "bookmarkCount": 454,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "2064756382386266136",
  "displayTextRange": [
    0,
    276
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "YichuanM",
    "url": "https://x.com/YichuanM",
    "twitterUrl": "https://twitter.com/YichuanM",
    "id": "1389164045782028288",
    "name": "Yichuan Wang",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/1717971585292775424/5D7hwdDV_normal.jpg",
    "coverPicture": "",
    "description": "",
    "location": "",
    "followers": 2152,
    "following": 2490,
    "status": "",
    "canDm": false,
    "canMediaTag": true,
    "createdAt": "Mon May 03 10:25:00 +0000 2021",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {}
    },
    "fastFollowersCount": 0,
    "favouritesCount": 307,
    "hasCustomTimelines": true,
    "isTranslator": false,
    "mediaCount": 19,
    "statusesCount": 429,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [
      "2064756382386266136"
    ],
    "profile_bio": {
      "description": "2nd-year eecs phd @ uc berkeley (skylab, bair).\nmlsys.\ncreator of leann: https://t.co/YWTe2nI530\nsjtu acm class alum.",
      "entities": {
        "description": {
          "urls": [
            {
              "display_url": "github.com/yichuan-w/leann",
              "expanded_url": "http://github.com/yichuan-w/leann",
              "indices": [
                73,
                96
              ],
              "url": "https://t.co/YWTe2nI530"
            }
          ]
        },
        "url": {
          "urls": [
            {
              "display_url": "yichuan-w.github.io",
              "expanded_url": "https://yichuan-w.github.io/",
              "indices": [
                0,
                23
              ],
              "url": "https://t.co/SJGx43LZlz"
            }
          ]
        }
      }
    },
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "additional_media_info": {
          "monetizable": true
        },
        "allow_download_status": {
          "allow_download": true
        },
        "display_url": "pic.twitter.com/VDVIRjglfB",
        "expanded_url": "https://twitter.com/YichuanM/status/2064756382386266136/video/1",
        "ext_master_playlist_only": [],
        "ext_media_availability": {
          "status": "Available"
        },
        "ext_playlists": [],
        "id_str": "2064754542718431232",
        "indices": [
          277,
          300
        ],
        "media_key": "13_2064754542718431232",
        "media_results": {
          "id": "QXBpTWVkaWFSZXN1bHRzOgwABAoAARyne1ErGzAAAAA=",
          "result": {
            "__typename": "ApiMedia",
            "id": "QXBpTWVkaWE6DAAECgABHKd7USsbMAAAAA==",
            "media_key": "13_2064754542718431232"
          }
        },
        "media_url_https": "https://pbs.twimg.com/amplify_video_thumb/2064754542718431232/img/9Lw_FElb2dLa5sDe.jpg",
        "original_info": {
          "focus_rects": [],
          "height": 720,
          "width": 1280
        },
        "sizes": {
          "large": {
            "h": 720,
            "w": 1280
          }
        },
        "type": "video",
        "url": "https://t.co/VDVIRjglfB",
        "video_info": {
          "aspect_ratio": [
            16,
            9
          ],
          "duration_millis": 55700,
          "variants": [
            {
              "content_type": "application/x-mpegURL",
              "url": "https://video.twimg.com/amplify_video/2064754542718431232/pl/6bL4KVX_EQQIt41U.m3u8?tag=27&v=cfc"
            },
            {
              "bitrate": 256000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2064754542718431232/vid/avc1/480x270/cgoChPv1a6zNBE9l.mp4?tag=27"
            },
            {
              "bitrate": 832000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2064754542718431232/vid/avc1/640x360/OysW7b1JTRfZYhRx.mp4?tag=27"
            },
            {
              "bitrate": 2176000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2064754542718431232/vid/avc1/1280x720/f0eANNvJnwFzvEH6.mp4?tag=27"
            }
          ]
        }
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {
    "hashtags": [],
    "symbols": [],
    "timestamps": [],
    "urls": [
      {
        "display_url": "github.com/StarTrail-org/…",
        "expanded_url": "https://github.com/StarTrail-org/PixelRAG",
        "indices": [
          915,
          938
        ],
        "url": "https://t.co/ssDF0nnVwZ"
      },
      {
        "display_url": "github.com/StarTrail-org/…",
        "expanded_url": "https://github.com/StarTrail-org/PixelRAG/blob/main/assets/pixelrag-paper.pdf",
        "indices": [
          946,
          969
        ],
        "url": "https://t.co/OIpQ26Vb8H"
      },
      {
        "display_url": "pixelrag.ai",
        "expanded_url": "https://pixelrag.ai/",
        "indices": [
          982,
          1005
        ],
        "url": "https://t.co/UdzM7GQmu3"
      }
    ],
    "user_mentions": []
  },
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "isLimitedReply": false,
  "communityInfo": null,
  "article": null
}