Datasets:

fromziro
/

Pixel2DOM-Raw

license: apache-2.0
task_categories:
  - image-to-text
  - text-to-image
language:
  - en
  - es
  - ja
  - fr
  - zh
  - ru
  - it
  - pt
  - nl
  - tr
  - pl
  - de
  - ga
  - vi
tags:
  - commoncrawl
  - image2text
  - vlm
  - text2image
  - html2image
  - image2html
  - webpages
  - fromzero
pretty_name: Pixel2DOM
size_categories:
  - 1K<n<10K

NOTE: THIS DATASET IS STILL IN PRODUCTION

Pixel2DOM-Raw

Pixel2DOM-Raw is a large-scale dataset of webpage screenshots paired with their corresponding HTML.

We downloaded WARC files from the Common Crawl CC-MAIN-2026-30 snapshot, rendered each page with Playwright, and captured a screenshot of the rendered webpage.

This is the raw, unfiltered release. The source data comes directly from the public web, so it may contain disturbing, offensive, explicit, malicious or otherwise undesirable content. Furthermore, users are responsible for filtering the dataset according to their own requirements before using it in downstream applications or model training.