Dataset Card for PaintBerri

Dataset Summary

This dataset contains hand-drawn artwork collected from PaintBerri. The dataset includes images along with associated metadata such as publication dates, titles, descriptions, and dimensions.

Languages

The dataset is monolingual:

English (en): All image descriptions and metadata are primarily in English

Dataset Structure

Data Files

The dataset consists of:

Image files (stored across multiple ZIP files in the "images" directory, with approximately 1,000 images in each archive)
Corresponding metadata in JSONL format containing information about the images, including URLs, titles, dimensions, and publication dates
An archive index CSV file mapping image IDs to their respective archive files

Data Fields

image: URL of the full-sized image
published: Timestamp when the image was first published (UTC)
modified: Timestamp when the image was last modified (UTC)
description: Description of the image provided by the creator (may be empty)
title: Title of the image
is_nsfw: Boolean flag indicating whether the image contains not-safe-for-work content
thumbnail: URL of the thumbnail version of the image
height: Height of the image in pixels
width: Width of the image in pixels
creator_id: ID of the image creator (may be null)
short_id: Short unique identifier for the image

Data Splits

All images and metadata are in a single split with 68,860 entries.

Additional Information

Dataset Structure Supplemental

For easier access to individual images, an archive index file is provided:

archive_index.csv.zst: Maps each image file to its containing ZIP archive (format: "archive_name.zip","image_file_id.jpg")