Skip to content

Dataset · Spawning

12 million public domain image-text pairs

Spawning/pd12m-full

A ready-to-stream version of the PD12M corpus in webdataset format for training generative and multimodal models.

What's inside

This is a downloaded and repackaged tar-archive variant in webdataset format of the open Spawning/PD12M dataset with permission from its authors. It contains 10M to 100M pairs of English images and text descriptions. The data is organized for direct streaming via URL without full pre-downloading.

What you can do with it

  • Pre-training and fine-tuning text-to-image generation models
  • Training multimodal encoders like CLIP on copyright-cleaned data
Languages
English
Size
10M–100M rows
Downloads this month
29.1K
Likes
21

How to load it

from datasets import load_dataset

ds = load_dataset("Spawning/pd12m-full")