Dataset · Spawning
34.7M image-caption pairs from PD12M and Megalith-CC0
Spawning/pd-extended
Massive collection of images with synthetic English captions and CLIP-L14 embeddings for training generative and multimodal models.
What's inside
Contains around 34.7 million English image-text pairs. The data combines the PD12M and Megalith-CC0 datasets while preserving original licenses and source identifiers. Each record includes metadata (dimensions, MIME type) and precomputed CLIP-L14 vector representations.
What you can do with it
- Training or fine-tuning text-to-image models (such as Diffusion or LoRA) on cleanly licensed data
- Training multimodal encoders and text-image matching models
- Fast semantic search and image filtering using precomputed CLIP-L14 embeddings
- Languages
- English
- Size
- 10M–100M rows
- Downloads this month
- 26K
- Likes
- 5
How to load it
from datasets import load_dataset
ds = load_dataset("Spawning/pd-extended")