Skip to content

Dataset · wanng

Midjourney v5 prompt and image pairs

wanng/midjourney-v5-202304-clean

1.7M Midjourney v5 prompt and image pairs from April 2023 in a cleaned unofficial dump.

What's inside

An unofficial dataset collected from Midjourney v5 for April 2023 containing 1,701,420 prompt-image pairs. After cleaning, it is split into two parquet files: 1,255,812 pairs with grid images and 445,608 pairs with upscale versions. Languages specified are en and fr.

What you can do with it

  • Training text-to-image models or LoRA on Midjourney prompts
  • Fine-tuning image-to-text models for image captioning
  • Searching and analyzing prompt-image alignment
Tasks
Text → image, Image → text
Languages
English, French
Size
1M–10M rows
Downloads this month
149
Likes
90

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("wanng/midjourney-v5-202304-clean")