Switch language한국어
Back to the list

GPIC: A Giant Permissive Image Corpus for Visual Generation

TL;DR AI

Key summary

2 min read
  1. Researchers released GPIC, a 28-trillion-pixel image corpus for visual generation with permissive licensing and vision-language captions.

  2. The dataset includes safety filtering, deduplication, and centrally hosted splits for training, validation, and testing on Hugging Face.

  3. GPIC also comes with a benchmark protocol and a reference pixel-space flow matching baseline.

  4. The release provides a legally usable large-scale image resource for both research and commercial deployment.

Read the original