Unsplash open-sourced its photo hoard for researchers

What it does

The project provides documentation and loading recipes for two static snapshots of the Unsplash platform. The Lite dataset holds roughly 25,000 nature-themed photos, 25,000 keywords, and one million search records in a ~1 GB package cleared for commercial use. The Full dataset exceeds 6.5 million photos, one million keywords, and 160 million searches, but it weighs ~80 GB raw and is restricted to non-commercial research behind an access request.

The interesting bit

Most image collections label what is in the frame; this one also includes what people typed to find it. The pairing of real search queries with photographer metadata offers a rare map of the gap between human intent and actual pixels.

Key highlights

• Dual-tier access: a small commercial-friendly Lite set and a massive non-commercial Full set
• Over 160 million real search queries paired with photos and keywords
• Semantically versioned releases that expand fields and imagery over time
• Loading examples for PostgreSQL and Python environments
• Explicit terms: research only, no redistribution of the images themselves

Caveats

• The repository is documentation and glue code; the actual data lives behind external downloads and an access request wall
• Lite photos are nature-themed only, so they do not represent the full Unsplash catalog breadth
• Full dataset access requires agreeing to non-commercial terms and a manual request

Verdict

Researchers probing image-text alignment, keyword semantics, or search intent will find the query data as valuable as the photos. Commercial builders and anyone needing a redistributable corpus should look elsewhere—ideally at the Unsplash API.