Welcome

Lightdrift is an image search engine built to be called by software, not browsed by people. Describe the image you need in text and get ranked, usable images back. Every result carries a machine-readable rights answer, so your agent knows what it can do with the file before it uses it. Lightdrift hosts Creative Commons-licensed images on its own servers and returns file URLs with source, license, and attribution details. Retrieval only: no generation, no checkout, no marketplace. One API, billed per query.

​

What it does

• Text search — natural-language intent over the captioned image index.
• Similar-image expansion — start from any result and find more like it.
• Hard filters — commercial use, license, source, dimensions, orientation, format, year.
• Rights on every result — license, permission flags, and a ready-to-use attribution string, as declared by the source.

​

How search works

Default photo search uses Voyage multimodal embeddings built from each image and its caption, followed by text reranking over the captions. Search by subject, recognizable names, visible text, aesthetics, mood, composition, or intended use. The collection includes stock photography and specialist sources; coverage varies by subject. Queries accept text, an uploaded image, a public image URL, or an image plus text. Existing indexed asset IDs also support similar-image lookup. All search modes use the captioned Voyage index. Design partners help guide which subjects and sources we add next. Negative constraints and additional design-direction filters remain on the roadmap.

​

Start here

​

Collection size

The broader Lightdrift collection contains about 2 million images and icons. The current Voyage search index contains 1,852,370 captioned images at launch; icons are outside that index.