Loading editorial discovery…
Loading editorial discovery…
Semantic Search
You don't need to know the artist or the title. Describe what you remember (a mood, a scene, a colour, a feeling) and Retrievals finds it across 68,000 works in the National Gallery of Art's open collection.
Describe what you're looking for
Retrievals uses a multimodal vision-language model to embed every artwork in the NGA collection as a vector, a mathematical representation of its visual and semantic content. When you type a description, your words are embedded the same way and compared against all 68,000 vectors simultaneously.
The result is search that understands meaning, not just keywords. “Stormy sea at dusk” finds Turner and van Ruisdael without either name appearing in your query. “Woman in yellow light reading” finds Vermeer.
Most museum search boxes were built for lookup: artist name, title fragment, accession number, date range. If the words you remember don't overlap with catalogue vocabulary, keyword search returns nothing useful even when the work is sitting in the collection. Description search removes that requirement.
Identify a Painting You Cannot Name
Work out what a painting is from what you can see. Describe the subject, setting, and mood to search 68,000 National Gallery of Art works by meaning, then check the catalogue record.
Find Similar Paintings and Artworks
Find artworks that look and feel like one you already know. Search 68,000 National Gallery of Art works by visual and semantic similarity rather than by artist or period.