Amazon Nova Multimodal Embeddings
Nova Multimodal Embeddings maps several content types into one vector space. The Bedrock model card lists it as Active.
What the release provides
The model identifier includes nova-2 even though the displayed model name does not. Its purpose is retrieval and similarity, with vector output rather than generated prose.Amazon Nova Multimodal Embeddings Bedrock model card
How to use it well
Practical guidance. Use the same embedding model and configuration for both the stored collection and incoming queries. Preserve document, frame and audio-segment identifiers so each match can be opened as evidence. Compare retrieval quality across each modality you actually use.
Input types and segment limits
The Nova guide supports text, document images, images, video and audio in one space. It gives an 8K-token text limit or 30-second video and audio segments, and supports synchronous and asynchronous requests.Nova multimodal embedding guide
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | amazon.nova-2-multimodal-embeddings-v1:0 |