What is Amazon Nova Multimodal Embeddings?
Amazon Nova Multimodal Embeddings is Amazon’s embedding model that converts content into vector representations for search and retrieval use cases. The model card lists text, image, video, and audio as input modalities, with embeddings as the output.
The announcement describes the model as an embedding model for agentic RAG and semantic search, and as the first unified model enabling cross-modal retrieval across different content types. Per the model card, the model launched on October 28, 2025.
Core Features
- One model for several modalities: Text, image, video, and audio are mapped into the same vector space
- Cross-modal retrieval: Per the announcement, content of different types can be searched together
- Segmentation: Inputs up to 8K tokens and video and audio segments up to 30 seconds; larger files are segmented
- Selectable output dimensions: Per the announcement, multiple dimensions let you balance accuracy against storage and computational cost
- Synchronous and asynchronous processing: Near real-time for individual requests, asynchronous for high volumes
- Delivered through Amazon Bedrock: Invoked on the
bedrock-runtimeendpoint with model IDamazon.nova-2-multimodal-embeddings-v1:0
Typical Use Cases
Agentic RAG: Make knowledge sources of different media types searchable for agents.
Semantic search: Answer queries by meaning rather than by exact keyword matches.
Media archives: Index image, video, and audio holdings together with text documents.
Building vector indexes: Generate embeddings and store them in a vector database or vector index.
Benefits
- One model instead of separate embedding models per media type
- Cross-modal search across text, image, video, and audio
- Selectable output dimensions to control storage and compute effort
- Synchronous and asynchronous processing for different load profiles
- Integrated into Amazon Bedrock and therefore into existing AWS access and billing models
Note on Data Residency
The model card currently lists only the US East (N. Virginia) and AWS GovCloud (US-West) Regions for this model, and geo and global cross-Region inference are not supported. For workloads that require processing within the EU, the model is not suitable in this form. Check regional availability in the official documentation before using it.
Integration with innFactory
As an AWS Reseller, innFactory supports you with Amazon Nova Multimodal Embeddings: assessing the model against your retrieval requirements, designing segmentation and output dimensions, building vector indexes and RAG pipelines on AWS, and weighing the regional and data protection implications against models with EU availability.
Typical Use Cases
Technical Specifications
Frequently Asked Questions
What is Amazon Nova Multimodal Embeddings?
Per the model card, Amazon Nova Multimodal Embeddings is Amazon's embedding model that converts text, images, and video into vector representations for search and retrieval use cases. The model card lists text, image, video, and audio as input modalities, with embeddings as the output. The announcement describes it as an embedding model for agentic RAG and semantic search and as the first unified model enabling cross-modal retrieval.
Which input sizes are supported?
Per the announcement, the model supports inputs up to 8K tokens and video and audio segments up to 30 seconds. Larger files can be segmented.
In which Regions is the model available?
The model card lists in-Region inference in US East (N. Virginia, us-east-1) and AWS GovCloud (US-West, us-gov-west-1). Geo cross-Region and global cross-Region inference are not supported. No EU region is currently listed there.
How is the model invoked?
Per the model card, the model is invoked through the bedrock-runtime endpoint using model ID amazon.nova-2-multimodal-embeddings-v1:0 and the StartAsyncInvoke operation. The announcement additionally describes synchronous APIs for near real-time requests and asynchronous APIs for high-volume processing.
Which Bedrock features are not supported?
Per the model card, guardrails, knowledge bases, agents, model evaluation, response streaming, intelligent prompt routing, prompt management, flows, and count tokens are not supported for this model, among others. The available service tier is Standard.
What does it cost?
Billing runs through Amazon Bedrock. Refer to the Amazon Bedrock pricing page for binding rates.
Note: All product information on this page has been compiled with care, but is provided without guarantee and may be outdated or incomplete. Cloud services evolve rapidly — features, pricing, SLAs, and availability change frequently. Authoritative and up-to-date information can only be found on the official product page of AWS (official documentation). This page does not represent an offer by AWS.