Skip to main content
Learn how you can power your AI agents and multimodal search with Amazon Nova Multimodal Embeddings

Amazon Nova Foundation Models

Optimize your AI applications with powerful, cost-effective foundation models

Try Amazon Nova
Build with Amazon Nova

About Amazon Nova

Amazon Nova models deliver frontier intelligence and industry-leading price-performance. With the most comprehensive suite of customization capabilities for any proprietary model family, organizations can customize Nova to  deliver responses that reflect their industry expertise. Through Amazon Bedrock, organizations can seamlessly build and scale generative AI applications with Nova that are safe, reliable, and cost-effective. Get started today with the Amazon Nova user guide.


See pricing in Amazon Bedrock. 

Models for all the gen AI you need

Nova accelerates AI innovation across modalities and capabilities - including text, image, video, speech, and agentic AI.

Understanding Models

Amazon Nova understanding models are designed to comprehend diverse inputs, including text, images, video, documents, and code. These models excel in powering interactive chat interfaces, Retrieval-Augmented Generation (RAG) systems, agentic applications, video analysis, and UI workflow automation solutions. There are four Amazon Nova understanding models - Nova Micro, Nova Lite, Nova Pro, and Nova Premier - each optimized for different needs while delivering text outputs. Together, they provide users with flexible options to balance capability, accuracy, speed, and cost.

Learn more about understanding models

Creative Content Generation

Amazon Nova offers two models for creative content generation. Amazon Nova Canvas is a high-quality image generation model that offers broad customization and control features for builders in advertising, marketing, entertainment, and retail industries. Amazon Nova Reel is a video generation model designed to democratize creative content generation. It specializes in producing lifestyle videos and b-roll for consumer use cases, making it ideal for ambient content and product videos. Whether you need marketing assets, product visuals, or engaging video content, these models deliver results at accessible price points.

Learn more about creative content generation

Multimodal Embeddings

Amazon Nova Multimodal Embeddings is a state-of-the art embedding model designed to power agentic RAG and semantic search. It is the first unified embedding model that supports text, documents, images, video, and audio through a single model. It maps diverse content types into a unified space with leading accuracy, helping to break down data silos. Organizations can unlock the value of their unstructured data by replacing multiple specialized models with a single unified solution to power seamless search and retrieval across content types.

Learn more about multimodal embeddings

Speech-to-speech

Amazon Nova Sonic is a speech understanding and generation model that delivers natural, human-like conversational AI with low latency and industry-leading price-performance. The model offers fluid dialog handling, adaptive speech response, content moderation, API actuation, and RAG-based knowledge grounding while offering highly adaptive and engaging user experience. 

Learn more about speech-to-speech

Building Computer Agents

Amazon Nova Act is a new AI model trained to perform actions autonomously within a web browser—such as clicking elements, selecting dropdowns, searching, and answering questions about the screen. With the Amazon Nova Act SDK, available as a research preview, you can build reliable agents that break down complex browser workflows into smaller, dependable commands.

Learn more about building broswer agents

Compare Amazon Nova models

Text to Text

Amazon Nova Micro

A text-only model with the lowest latency at a very low cost. It has a max token capacity of 128k, supports 200+ languages, and can be fine-tuned.

Learn more
Understanding

Nova Lite

A multimodal model that processes image, video, and text with lightning-fast speed. Max token of 300k, supports 200+ languages, and can be fine-tuned.

Learn more
Understanding

Nova Pro

Our best combination of accuracy, speed, and cost for a wide range of tasks. It has a max token of 300k, supports 200+ languages, and can be fine-tuned.

Learn more
Understanding

Nova Premier

Our most capable model designed for complex tasks and serves as a teacher for model distillation. It has a max token of 1M and supports 200+ languages.

Learn more
Creative

Nova Canvas

An image generation model with customization and control features. It accepts up to 1,024 input characters, supports English, and fine-tuning.

Learn more
Creative

Nova Reel

A video generation model from text or image inputs. It accepts up to 512 input characters and supports English language.

Learn more
Speech

Nova Sonic

Real-time, human-like voice conversations with natural interactions. It has a max token of 300k and supports English (US, UK), Spanish, French, Italian, and German.

Learn more
Agents

Nova Act

Available through an SDK, this model is trained to perform actions autonomously within a web browser, including clicking elements, searching, and answering questions.

Learn more
Embeddings

Nova Multimodal Embeddings

A unified, state-of-the-art multimodal embeddings model for agentic RAG and semantic search applications.

Learn more

Browse more customer stories

Read more

FAQ

To start building generative AI applications with Amazon Nova, you can access the models through an API using Amazon Bedrock. Learn more about building applications with Amazon Nova. 

Amazon Nova understanding models support 200+ languages, and are optimized for 15 languages: English, German, Spanish, French, Italian, Japanese, Korean, Arabic, Simplified Chinese, Russian, Hindi, Portuguese, Dutch, Turkish, and Hebrew.

Common use cases with Amazon Nova are aligned to understanding behaviors, including chat assistants, summarization, and intelligent document processing. In addition, customers are rapidly generating images, videos, and speech. 

Nova understanding models support text and vision fine-tuning, to be able to understand and generate custom text, images, and video content. Customers can fine-tune Amazon Nova on their proprietary data – including unique datasets and brand characteristics– to generate fully customized outputs and creative content that align with your specific requirements and style guidelines. 

  • Amazon Nova comes with built-in content moderation filters to limit generation of harmful content and misuse.
  • Amazon Nova Canvas, Nova Reel, and Nova Sonic come with invisible watermarks embedded within the visual or audio content. These can be used for traceability and accountability for AI generated content, and to reduce spread of disinformation.
  • AWS offers uncapped intellectual property (IP) indemnity coverage for outputs of Amazon Nova models – protecting customers from third-party claims alleging IP infringement or misappropriation (including copyright claims) by the outputs generated by these Amazon Nova models.
  • Amazon Nova integrates seamlessly with Amazon Bedrock, leveraging advanced automated reasoning in Bedrock Guardrails to establish and enforce robust safety boundaries. This integration helps ensure that AI-generated outputs consistently comply with organizational policies, security protocols, and regulatory requirements while maintaining operational efficiency.