Amazon Nova
Amazon Nova is a new generation of state-of-the-art foundation models from Amazon that deliver a compelling combination of accuracy, speed, and cost efficiency. Amazon Nova models are accessible through Amazon Bedrock and support text, image, video, speech understanding and generation across a range of model types: Nova Premier (1M context), Nova Pro, Nova Lite, Nova Micro (text-only), Nova Canvas (image generation), Nova Reel (video generation), and Nova Sonic (speech).
APIs
Amazon Nova API
The Amazon Nova API provides programmatic access to Amazon Nova foundation models through Amazon Bedrock for text, image, and video generation, understanding, and reasoning task...
Features
Seven specialized models: Nova Premier (1M context), Nova Pro, Nova Lite, Nova Micro (text), Nova Canvas (image), Nova Reel (video), Nova Sonic (speech).
Supports text, images, video, documents (PDF, CSV, DOCX, XLS, HTML), and speech as input modalities.
Up to 1 million token context window in Nova Premier; 300k tokens in Nova Pro and Lite; 128k in Nova Micro.
All understanding models support streaming for real-time interactive applications.
All understanding models support batch processing for high-volume offline workloads.
Nova Pro, Lite, and Micro support fine-tuning for domain-specific customization.
Nova Premier can serve as a teacher model for distillation into Pro, Lite, and Micro.
Natively integrated with Amazon Bedrock Knowledge Bases, Agents, Guardrails, Evaluations, and Prompt Flows.
Use Cases
Build conversational AI applications with long context awareness using Nova Pro, Lite, or Micro.
Enhance knowledge retrieval accuracy by combining Nova models with Bedrock Knowledge Bases.
Build autonomous AI agents that reason and act using Nova models with Bedrock Agents.
Analyze video content and complex documents (PDF, DOCX, XLS) with Nova Pro and Premier.
Generate and edit high-quality images programmatically with Amazon Nova Canvas.
Create short video clips from text or image prompts using Amazon Nova Reel.
Build voice-enabled customer service and assistant applications with Nova Sonic speech model.
Automate UI interactions and screen navigation workflows using Nova vision capabilities.
Integrations
Primary access method; all Nova models are served through the Bedrock InvokeModel and InvokeModelWithResponseStream APIs.
Connect Nova models to structured and unstructured data sources for RAG applications.
Orchestrate multi-step agentic workflows with tool use and memory using Nova as the reasoning engine.
Apply safety guardrails to Nova Premier, Pro, and Lite model outputs for content filtering.
Build visual prompt chaining workflows connecting Nova models with other services.
Evaluate Nova model performance on custom benchmarks and safety criteria.
Store and access training data, batch inference inputs/outputs, and generated media artifacts.
Control access to Nova model invocations through fine-grained IAM policies and Bedrock model access settings.