Nova Series by Amazon
The Nova Series is Amazon’s proprietary family of state-of-the-art foundation models designed for text generation, multimodal reasoning, image and video generation, speech-to-speech interaction, and enterprise-grade AI workflows. Integrated into Amazon Bedrock, the Nova Series combines cutting-edge generative AI capabilities with seamless platform access for experimentation, fine-tuning, and deployment. These models are optimized for speed, cost efficiency, and customization across diverse applications in industries such as media, customer support, and financial analysis.
Current Models in the Nova Series:
Text-Only Models:
Nova Micro:
A lightweight text-only model optimized for low latency and cost efficiency.
Ideal for tasks such as translation, summarization, brainstorming, content classification, interactive chat, simple mathematical reasoning, and basic coding.
Multimodal Models:
Nova Lite:
A cost-efficient multimodal model capable of processing text, images, and videos to generate text outputs.
Handles up to 300k tokens of input and excels at real-time customer interactions, document analysis, visual question answering (VQA), and multimodal fine-tuning.Nova Pro:
A highly capable multimodal model combining accuracy, speed, and cost efficiency for a wide range of tasks.
Optimized for instruction-following workflows, financial document analysis, coding tasks (15k+ lines of code), and agentic AI workflows.Nova Premier (In Development):
The most advanced multimodal model in the series, currently in training with expected release in early 2025.
Designed as a "teacher" model for distilling custom versions of other Nova models.
Creative Content Generation Models:
Nova Canvas:
A state-of-the-art image generation model capable of creating studio-quality visuals from text or image prompts.
Designed for industries like advertising and entertainment with tools for color adjustments and background edits.Nova Reel:
A video generation model that produces short clips from text or image prompts.
Supports natural language commands for manipulating camera angles and movements.
Speech-to-Speech Model:
Nova Sonic:
A real-time speech-to-speech model that enables natural, back-and-forth voice conversations.
Offers expressive English voices, low-latency streaming via API, and supports function calling and RAG-based knowledge grounding through Amazon Bedrock.
Future Model (In Development):
Multimodal-to-Multimodal ("Any-to-Any") Model:
Capable of taking text, images, audio, and video as input to generate outputs in any modality.
Designed to simplify cross-modal workflows such as translation, editing, and interactive storytelling.
Key Features Across the Series:
Multimodal Capabilities: Processes text, images, video, and speech across the series.
Cost Efficiency: Nova Micro offers responses at up to 75% lower cost than comparable models on Amazon Bedrock.
Customization: All models support fine-tuning with proprietary data for tailored enterprise applications.
Distillation & Deployment: Nova Premier will act as a base for training smaller models for specialized use cases.
Example Use Cases:
Automating customer support with real-time multilingual chatbots powered by Nova Micro or Lite.
Generating marketing visuals or video content using Nova Canvas or Reel.
Conducting financial document analysis with high accuracy using Nova Pro’s advanced reasoning capabilities.
Building intelligent AI agents capable of executing multistep tasks using custom-tuned Nova Pro or Premier.
Enabling voice-based customer interactions using Nova Sonic’s speech-to-speech capabilities.