(Illustration: On AWS PartyRock, you can use the Amazon Nova Canvas model to create images or design various interactive games with limited free usage. Image source: ChatRPG with Amazon Nova on AWS PartyRock.)
Briefing
At the 2024 AWS re:Invent developer conference, Amazon announced a new series of foundation models— Amazon Nova— designed to provide advanced artificial intelligence solutions for enterprises. These models are available through Amazon Bedrock and support the generation and understanding of text, images, and videos to meet various business needs.
Amazon Nova Models
- Amazon Nova Micro: Focused on text processing, it offers extremely low-latency responses and is cost-effective, suitable for applications requiring rapid text generation.
- Amazon Nova Lite: A low-cost multimodal model capable of quickly processing text, image, and video inputs to generate text outputs, ideal for real-time customer interactions and document analysis.
- Amazon Nova Pro: A high-performance multimodal model that balances accuracy, speed, and cost, suitable for a wide range of tasks involving complex workflows.
- Amazon Nova Premier: The most powerful multimodal model, designed for complex reasoning tasks and fine-tuning custom AI solutions. (Expected to launch in early 2025.)
- Amazon Nova Canvas: An advanced image generation model that creates professional-grade images based on text or image prompts, with features for adjusting colors and layouts, applicable in advertising, marketing, and entertainment.
- Amazon Nova Reel: An advanced video generation model that enables users to produce high-quality short videos from text and image prompts, with natural language controls over visual style and pacing, suitable for marketing and content creators.
| Model | Input Types | Output Types | Context Length | Languages Supported | Key Features | Availability |
|---|---|---|---|---|---|---|
| Nova Micro | Text | Text | 128K tokens | 200+ | Lowest latency, cost-effective, ideal for rapid text generation. | Available now |
| Nova Lite | Text, Image, Video | Text | 300K tokens | 200+ | Low-cost, fast processing of multimodal inputs, suitable for real-time interactions. | Available now |
| Nova Pro | Text, Image, Video | Text | 300K tokens | 200+ | High performance, balances accuracy, speed, and cost for complex workflows. | Available now |
| Nova Premier | Text, Image, Video (planned) | Text | Over 2M tokens | 200+ | Most capable, designed for complex reasoning tasks and custom AI solution fine-tuning. | Early 2025 (planned) |
Key Features
- Multilingual Support: Amazon Nova models support over 200 languages, catering to global business operations.
- High Performance and Cost Efficiency: According to Amazon, the Amazon Nova Micro, Lite, and Pro models offer at least 75% cost savings compared to other leading models in their respective intelligence categories.
- Customization: These models support fine-tuning for both text and multimodal inputs, allowing enterprises to optimize them according to specific needs, enhancing accuracy and relevance.
- Safety and Responsibility: Amazon Nova Canvas and Amazon Nova Reel come with built-in watermarking and content moderation features to ensure the safe and responsible use of AI.
The introduction of Amazon Nova marks a significant advancement in Amazon’s artificial intelligence capabilities, providing enterprises with robust and flexible tools to tackle various generative AI application challenges.
Pricing
Performance vs Price Comparison
(Excerpt from Artificial Analysis Highlights, 2024-12-31.)
(Excerpt from Artificial Analysis Quality vs Price, 2024-12-31.)
Reference
Docs
- What is Amazon Nova?, 2024-12-07, by AWS Official Docs, at AWS re:Invent 2024
Getting Started
- AWS re:Invent 2024 - [NEW LAUNCH] Practical generative AI using Amazon Nova (AIM398-NEW), 2024-12-07, by AWS Events, at AWS re:Invent 2024
- AWS re:Invent 2024 - [NEW LAUNCH] Amazon Nova understanding models (AIM395-NEW), 2024-12-07, by AWS Events, at AWS re:Invent 2024