AWS re:Invent 2025 - [NEW LAUNCH] Amazon Nova 2 Omni: A new frontier in multimodal AI (AIM3324)

AWS re:Invent 2025 - Amazon Nova 2 Omni: A New Frontier in Multimodal AI

Overview of the Amazon Nova Family of Models

• The Amazon Nova family of foundation models was launched at re:Invent 2024, including:
  • Nova Understanding models for text, image, and video understanding
  • Nova Canvas for image generation
  • Nova Real for video generation
  • Nova Sonic for real-time conversational AI

Introduction to the Amazon Nova 2 Family

• Four new models in the Amazon Nova 2 family were introduced:
  • Nova 2 Light: A fast, cost-effective reasoning model for everyday workloads
  • Nova 2 Pro: A higher-performance reasoning model for complex tasks
  • Nova 2 Omni: A unified multimodal reasoning and image generation model
  • Nova 2 Sonic: An improved version of the conversational AI model

Key Capabilities of the Amazon Nova 2 Omni Model

• Multimodal understanding and generation: Can process and generate content across text, images, video, and audio
• Hybrid reasoning: Developers can control the level of reasoning the model applies
• Powerful multimodal perception: State-of-the-art performance on tasks like document understanding, audio understanding, and video understanding
• High-quality image generation: Improved text rendering and spatial understanding compared to previous models
• Broad language support: Understands over 200 languages, including up to 10 languages for audio/speech

Technical Performance of the Amazon Nova 2 Omni Model

• Highly competitive on benchmarks measuring language understanding, reasoning, instruction following, and tool calling
• Outperforms other leading models on the Artificial Analysis Index, a consolidated metric across 10+ benchmarks
• Achieves state-of-the-art results on document understanding tasks like OCR and key information extraction
• Ranked #2 on the MMAU leaderboard for audio understanding and reasoning

Business Applications and Use Cases

Document Understanding

• Accurately extracts text, images, and structured information from complex documents
• Can identify inconsistencies and perform calculations within the documents

Audio Understanding

• Transcribes speech, summarizes audio content, and answers questions about audio files
• Supports multi-speaker diarization and multiple languages

Image and Video Understanding

• Excels at perception tasks like object detection, scene understanding, and temporal reasoning
• Outperforms other models on benchmarks like the Video Benchmark and the new Mavericks benchmark

Image Generation and Editing

• Generates high-quality, realistic images from text prompts
• Supports a wide range of image editing operations like adding, altering, and replacing objects

Customer Spotlight: Densu Digital's Use Cases

• Densu Digital, a leading advertising agency, is using the Amazon Nova 2 Omni model in several ways:
  • Ad creative generation, performance prediction, and improvement suggestion
  • Automating marketing workflows and agent-based applications
  • Connecting in-store and digital experiences through persona-based interactions

Key Takeaways

• The Amazon Nova 2 Omni model represents a significant advancement in multimodal AI, with state-of-the-art performance across a wide range of perception, reasoning, and generation tasks.
• The model's ability to understand and generate content across text, images, video, and audio enables new classes of applications and workflows that were previously difficult to achieve.
• Customers like Densu Digital are already leveraging the power of the Nova 2 Omni model to streamline creative processes, automate marketing operations, and create more immersive customer experiences.
• The technical performance and real-world business impact demonstrate the transformative potential of this new frontier in multimodal AI.