Amazon Rolls Out Nova Sonic and Nova Reel 1.1 for Generative Voice and Video AI

Amazon has introduced two new additions to its generative AI portfolio—Amazon Nova Sonic, a foundation model for voice-based applications, and Amazon Nova Reel 1.1, an updated model for text-to-video generation.

Nova Sonic integrates speech recognition, understanding and generation into one model, removing the need for separate components. Traditional voice systems involve complex pipelines—converting speech to text, processing through a large language model, and converting the response back to speech. According to Amazon, this approach “fails to preserve crucial acoustic context and nuances.”

“Nova Sonic takes a new approach,” the company said. “It unifies the understanding and generation capabilities into a single model.” The result is a voice agent that not only understands user input but also responds with an appropriate tone, pace, and style.

The model is available through Amazon Bedrock. It supports applications in customer service, travel, education, healthcare and entertainment. In one example shared by Amazon, a virtual travel assistant shifts its tone in response to a customer’s change in emotion—moving from enthusiastic to reassuring when concerns about cost are raised. Another use case includes an enterprise dashboard assistant that grounds answers in company data and maintains multi-turn dialogue without requiring users to reset context.

Nova Sonic also generates transcripts of user speech. This feature allows developers to integrate external APIs and tools, enabling AI agents to perform tasks such as retrieving flight options or accessing internal dashboards.

On the other hand, Nova Reel 1.1 enables multi-shot videos up to two minutes in length, with consistent visual style across 6-second segments. It improves on the previous version in terms of generation speed and coherence. Users can choose to provide a single prompt for the entire video or set individual prompts per shot for more control.

The model supports use cases like marketing campaigns, product design showcases, and social media content creation. “Nova Reel enhances creative productivity,” Amazon said, “while helping to reduce the time and cost of video production using generative AI.”

To get started with Amazon Nova Reel 1.1, users need to visit the Amazon Bedrock console and request access to the model. In the left-hand navigation panel, they should select “Model access” and then locate Amazon Nova Reel in the list of available models. Requesting access here provides permission to use both version 1.0 and 1.1 of the model. Once access is granted, users can begin using Amazon Nova Reel 1.1 through the Amazon Bedrock console, the AWS SDK, or the AWS Command Line Interface (CLI).

The releases are part of Amazon’s broader Nova model family, introduced at re:Invent 2024, which also includes Nova Micro, Lite, and Pro that generate text from different modalities