During the re:Invent conference held on Tuesday, Amazon’s CEO Andy Jassy unveiled an array of six new generative models, collectively known as Amazon Nova. These models are capable of producing text, images, and videos.
This innovative suite of multimodal generative AI comprises different variants: Nova Micro, a text-exclusive model optimized for quick and cost-efficient responses; Nova Lite, a budget-friendly multimodal model capable of processing text, image, and video inputs; and Nova Pro, designed for versatility and efficiency across various tasks, emphasizing speed, accuracy, and affordability, as highlighted in the official announcement. Additionally, Nova Premier stands out as Amazon’s most advanced model for complex reasoning tasks, Nova Canvas is a specialized text-to-image generator, and Nova Reel is dedicated to video creation.
These text-based models have been refined to support 15 different languages. Micro provides a context window of 128,000 tokens, while Lite and Pro models can handle up to 300,000 tokens, equivalent to about 225,000 words or around 30 minutes of video content. There are plans to enhance the context capacities of the larger models to reach 2 million tokens by early next year.
Canvas allows users to create and modify images through natural language prompts. Reels has the capability to produce clips up to six seconds long, utilizing both text prompts and reference images. Moreover, the video generator includes features for controlling camera motions, such as zooms and pans.
Jassy expressed, “We’ve been refining our frontier models, and they have shown remarkable advancements over the last four to five months. We believe that if we’ve found value in these developments, you will likely benefit from them as well.”
Additionally, Jassy asserted that these models rank among the fastest and most cost-effective in their category, although specific benchmark data backing these assertions has not yet been made public. “We’ve designed these models to integrate seamlessly with proprietary systems and APIs, enabling a more efficient orchestration of automated operations,” he explained. “I believe these models present very appealing options.”
The Micro, Lite, and Pro models, along with Canvas and Reels, are currently accessible to AWS users. Nova Premiere is expected to launch in the first quarter of 2025.
