ALBAWABA - During the AWS re:Invent conference, Amazon Web Services (AWS) displayed its latest AI breakthroughs; the Nova family of multimodal generative AI models. In an attempt to compete with the rapidly expanding AI market, AWS CEO Andy Jassy unveiled the Nova portfolio, which consists of six models made for text, picture, and video generation.
Four text-generating models are available in the Nova family: Micro, Lite, Pro, and Premier. Although English is the main emphasis, these models are optimized for 15 more languages. Micro, Lite, and Pro are available immediately through AWS Bedrock, Amazon’s AI development platform, while Premier is slated for release in early 2025.
Each model has unique set of features to meet a variety of requirements. Amazon claims that Micro, the smallest model, can handle up to 100,000 words via its 128,000-token context window and analyzes text with unparalleled speed and low latency.
Lite and Pro are appropriate for jobs like document summarization, coding, and video content analysis since they have bigger 300,000-token context windows and can evaluate text, pictures, and video. The most sophisticated model, Premier, is designed to support intricate workloads and serve as a basis for creating customized models.
By 2025, AWS says it intends to improve these models even further, expanding their context window capabilities to more than 2 million tokens.
AWS also unveiled two media-focused models, Canvas and Reel, to go along with the text-generating models. Canvas gives users the ability to easily create and alter pictures by providing tools to change color schemes and layouts.
With Reel, users may utilize reference photos or instructions to make brief video segments up to six seconds long. It has tools for modifying camera movements, including 360-degree rotations and pans. AWS has additionally revealed intentions to soon extend Reel's functionality to create two-minute videos.
Citing measures to prevent exploitation and false information, AWS reaffirmed its commitment to safety. Reel and Canvas both claimed to include robust security features, such as content monitoring and watermarking, to encourage responsible usage and deter the creation of dangerous material, Amazon says.
