At its re: Invent conference, Amazon Web Services (AWS) introduced Nova, a new family of multimodal AI models designed to push the boundaries of generative technology. In total, there are four text-generating models: Micro, Lite, Pro, and Primer, each customized for diverse workloads.
Micro, Lite, and Pro models will be available to AWS customers on Tuesday, while Primer will be launched in early 2025.
Micro, Lite, Pro, and Primer
Talking about the uses and features of these four models, first, we have Micro, which is optimized for speed and handles text with the lowest latency. Lite and Pro process text, videos, and images and balance speed, accuracy, and cost.
On the other hand, Primer is the most advanced tool, used for complex tasks and allowing customized model training.
Regarding the capabilities of these models, Micro has 128,000-token context windows, which means that it can process around 100,000 words. Lite and Pro have 300,000-token context windows, which work around 225,000 words, 30 minutes of footage, and 15,000 lines of computer code.
Furthermore, AWS said that Nova models’ performance will expand, and context windows will support over 2 million tokens.
AWS Introduced Nova Canvas and Nova Reel
In addition to this conference, AWS also introduced Nova Canvas and Nova Reel, both of which focus on generative media. Nova Canvas allows users to create and edit images and offers features like background removal and style adjustment.
On the other hand, Nova Reel generates short videos of up to 6 seconds. It allows users to adjust the camera motion and generate videos with pans, zoom, and 360-degree rotations. Currently, Nova Reel is limited to six-second videos, which take only 3 minutes to generate. According to AWS, it will create two-minute videos, and an upgraded version is expected to be released soon.
