his Tuesday, Amazon Web Services, the cloud computing division of the giant Amazon, announced at the re:Invent conference a new generation of multimodal AI
models called Nova. In this launch, Amazon Web Services released four text-generating models, the Micro, Lite, Pro, and Premier. The first three generative AI models are already available for AWS subscribers, but the Premier version of the Amazon AI Nova is planned to show up early next year, as the CEO Andy Jassy states.
“We’ve continued to work on our own frontier models, and those frontier models have made a tremendous amount of progress over the last four to five months. And we figured, if we were finding value out of them, you would probably find value out of them.”, the Amazon CEO stated.
The new four text-generating Amazon Nova AI models are optimized primarily for English but it is also available for other 15 languages and features various capabilities. The Micro Nova AI model is able to process and generate just text but stands out with the highest process speed, offering faster responses compared with any model in this launch.
The Lite Nova AI model is able to handle image, video, and text queries while the Pro version of the Amazon AI Nova can provide at the same time a mix of accuracy, fast time, and cost for various tasks.
It’s important to mention that from all of these, the Premier Amazon Nova AI model is the most skilled and capable, and will be created for more complex tasks. So, that’s probably why it will be available early next year, in order to design it without flaws. Similar to the Lite version, Pro and Premier can analyze text, images, and even video, and are ideal to be used for various tasks like document analysis, summarizing charts, or diagrams.
Also, Amazon announced that Micro Nova AI has 128,000 tokens, which means that is able to process about 100,000 words, while Lite and Pro have around 300,000 token context windows which makes them able to handle about 225,000 words, 15,000 lines of computer code and also 30 minutes of video footage.
Subscribe to our newsletter
According to Amazon Web Services, early next year some of these Nova AI models will be expanded up to 2 million tokens to support even more complex tasks.
Even more so, in addition to these four generative Nova AI models, Amazon also announced at the re:Invent conference two new models, an image-generation model called Nova Canvas, and a video-generation one called Nova Reel.
Canvas allows its users to generate and edit certain images based on prompts and offers an option to change the color schemes and layouts of the generated images. On the other hand, Nova Reel can generate videos of up to six seconds long based on prompts or reference images. It’s important to note that its users are able to adapt the camera motion in order to generate videos with for example 360-degree rotation or zoom.
At the same conference, soon AWS is planning to launch in Q1 2025 for the Nova AI app a speech-to-speech model that will take speech in order to offer an improved version of it. Also, an “any-to-any” model is scheduled to be launched in 2025, but for now, AWS chose to not reveal further details about it.
Stay tuned for more updates!
By
Raluca Matei
•
December 4, 2024 10:00 AM
.webp)
