Piling on to the checklist of bulletins from Amazon at AWS re:Invent, the corporate introduced Amazon Nova, a household of basis fashions that promise “frontier intelligence and trade main value efficiency.”
They will deal with typical generative AI duties, resembling analyzing paperwork and movies, understanding charts, producing video content material, or constructing superior AI brokers.
“Whether or not you’re growing doc processing purposes that have to course of pictures and textual content, creating advertising and marketing content material at scale, or constructing AI assistants that may perceive and act on visible data, Amazon Nova supplies the intelligence and suppleness you want with two classes of fashions: understanding and inventive content material technology,” Danilo Poccia, chief evangelist at AWS, wrote in a weblog put up.
Its fashions are divided into “understanding fashions” and “content material technology fashions.” There are three understanding fashions at present obtainable, they usually can absorb textual content, picture, or video as enter and generate textual content as output. The three fashions embody:
- Amazon Nova Micro: The bottom latency mannequin that’s optimized for pace and price, succesful solely of working with textual content.
- Amazon Nova Lite: A low-cost multimodel that may course of pictures, video, and textual content and generate textual content
- Amazon Nova Professional: Essentially the most succesful mannequin of the three that may course of as much as 300K enter tokens, making it in a position to course of codebases of over 15,000 traces of code.
A fourth mannequin, Amazon Nova Premier, can also be at present being skilled and is anticipated in early 2025. The corporate says this mannequin will likely be extra succesful and be designed for advanced reasoning duties, and can be utilized as a instructor mannequin for different customized fashions, as can Amazon Nova Professional.
“What makes Amazon Nova significantly highly effective for enterprises is its customization capabilities,” Poccia wrote. “Consider it as tailoring a swimsuit: you begin with a high-quality basis and alter it to suit your actual wants. You possibly can fine-tune the fashions with textual content, picture, and video to grasp your trade’s terminology, align together with your model voice, and optimize on your particular use circumstances. As an example, a authorized agency would possibly customise Amazon Nova to higher perceive authorized terminology and doc constructions.”
The inventive content material technology fashions embody two fashions, and may settle for textual content and picture inputs and output pictures or movies:
- Amazon Nova Canvas can produce studio-quality pictures and offers modifying options like inpainting, outpainting, and background elimination
- Amazon Nova Reel can be utilized to provide quick movies primarily based on textual content or picture prompts, and permits the person to manage visible fashion and pacing.
In line with Amazon, all of those fashions can be found on Amazon Bedrock and embody built-in security controls. The content material technology fashions additionally embody watermarking capabilities so promote accountable use of AI.
They’re at present obtainable within the US East (N. Virginia) AWS Area, and Amazon Nova Micro, Lite, and Professional can be found in US West (Oregon) and US East (Ohio) as effectively. The fashions may also perceive and generate content material in additional than 200 languages, resembling English, German, Spanish, French, Italian, Japanese, Korean, Arabic, Simplified Chinese language, Russian, Hindi, Portuguese, Dutch, Turkish, and Hebrew.
