Amazon has announced a new group of AI models under the ‘Nova’ moniker. The Amazon Nova foundation models can be loaded as part of the Amazon Bedrock model in AWS. Nova Pro, Lite, and Micro handle various automation tasks, while Canvas and Reel will be Amazon’s answer to AI image and video generation.
The Amazon Nova models
The three automation models break down as follows. Amazon Nova Micro is a text model “optimised for speed and cost.” Amazon claims Micro excels at text summarisation, translation, interactive chat, and brainstorming, as well as simple mathematical reasoning and coding. It also supports proprietary data customisation via fine-tuning so users can make it more accurate.
Nova Lite is a “very low-cost multimodal model” capable of processing images, video, and text inputs to generate text outputs. Amazon claims Nova Lite handles real-time customer interactions, and document analysis with high accuracy. It can analyze multiple images and up to 30 minutes of video. It’s also capable of fine tuning.
Nova Pro is a multimodal model with “the best combination of accuracy, speed, and cost for a wide range of tasks.” Amazon claims Nova Pro shows strong capabilities in processing both visual and textual information, while also excelling at analyzing financial documents. Nova Pro also serves as a “Teacher model to distill custom variants of Amazon Nova Micro and Lite.”
Amazon is also dropping two content generation models – Canvas, Reel, and Premier. Canvas can produce quality images with control over style and content, allowing for editing features like inpainting, outpainting, and background removal.
On Amazon Nova Reel, users can produce short videos with text and image prompts, manipulate the visual style and pacing. Amazon says Canvas can generate professional-quality video and outperforms existing models on “human evaluations of video quality and consistency.” Later in the year, Amazon wants to release a speech-to-speech model. Nova Premier seems to function like a content creation verson of Nova Pro.
These announcements all come from the AWS re:Invent conference in Las Vegas. There the company also announced the construction of a huge AI compute cluster using its Trainium 2 chips. Once completed, the company expects it to be the world’s biggest. Amazon subsidiary Anthropic has essentially been bought the biggest playground of any kid in the block.
The company is hard at work playing catch up to companies like Microsoft’s OpenAI, and Google’s Gemini. The company is also working on a newer AI-powered Alexa, which was supposed to drop a few months ago but has been pushed to 2025.
