Amazon Nova: Six new AI models for partners and sellers

The online giant Amazon has been involved in the AI sector for some time now. After releasing the shopping assistant Rufus for Amazon customers at the end of October, the company is now taking the next step. However, this is once again moving away from the end customers and towards the numerous suppliers who sell their products or place adverts on Amazon.

Amazon Nova: these are the new AI models

In an announcement, Amazon shows six variants of a new AI model called Nova. The variants differ in terms of their functions and complexity. For example, there is Amazon Nova Micro, a purely text-based AI model that is primarily trimmed for speed and low costs. Amazon Nova Lite is also designed to be cost-efficient, but can also process images and videos in addition to text. There are also Pro and Premier versions of Amazon Nova. These are multimodal models that should even be able to solve complex tasks.

Last but not least, customers will receive Amazon Nova Canvas, a model for image generation, and Amazon Nova Reel, an AI model for generating videos. Almost all models are now included in Amazon Bedrock in the USA. Only Amazon Nova Premier will be activated in the first quarter of 2025. It is not known if and when the AI models will also be available in this country. All Nova AI models are expected to understand 200 languages. Amazon Nova Lite, Micro and Pro are also expected to be around 75 per cent cheaper than comparable AI models available via Amazon Bedrock.

What can Amazon Nova do?

As already mentioned, Amazon Nova is primarily aimed at sellers, advertising partners and companies. According to Amazon, the AI models are intended to ‘remove barriers for sellers and advertisers and enable a new level of creativity and innovation’. This includes video advertising, for example, as Amazon shows with a clip generated by the AI. The video advertises a fictitious pasta brand by generating an entire city of pasta using AI.

Another possible application is descriptions that the AI can create based on a video. For example, Amazon Nova was shown a clip of a football match without sound. The AI was asked to describe what it ‘saw’. Amazon Nova was able to output the different team colours and even individual plays as text. Advertisers could therefore have their videos translated into text by the AI and then use it to create suitable advertising or social media texts.

What's next for Amazon Nova

Amazon already has further plans for Nova for 2025. The AI model will have two further variants, which have not yet been finalised. One of the two models will support speech as input and output, while the other will be a multimodal model that can also write its output multimodally.

For the speech model, Amazon promises that it will even be able to register non-verbal input - such as tone of voice and speech rhythm. The multimodal model, on the other hand, will form the basis for AI agents. Amazon Nova would then be able to solve several tasks in succession or even control end devices independently.