As mentioned, the AWS re:Invent event has brought significant news in terms of generative artificial intelligence. Amazon has announced Novaits new family of basic models with multimodal capabilities, and has also introduced New canvas y Nova Reelstwo models dedicated to creating images and videos from text.
The Amazon Nova portfolio consists of four basic AI models: Micro, Lite, Pro and Premier. The first three are already available through the Amazon Bedrock platform, while the last one is still in the training phase. However, the company aims to launch it early next year.
Amazon Nova Micro works only with text as an input and output method, while the more powerful variants also allow you to work with photos and videos. The firm headed by Andy Jassy seeks to provide solutions that fit the diverse needs of users, at reasonable costs.
These are the most notable features of Amazon’s new AI:
- Amazon Nova Micro: As we have shown before, it only works with text as input and output method. According to the company, this allows it to operate with the lowest possible latency and at a very low cost. It provides support for more than 200 languages and the maximum context length is 128,000 characters. It is primarily designed for translation tasks, generating summaries and programming, among other possibilities.
- Amazon Nova Lite: Unlike the previous one, it is not limited to text as an input method, as it also supports the use of images and videos. However, it only provides answers in text format. It supports more than 200 languages and can process instructions of up to 300,000 arguments. Amazon defines it as a low-cost multimodal AI with capabilities designed specifically for machine learning tasks to transfer knowledge from a large model to a smaller model.
- Amazon Nova Pro: It is the most powerful option of the three available today. It can also process instructions of up to 300,000 tokens, but stands out for its speed and accuracy. According to Amazon, it performs extremely well in analyzing financial documents, as well as creating video summaries, software development and mathematical reasoning.
- Amazon Nova Premier: This will be the company’s most advanced foundational model, with capabilities focused on complex reasoning tasks. However, not many details have been given regarding this. It is expected to debut in the first months of 2025.
Nova Canvas and Nova Reels: Amazon doubles down on image and video creation
Beyond the new basic models, Amazon has introduced two other generational models dedicated to creating images and videos from text: Nova Canvas and Nova Reels, respectively.
The former promises to generate professional-quality images through requests up to 1024 characters long. In addition, it includes multiple built-in tools for removing backgrounds and adjusting the color scheme, among other editing options. One of its limitations is that, at the moment, it only works in English.
Nova Reels, meanwhile, is Amazon’s answer to platforms like Sora, from OpenAI, and Movie Gen, from Meta. It supports requests up to 512 characters long, in English only, and offers interesting settings for zoom, rotation and various camera panels. You can also use a reference still image to create a clip. At the moment it is limited to generating videos up to 6 secondsbut Amazon promises that later they will last up to two minutes.
Amazon’s idea for next year is to expand the offer of Nova models. This is not only limited to the launch of Premier but also for presentation a voice pattern in speech and another multimodal version with no restrictions on input and output formats.
