Amazon Announces 'Amazon Nova,' a Multimodal Generative AI Model Available on AWS
Amazon Web Services (AWS), Amazon's cloud computing division, announced its own multimodal generative AI model, Amazon Nova , at the event '
Introducing Amazon Nova: Frontier intelligence and industry leading price performance | AWS News Blog
https://aws.amazon.com/jp/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/
Amazon Nova: Meet our new foundation models in Amazon Bedrock
https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws
Amazon announces Nova, a new family of multimodal AI models | TechCrunch
https://techcrunch.com/2024/12/03/amazon-announces-nova-a-new-family-of-multimodal-ai-models/
Amazon unveils 'Nova' AI models, looking to make its mark in the generative AI revolution – GeekWire
https://www.geekwire.com/2024/amazon-unveils-nova-ai-models/
'Amazon Nova helps reduce the cost and latency of nearly any generative AI task,' AWS said. 'With Amazon Nova, you can build a range of intelligence classes optimized for enterprise workloads to analyze complex documents and video, understand charts and diagrams, generate compelling video content, and build sophisticated AI agents.'
Amazon Nova will be available in two types, an understanding model and a creative content model, and will be offered through Amazon BedRock , a service that makes generative AI available on AWS.
There are four understanding models: The understanding models are multilingual models that support more than 200 languages, and are particularly effective in English, German, Spanish, French, Italian, Japanese, Korean, Arabic, Simplified Chinese, Russian, Hindi, Portuguese, Dutch, Turkish, and Hebrew.
・Amazon Nova Micro
A low-cost, low-latency text-only model with a context length of 128,000 tokens that can efficiently perform text summarization, translation, content classification, chat, simple mathematical reasoning and coding. It also supports fine-tuning and model distillation on your own data.
・Amazon Nova Lite
A low-cost multi-modal model that processes image, video and text inputs at lightning speed. It can process inputs of up to 300,000 tokens, analyze multiple images and up to 30 minutes of video in a single request, and allows fine-tuning on both text and multi-modality.
・Amazon Nova Pro
A high-performance multi-modal model that provides the best balance of accuracy, speed and cost. It can process inputs of up to 300,000 tokens and execute complex workflows using APIs and tools. It is particularly good at analyzing financial documents and can handle code bases of 15,000+ lines.
・Amazon Nova Premier
A top-level model intended to be used as an optimal 'teacher' model for extracting custom models for complex inference tasks, rather than as a stand-alone model.
There are two creative content models: At the time of writing, only English prompts are supported.
・Amazon Nova Canvas
A generative model that can create studio-quality images, with precise control over style and content. Below are the prompts and generated images that were actually entered into Amazon Nova Canvas.
・Amazon Nova Reel
It can generate short videos from text prompts and images. It is a video generation model that can adjust the visual style and pace to create professional quality videos for marketing, advertising and entertainment. You can get a good idea of what the video generated by Amazon Nova Reel looks like by watching the movie below.
According to AWS, all Amazon Nova models have built-in safety and content moderation features, and in particular the creative content model has a watermarking function, which AWS claims promotes responsible use of AI and prevents the generation of inappropriate content.
Of the Amazon Nova models, 'Amazon Nova Micro,' 'Amazon Nova Lite,' 'Amazon Nova Pro,' 'Amazon Nova Canvas,' and 'Amazon Nova Reel' will be available from December 3, 2024, and 'Amazon Nova Premier' is scheduled to be released in the first quarter of 2025 (January to March).
Amazon Nova is available primarily in AWS' US East (Northern Virginia) region, with the Micro, Lite, and Pro models also available in the US West (Oregon) and US East (Ohio) regions through cross-region inference. Pricing is based on a pay-as-you-go model, just like other Amazon Bedrock services, and charges are based on actual usage.
In addition, Amazon announced plans to introduce two additional Amazon Nova models by 2025. One is a 'speech-to-speech model' that aims to realize more natural, human-like interactions by understanding voice input in natural language and interpreting linguistic and non-linguistic cues such as tone and rhythm.
The second is the 'native multimodal-to-multimodal model,' also known as the 'any-to-any' modality model, which can handle different forms of data, such as text, images, audio, and video, on both input and output. This enables content transformation between different modalities, content editing, and even AI agents that can understand and generate all modalities.
AWS stated that these developments are 'just the beginning' and that the introduction of these two models will enable a wide range of tasks to be performed with a single model, greatly simplifying application development. It also indicated that it will continue to innovate to deliver real value to customers.
Related Posts:
in Software, Web Service, Video, Posted by log1i_yk