Amazon Web Services (AWS) has announced a new family of multimodal generative AI models called Nova, designed to revolutionize cloud computing. The four text-generating models – Micro, Lite, Pro, and Premier – are available now, with the image-generation model, Nova Canvas, and video-generating model, Nova Reel, also launched today.
Nova’s CEO Andy Jassy emphasized that these frontier models have made significant progress in just four to five months, demonstrating their value. The text-generating models can be optimized for 15 languages, with varying sizes and capabilities. Micro focuses on low latency, while Premier is designed for complex workloads. Lite offers a balanced combination of accuracy, speed, and cost.
Nova Canvas allows users to generate and edit images using prompts, while Nova Reel creates videos up to six seconds in length from prompts or reference images. Jassy highlighted the importance of responsible use, including watermarking and content moderation, to combat misinformation and child sexual abuse material.
AWS has also expanded its safety measures to prevent the spread of harmful content. However, the company remains vague about its training data, citing competitive advantages and potential IP-related lawsuits.
Looking ahead, AWS is working on a speech-to-speech model for Q1 2025 and an “any-to-any” model for mid-2025. These models will enable users to input multiple formats of data and output corresponding formats, powering applications from translators to content editors to AI assistants.
Nova’s generative capabilities are set to transform the future of cloud computing, with Jassy stating that these models will be built and consumed in a way that’s “fundamentally different” from current AI approaches.
Source: https://techcrunch.com/2024/12/03/amazon-announces-nova-a-new-family-of-multimodal-ai-models