Amazon has launched Nova, a new family of artificial intelligence (AI) models designed to fulfill automation tasks with affordable costs and performance above the market average, according to the firm. The algorithms will be available to developers and businesses gradually through the Amazon Web Services (AWS) Bedrock library.
“Amazon Nova can analyze complex documents and videos, understand graphs and diagrams, generate engaging video content, and create sophisticated AI agents, thanks to a variety of intelligent solutions optimized to respond to enterprise workloads,” explains the company.
What is Nova from Amazon?
The proposal includes two categories of models: one focused on understanding and another on the production of creative content. The developments included in the first segment are foundational and are offered in four modalities, one of them not yet available:
- Amazon Nova Micro– Processes only text, supports 200 languages, and has a context span of up to 128,000 tokens. It effectively responds to tasks such as synthesis, translation, content classification, digital conversations, basic mathematical reasoning, and coding. Supports custom training data to improve the accuracy of results.
- Amazon Nova Lite– Receives written, image, and video prompts, but responds only with text. Handle interactions in real time, analyze documents and multimedia materials and process up to 300,000 tokens. It is ideal for machine learning and detailed analysis tasks.
- Amazon Nova Pro– Large-scale multimodal model with processing capacity of 300,000 tokens. It is capable of executing highly complex natural language and computer vision tasks. Stand out in benchmarks such as TextVQA and VATEX and serves as a basis for distilling smaller models.
- Amazon Nova Premier: it will be the most advanced model in the family. It is currently in development. It will have “complex reasoning” capabilities and is expected to be available in early 2025.
The creative content section is made up of the Nova Canvas and Nova Reels systems. The first resource is focused on generating high quality images through prompts up to 1,024 characters. Add various editing features to remove backgrounds, add objects, and adjust color or lighting schemes. So far, it only accepts requests in English.
For its part, Nova Reels has similar characteristics to Sora from Open AI or Movie Gen from Meta. You can create six-second videos in high resolution through text and image prompts. It also controls the visual style and pace of multimedia content. The Amazon team assures that it is ideal for tasks of marketingadvertising and entertainment. He states that “outperforms existing models in human assessments of video quality and consistency”.
The organization founded by Jeff Bezos emphasizes that its new family of AI models incorporates strict security mechanisms to safeguard intellectual property and guarantee its responsible use. The pieces generated with these resources will be identified with a watermark to show their origin.
Amazon has increased its commitment to compete in the emerging AI business. The launch of Nova represents one of the riskiest steps of the big tech to guarantee a privileged place in the sector, according to analysts. The giant of e-commerce has built a strong infrastructure around the cloud using AWS services. The platform is used by other companies such as Apple to support various AI products. Experts point out that the great challenge will be to bring this innovation to the mass consumption niche. Reports from various media speculate that the company plans to launch an AI-optimized version of Alexa under a subscription scheme. The plan has been put in place due to various technical problems, according to unofficial information.
#Amazon #announces #Nova #family #models #generate #text #images #videos
