At the re:Invent conference, Amazon Web Services (AWS) introduced Nova, a series of multimodal generative AI models. The lineup includes four text-generating models—Micro, Lite, Pro, and Premier—and two generative media models, Nova Canvas for images and Nova Reel for videos.
Text Models:
Micro offers the fastest processing for text-only tasks, while Lite, Pro, and Premier handle text, images, and videos. Pro balances speed, accuracy, and cost, and Premier excels in advanced workloads, serving as a customizable “teacher” model. Premier is set to launch in early 2025, while others are already accessible. These models feature vast token context windows, with planned expansions exceeding 2 million tokens.
Generative Media Models:
Canvas generates and edits images, while Reel creates six-second videos with options for camera effects. AWS plans to extend Reel’s capabilities to two-minute videos soon. Both models include safeguards like watermarking to prevent misuse.
AWS emphasizes Nova’s efficiency, affordability, and versatility for developing and fine-tuning AI systems via its Bedrock platform. Future innovations include a speech-to-speech model in Q1 2025 and an any-to-any model by mid-2025, aiming to revolutionize AI’s adaptability across formats.