Amazon’s New Multimodal AI Can Produce Towns Made of Pasta

Amazon has rolled out a new family of “understanding” AI multimodal models that can generate text, image, and video from varying inputs.

Collectively known as Nova, Amazon says it is able to “analyze complex documents and videos, understand charts and diagrams, generate engaging video content, and build sophisticated AI agents.”

Part of Nova is an AI image generator and a video generator. Amazon Nova Canvas is a “state-of-the-art image generation model” that can produce “studio-quality images” offering precise controls over style and content. It has features like inpainting, outpainting, and background removal.

Amazon Nova Reel can produce short videos from text prompts or image inputs. Amazon says it can “generate professional-quality video content for marketing, advertising, and entertainment.” The company released an ad for a fake pasta brand made by Nova Reel which shows a town made of pasta.

It also shared a bunch of other short clips (see below). The company stresses that “all Amazon Nova models include built-in safety controls and creative content generation models include watermarking capabilities to promote responsible AI use.”

In its press release, Amazon says there are three understanding models about to be rolled out with a fourth one coming soon.

The first model is Amazon Nova Micro, a text-only model that operates similarly to ChatGPT. The second is Amazon Nova Lite, a “low-cost” multimodal model that can process image, video, and text inputs to generate text output. Finally, there is Amazona Nova Pro, a “highly capable multimodal model”.

The fourth one, Amazon Nova Premier, will be its “most capable multimodal model for complex reasoning tasks.” Amazon says Nova Premier should be available in early 2025.

The new Nova AI foundation models will be available exclusively as part of the Amazon Bedrock model library in Amazon Web Services (AWS).

Nova AI was announced during the company AWS re:Invent conference taking place in Las Vegas. The Verge notes that the company also said it is building up a massive AI compute cluster through its partnership with Anthropic that relies on its Trainium 2 chips.

“When completed, it is expected to be the world’s largest AI compute cluster reported to date available for Anthropic to build and deploy its future models on,” adds Amazon.