Video and Image Search in Amazon Bedrock Using Marengo 3.0

Amazon Bedrock has announced the general availability of TwelveLabs Marengo Embed 3.0 as an embedding model within its Knowledge Bases service. This development allows users to perform natural language searches across video, audio, and image content without the need for complex infrastructure. Marengo Embed 3.0 encodes multimedia data into a compact, 512-dimensional vector space, facilitating efficient storage and retrieval.

The service supports various media formats, including MP4 and MOV videos, JPEG and PNG images, and audio tracks, with native connectors for storage solutions like Amazon S3, SharePoint, and Confluence. Users can create a knowledge base, ingest media assets, and run semantic queries to locate specific moments or segments, such as a penalty kick in a soccer match.

The process involves uploading media to an Amazon S3 bucket, configuring a knowledge base with the Marengo Embed 3.0 model, and syncing the data for automatic extraction and embedding. Once ingested, users can test queries directly within the console, retrieving relevant video segments with associated metadata. The service is available in the US East and US West AWS regions.

This enhancement aims to support industries such as sports analytics, media management, security, education, and retail by enabling more efficient and accurate media content search and analysis. Pricing is based on storage and retrieval, with charges for embedding generation aligned with Amazon Bedrock's standard rates.