Amazon Bedrock Managed Knowledge Base now supports multimodal embeddings for video, audio, and image content with TwelveLabs Marengo 3.0
Amazon Bedrock Managed Knowledge Base now supports TwelveLabs Marengo 3.0 for multimodal embeddings, enabling direct video, audio, and image search without transcription reliance.
Amazon Web Services announced the integration of TwelveLabs Marengo 3.0 as an embedding model within Amazon Bedrock Managed Knowledge Base, expanding capabilities beyond text-based search. The update allows customers to generate multimodal embeddings that encode visual scenes, speech, and video cues directly from uploaded media assets. Users can upload content from sources such as Amazon S3 and perform natural language searches without managing underlying infrastructure. This enhancement captures meanings that transcription alone cannot, improving search accuracy for non-textual content.
Marengo 3.0 produces 512-dimensional vectors optimized for retrieval performance, delivering state-of-the-art accuracy in locating relevant media segments. The model includes segment start and end times, enabling applications to navigate directly to specific moments within videos. This feature supports use cases across industries such as sports analytics, media and entertainment, security, education, and retail by improving precision in content retrieval.
Configurable segmentation options allow users to align the model’s processing with their content structure, enhancing flexibility for different types of media. The integration eliminates the need for manual transcription or separate processing pipelines, streamlining the workflow for multimodal search. Customers can now search video, audio, and image libraries using natural language queries, reducing reliance on keyword-based methods.
AWS provides documentation and resources for the new embedding model, including a user guide for integration and a product page for Amazon Bedrock Knowledge Bases. The announcement highlights the model’s compact vector output and improved retrieval capabilities as key benefits for developers and enterprises. No additional infrastructure management is required, simplifying deployment and scaling for media search applications.