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, enabling multimodal embeddings for video, audio, and image content by directly encoding visual scenes and speech, capturing meaning beyond transcription.
Amazon Bedrock managed knowledge base now offers TwelveLabs Marengo 3.0 as an embedding model, generating multimodal embeddings from video, audio, and images. It encodes visual scenes, audio, and video cues directly, capturing meaning beyond traditional transcription-based text embeddings. Upload media assets from sources like Amazon S3, sync them, and search in natural language without managing infrastructure. Marengo 3.0 produces compact 512-dimension vectors for state-of-the-art search accuracy, with results including segment start and end times for direct video jumps. This enables use cases in sports analytics, media and entertainment, security, education, and retail. The model provides content-structure-aligned segmentation options. For details, see the TwelveLabs Marengo 3.0 embedding model integration in the Amazon Bedrock knowledge base user guide or the Amazon Bedrock knowledge base product page.
Why it matters
Amazon Bedrock Managed Knowledge Base simplifies the integration of AI models with external data. This update enables more accurate searching of video, audio, and image content.