TwelveLabs Pegasus 1.5 Video Model Launches on Amazon Bedrock
Amazon Bedrock now supports TwelveLabs’ Pegasus 1.5, giving developers advanced capabilities to process long-form video content into structured JSON data without prior indexing pipelines.
Radwan AbdulhadiEditor and technology writer
Overview of TwelveLabs Pegasus 1.5 on Amazon Bedrock
AWS announced the integration of <a href="https://aws.amazon.com/blogs/machine-learning/pegasus-1-5-from-twelvelabs-is-now-available-in-amazon-bedrock/">TwelveLabs’ Pegasus 1.5 on Amazon Bedrock</a>, expanding enterprise video artificial intelligence capabilities. Video represents a major category of unstructured data that has traditionally remained difficult to search and analyze without extensive manual review or complex toolchains. According to the official announcement, the video language model allows developers to turn long-form video recordings into structured, timestamped metadata by defining specific moments that matter to their workflows.
Building on earlier versions of the technology that primarily answered questions about specific clips, the new model analyzes entire video files and extracts every matching segment. Developers can access comprehensive details regarding pricing and model specifications by visiting the <a href="https://aws.amazon.com/bedrock/twelvelabs/">TwelveLabs models page on Amazon Bedrock</a>.
Key Feature Enhancements and Capabilities
Pegasus 1.5 delivers several key updates compared to predecessor models, focusing heavily on operational flexibility and analytical depth. A primary addition is time-based metadata generation, which lets users define domain-specific events, editorial segments, or brand appearances. The model subsequently returns every matching segment structured in JSON format complete with start and end timestamps.
Furthermore, the platform introduces multimodal prompting, enabling segment definitions to combine text descriptions with visual reference images. This functionality helps systems locate specific logos, products, or individuals along with their on-screen actions. To review the formal technical parameters and system boundaries, developers can consult the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-twelvelabs-pegasus-v1-5.html">Pegasus 1.5 model card</a> before deployment.
Processing Architecture and On-the-Fly Analysis
The integration bypasses traditional ingestion pipelines by supporting on-the-fly processing. Videos can be analyzed directly from an Amazon Simple Storage Service (Amazon S3) URI or transmitted as a base64 string without requiring any pre-indexing step. AWS notes that data processed through this architecture remains securely within the official AWS data boundary.
Processing capacity has also doubled. Pegasus 1.5 can successfully analyze videos up to two hours long in a single request, doubling the one-hour limitation found in previous versions like Pegasus 1.2. The model retains essential core features including video-to-text generation, granular visual comprehension of on-screen text and objects, temporal grounding, and broad multilingual support that includes full English capabilities and partial support for multiple other languages.
Implementation and Getting Started
To begin building applications with the new video language model, developers need an active AWS account with access configured in the <a href="https://console.aws.amazon.com/bedrock/">Amazon Bedrock console</a>. Prerequisites include Python 3.10 or later, the AWS SDK for Python (Boto3), and appropriate IAM permissions for invoke actions.
Requests are processed through the bedrock-runtime endpoint using the InvokeModel API or InvokeModelWithResponseStream for streaming responses. Teams can review regional availability details across the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html">full Region list</a> to verify service deployment options for specific cloud environments.
Sources
- AWS Machine Learning BlogPegasus 1.5 from TwelveLabs is now available in Amazon Bedrock
Continue chronologically
Related entity coverage
Related stories
Agentic Retrieval with LangChain and Amazon Bedrock
Explore building Retrieval Augmented Generation applications using LangChain and Amazon Bedrock Managed Knowledge Base for complex multi-part questions.
Query Claims in Natural Language With Amazon Bedrock
Learn how to build a conversational claims assistant using Amazon Bedrock Knowledge Bases to answer natural-language queries with citations.
AWS Introduces Agentic Conversational Video Intelligence
Amazon Web Services details a new agentic conversational video intelligence solution using Amazon Bedrock, Rekognition, and Transcribe.
Condé Nast Builds Multimodal Video Discovery with AWS
Condé Nast partnered with the AWS Generative AI Innovation Center to build an AI-powered multimodal video discovery solution on Amazon Bedrock.
