TwelveLabs Releases Pegasus 1.5: Video Language Model Now Available on Amazon Bedrock
Summary:TwelveLabs has released Pegasus 1.5, a video language model, and announced its availability on Amazon Bedrock. The model transforms long-form videos into structured, timestamped metadata, featuring video segmentation, multimodal prompting, and on-the-fly processing for videos up to two hours long. This release enhances video content analysis for applications in news, sports, streaming, and more, offering a more efficient and intelligent solution for video understanding.
Core Features and Technological Innovations of Pegasus 1.5
Pegasus 1.5, developed by TwelveLabs, is a video language model designed to address the challenges of searching and utilizing video content. Its key features include:
- Timestamped Metadata Generation: Users can define key events or segments in a video, and Pegasus 1.5 returns the matching segments with timestamps and requested fields in JSON format.
- Multimodal Prompting: Supports queries text descriptions with visual references, such as locating specific products, logos, or people and their on-screen actions.
- On-the-fly Processing: Eliminates the need for pre-indexing or preprocessing, allowing users to send videos directly from Amazon S3 and receive real-time results.
- Extended Processing Capacity: Can process videos up to two hours long in a single request, doubling the capacity of previous versions.
Additionally, Pegasus 1.5 retains the core capabilities of its predecessors, including video-to-text generation, granular visual comprehension, temporal grounding, and multimodal understanding.
Technical Mechanism Analysis and Engineering Trade-offs
The technical architecture of Pegasus 1.5 is based on deep learning models that combine visual and language processing technologies. Its core mechanisms include:
- Video Segmentation and Timestamp Alignment: Efficient segmentation algorithms decompose videos into semantic units and generate precise timestamps.
- Multimodal Fusion: Integrates visual, audio, and textual information to achieve comprehensive video content understanding.
- Efficient Inference Engine: Optimizes the inference process to support real-time processing of large-scale video data.
In terms of engineering trade-offs, Pegasus 1.5 requires significant computational resources when processing long videos, especially in high-resolution and multimodal data fusion scenarios. Moreover, while real-time processing enhances user experience, it also imposes higher demands on network bandwidth and server load.
Developer Implementation and Deployment Recommendations
For developers, deploying Pegasus 1.5 is relatively straightforward, with the following main steps:
- Environment Setup: Ensure that the AWS account has access to Amazon Bedrock and install the Boto3 SDK.
- Model Invocation: Use the Bedrock Runtime API to call Pegasus 1.5, passing in the video source and prompt information.
- Result Processing: Parse the returned JSON data to extract the required information.
Developers are advised to choose the appropriate analysis mode based on the specific application scenario, such as using the time-based metadata mode when detailed timestamp information is needed. Additionally, it is important to configure request parameters, such as temperature and maximum output tokens, to balance the quality of the generated results and computational costs.
Use Cases and Future Prospects
Pegasus 1.5 is suitable for scenarios that involve processing large volumes of long-form video data, such as:
- News and Broadcast Archives: Quickly generate structured summaries of video content to improve retrieval efficiency.
- Sports Event Analysis: Automatically identify key events and scoring moments to simplify post-production workflows.
- Streaming Platforms: Provide scene-based recommendations and ad placements to enhance user experience.
In the future, Pegasus 1.5 is expected to find applications in more areas, such as video surveillance, intelligent security, and virtual reality. As the technology continues to evolve, its processing capabilities and accuracy will also improve.
— END —Source: AWS Machine Learning Blog (2026-10-19)
Tags: #TwelveLabs #Video Language Model #Amazon Bedrock #Multimodal AI #Real-time Processing
Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.
Community Comments