The Pulse
TwelveLabs Brings Pegasus 1.5 Video Segmentation to Amazon Bedrock
AWS made TwelveLabs’ Pegasus 1.5 available through Amazon Bedrock, where it can return timestamped segments and structured metadata from video. The model can process videos up to two hours long, but Bedrock access requires cross-Region infe
AI.info Team · Written with AI · fact-checked ·
Pegasus 1.5 searches the full video, not just a clip
TwelveLabs’ Pegasus 1.5 can scan long-form video for moments an organization defines, then return those matches as structured, timestamped data. AWS announced the model’s Amazon Bedrock availability on October 9; a technical guide published October 19 describes how developers can call it through the service.
The shift is from asking questions about a selected clip to defining the events or segments to find across a whole video. AWS says the model can identify editorial sections, sports plays, speaker changes and brand appearances, returning requested fields such as summaries and topics alongside start and end times.
“Now, Pegasus 1.5 analyzes the entire video, finds every segment that matches user-defined criteria, and returns structured, timestamped metadata.”
Ron Tal, Narcisse Zekpa, Badri Raghunathan Srikumar, Shannon Hong, Nate Stecz and Suheel Farooq, AWS Machine Learning Blog
Users can combine written descriptions with reference images to look for a particular product, logo or person and describe what appears on screen. Pegasus 1.5 also analyzes visual content, audio and on-screen text, and AWS says it needs no advance indexing or separate ingestion pipeline.
Two-hour videos meet a cross-Region requirement
The model handles videos up to two hours long in a single request, twice the one-hour limit AWS gives for Pegasus 1.2. Developers can provide a video from Amazon S3, up to 2 GB, or send an inline base64 video up to 25 MB. Requests use Bedrock’s InvokeModel API, with InvokeModelWithResponseStream available for streamed responses.
That reach comes with a deployment constraint: Amazon Bedrock’s model card says Pegasus 1.5 supports cross-Region inference only, not in-Region on-demand inference. The available profiles route requests within the United States or globally, so teams must choose a routing option that fits their data-residency requirements. AWS says video sent through Bedrock stays within the AWS data boundary, but the model is not confined to a single AWS Region.
Structured output still needs review
Pegasus 1.5 offers a general mode for prompt-based analysis and a time-based metadata mode for user-defined segments. In the latter, developers specify what a matching segment looks like and which fields to extract; the model returns matching sections as JSON. That format can feed video archives, highlight-production workflows or scene-level search without requiring teams to build a separate indexing step first.
AWS describes the intended change in labor as a move from manually finding and tagging every moment to reviewing and refining model-generated metadata. The guide does not present the output as a substitute for human review. Charges run through AWS Marketplace and appear on the AWS bill under TwelveLabs, rather than under Amazon Bedrock; AWS directs customers to its pricing page for rates.
Availability and billing depend on AWS setup
The October 9 announcement made Pegasus 1.5 available through US and global cross-Region inference profiles. AWS documentation lists the model under the ID twelvelabs.pegasus-1-5-v1:0 and instructs developers to use a geographic or global inference profile for requests. Teams can begin in the Bedrock console or invoke the model programmatically, subject to account access and permissions.
For media teams, the concrete offer is automated, timestamped segmentation of long video inside Bedrock, with custom definitions and image references. The trade-off is equally concrete: workloads must use cross-Region inference, and AWS bills the model through Marketplace under TwelveLabs.