Pegasus 1.5 from TwelveLabs is now available in Amazon Bedrock

AWS has added TwelveLabs’ Pegasus 1.5 to Amazon Bedrock, and the headline change is simple: the model no longer has to be asked about one clip at a time. It can look across an entire video, find every segment that matches what a user defines, and hand back structured metadata with timestamps attached.

That matters because video is a nasty kind of unstructured data. Most of it still sits there, hard to search and harder to turn into something useful without humans scrubbing through it or stitching together a pile of tools. Pegasus 1.5 is aimed straight at that problem.

The model now supports time-based metadata generation, so a user can define moments such as editorial segments, sports plays, speaker changes, or brand appearances. It returns matching segments as JSON, with start and end times plus whatever fields were requested. It also adds multimodal prompting with reference images, so a segment can be defined with both text and a visual cue.

AWS says Pegasus 1.5 can process video directly from an Amazon S3 URI or a base64 string, with no pre-indexing step. Data stays within the AWS data boundary. The model can handle videos up to two hours long in a single request, which is double the limit listed for Pegasus 1.2.

There’s also a more practical developer angle here. Pegasus 1.5 on Bedrock can be called with InvokeModel or InvokeModelWithResponseStream, and in general mode it can answer in plain text or JSON. For people building workflows around video, that’s the real appeal: not a clever demo, but a way to make long footage searchable without first building a search system around it.