Amazon Bedrock now offers TwelveLabs’ Pegasus 1.5, a video-language model that converts long-form footage into structured, timestamped metadata. Instead of answering one question about a clip, the model can scan a video for every segment that matches a user-defined condition and return the results as JSON.
A developer can define editorial sections, sports plays, speaker changes, brand appearances, or other events, then request start and end times plus selected fields. Pegasus 1.5 also adds video segmentation and multimodal prompts, allowing a text description to be paired with a reference image. Bedrock accepts video from an Amazon S3 URI or as a base64 payload and can process as much as two hours on the fly without a separate indexing step.
AWS says data remains within its cloud boundary, which may simplify adoption for teams already using Bedrock. The release reduces manual review and custom video-search plumbing, but useful output still depends on clearly defining the events to detect and validating the returned segments. Availability through a managed service also does not independently establish accuracy across every type of footage or domain.