TwelveLabs Pegasus 1.5 is now available in Amazon Bedrock
TwelveLabs Pegasus 1.5 is now on Amazon Bedrock, allowing users to convert long-form video into structured, timestamped JSON metadata.
TwelveLabs has released Pegasus 1.5 on Amazon Bedrock, a video language model that converts long-form video into structured, timestamped metadata. This update helps developers transform unstructured video content into searchable, actionable JSON information.
A video language model like Pegasus 1.5 acts as an intelligent observer that can "watch" a video and describe what is happening. It translates visual and audio information into text or structured data that computers can easily process and organize.
What is new in Pegasus 1.5?
- Time-based metadata generation: Users can define specific segments, such as sports plays, editorial segments, or brand appearances, and the model returns them as JSON with start and end timestamps.
- Multimodal prompting: The model can combine text descriptions with visual reference images to locate specific people, logos, or products.
- On-the-fly processing: Videos can be analyzed directly from an Amazon S3 URI or a base64 string without requiring a pre-indexing step or ingestion pipeline.
- Extended capacity: Pegasus 1.5 can process videos up to two hours long in a single request, doubling the capacity of the previous version.
Why does video metadata matter?
Video represents a massive category of unstructured data that often remains unsearchable and underutilized. Historically, extracting specific insights from video required manual review or fragmented toolchains. By providing structured, machine-readable metadata, Pegasus 1.5 allows businesses to automate the extraction of insights, making large video libraries much more useful for automated workflows.
How does Pegasus 1.5 work on Amazon Bedrock?
The model is available via the bedrock-runtime endpoint using the InvokeModel or InvokeModelWithResponseStream APIs. The Model ID is twelvelabs.pegasus-1-5-v1:0.
Each request can be run in one of two analysis modes:
- General mode: This is the default mode where a text prompt in the
inputPromptreturns a text answer in themessagefield, or JSON if a schema is supplied. - Time-based metadata mode: Users provide segment definitions in the
responseFormat, and the model returns a list of segments including start and end times.
The model retains core capabilities such as video-to-text generation, granular visual comprehension of objects and text, and temporal grounding. It offers full support for English and partial support for several other languages, including Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish, Thai, and Vietnamese. Your data remains within the AWS data boundary during processing.
Where is Pegasus 1.5 available?
Pegasus 1.5 is currently available in the US West (Oregon) region via Amazon Bedrock.
Source: aws.amazon.com