A video model capable of analyzing and segmenting long content without pre-indexing

TwelveLabs made its video model Pegasus 1.5 available on the Amazon Bedrock platform starting October 9, 2026, according to official announcements relayed by the AWS Machine Learning Blog and syncai.news. Unlike its predecessors, this model converts long videos (up to 2 hours, compared to 1 hour for Pegasus 1.2) into structured and timestamped metadata without requiring a pre-indexing or prior ingestion step.

The key innovation lies in its ability to automatically segment specific events within a video, such as speaker changes or brand appearances, and return these segments in the form of JSON with precise timestamps. For example, a query like ‘extract all moments where a logo appears’ will generate a list of video segments with their timestamps, as explained in the functional description cited by the sources: ‘You describe the moments that matter to your workflow, and the model returns every matching segment as JSON.’

You describe the key moments for your workflow, and the model returns each corresponding segment in the form of JSON with precise timestamps.

Multimodal Prompting and Real-Time Processing: Features Tailored to Dynamic Video Streams

Pegasus 1.5 introduces two key features for applications requiring real-time video analysis: multimodal prompting and on-the-fly processing. Multimodal prompting enables the association of visual references (up to 30 images per query, according to Logeshwaran.org) with textual instructions to pinpoint specific elements, such as products or faces. This approach eliminates the need for a complex ingestion pipeline, as highlighted by the technical detail shared across all sources: "No ingestion pipeline or pre-indexing step is required."

Real-time processing paves the way for applications where latency is critical, such as content moderation or live video stream analysis. Videos can be provided directly via an Amazon S3 URI or a base64 stream (limited to 25 MB), simplifying integration with existing systems. Two analysis modes are offered: a general mode for textual or structured JSON responses, and a time_based_metadata mode to extract timestamped segments based on user-defined criteria.

Multilingual Compatibility and Technical Accessibility: Advantages for African Organizations

Pegasus 1.5 retains its predecessor’s multilingual capabilities, with full support for English and partial support for 12 other languages, including French, Arabic, Portuguese, and Russian, according to the technical details shared by Logeshwaran.org. This feature is particularly relevant for businesses and government agencies in West Africa, where French and local languages play a central role in institutional communication and media.

In terms of accessibility, the model is available via the InvokeModel and InvokeModelWithResponseStream APIs on the bedrock-runtime endpoint, accessible in the us-west-2 region. Users must enable the model in the Amazon Bedrock console and configure specific IAM permissions. Costs are tracked through application inference profiles and ARN tags, enabling granular billing based on video seconds analyzed and output tokens, as outlined by AWS.

What this changes here: opportunities for media, security, and governance in West Africa

For media organizations in Benin and West Africa, Pegasus 1.5 could revolutionize video archive management. Television channels and streaming platforms could automate content indexing, simplifying keyword or timestamped segment searches. For example, a Beninese broadcaster could automatically extract all segments from a political speech or live broadcast, reducing time spent on manual editing and analysis.

In the fields of security and governance, administrations could leverage video segmentation to analyze surveillance footage or institutional meetings. For instance, a ministry could quickly identify key moments in a press conference or detect suspicious behavior in public security videos without requiring heavy video processing infrastructure. Multilingual support, including French, would also enhance the inclusivity of these tools for Francophone populations.

Finally, local businesses operating in sectors like e-commerce or advertising could benefit from multimodal prompting to analyze promotional videos or customer feedback. For example, an online sales platform could automate the extraction of highlighted products in demonstration videos, optimizing their search rankings and visibility.