Marengo 3 is a new release from TwelveLabs that indexes long video files and enables multimodal search across visual and textual cues.

Key features and capabilities

The service supports indexing of video files up to 4 hours, enabling analysis of extended footage without manual segmentation.

It provides multimodal search that combines textual queries with an image reference to find matching moments in the footage.

• Create entities from 1–5 photos to tag and locate specific people or objects within video timelines.
• Supports 36 languages, including Russian, for both search queries and transcribed content.
• Recognizes domain-specific film and sports terminology such as zooms, pan, and tracking shot for more precise retrieval.
• Offers an SDK for pipeline integration and an API billed per request for production workflows.

Tester experience

The author previously wrote about the company and tested Marengo 3 by uploading a file of just over 40 minutes and creating an entity from five photos.

Indexing completed in a matter of minutes, and searches returned the relevant clip with negligible delay during the trial.

The free tier includes 600 minutes for initial testing, allowing evaluation of indexing speed and retrieval accuracy before committing to paid requests.

Use cases

Marengo 3 suits scenarios such as locating spoken phrases, finding athlete actions like goals, and identifying brand placements within creator footage.

By combining visual references and text queries, teams can reduce manual review time and automate extraction of moment-level metadata for large video archives.