Typical workflows
This page provides an overview of common workflows for using the TwelveLabs Video Understanding Platform. Each workflow consists of a series of steps, with links to detailed documentation for each step.
All workflows involving uploading video content to the platform require asynchronous processing. You must wait for the video processing to complete before proceeding with the subsequent steps.
Prerequisites
- To use the platform, you need an API key.
Search
Follow the steps in this section to search through your video content and find specific moments, scenes, or information. For an interactive implementation using the Python SDK, see the Quickstart Search Jupyter notebook.
Steps:
- Create an index, enabling the Marengo video understanding engine.
- Upload videos and monitor the processing.
- Perform a search request, using text or images as queries.
Notes
- The search scope is an individual index.
- Results support pagination, filtering, sorting, and grouping.
Generate text from videos
Follow the steps in this section to generate texts based on your videos. For an interactive implementation using the Python SDK, see the Quickstart Generate Jupyter notebook.
Steps:
- Create an index, enabling the Pegasus video understanding engine.
- Upload videos and monitor the processing.
- Depending on your use case, generate one of the following:
Note
When generating open-ended texts, the platform supports streaming responses.
Classify
Deprecation notice
The Classify API has been deprecated on Feb 28, 2025.
Recommended alternative: Update to the 1.3 version of the API and use the Pegasus video understanding model to classify videos.
Resources: Migration guide > Use Pegasus to classify videos.
Create text, image, and audio embeddings
This workflow guides you through creating embeddings for text.
Steps:
Note
Creating text embeddings is a synchronous process.
Create video embeddings
This workflow guides you through creating embeddings for videos. For an interactive implementation using the Python SDK, see the Quickstart Embed Jupyter notebook.
Steps:
Extract video data
Follow the steps in this guide to extract various types of data from your videos, such as transcriptions, on-screen text, logos, and thumbnails.
Steps:
- Create an index, enabling the Marengo video understanding engine.
- Upload videos and monitor the processing.
- Depending on your use case: