The platform analyzes videos to generate text based on their content using a multimodal approach. This method analyzes the visuals, sounds, spoken words, and relationships between them.

The page uses a two-panel layout. Enter your prompt and configure parameters in the left panel. Generated text appears in the right panel.

Notes

This feature is available only for indexes with the Pegasus video understanding engine enabled. See the Create an index section for details.
Your prompts can be instructive, descriptive, or phrased as questions. For guidance, see Prompt engineering.
The maximum prompt length is 2,000 tokens.

Procedure

Open the Indexes page.
Select the index containing the video you want to analyze.
Select the Analyze tab.
Select a video.
In the left panel, enter a prompt or select a suggested prompt.
(Optional) Adjust the parameters under Advanced Settings. To reset all parameters to their default values, select the Reset button.
(Optional) To receive a structured JSON response, use the structured_output section under Advanced Settings. Select a template from Start with templates, or enable the structured_output toggle.
- If you enable the toggle, the Builder panel opens immediately. Define your schema in the Builder panel (visual interface) or the Editor panel (raw JSON). Changes in either panel reflect immediately in the other.
- If you select a template, a template dropdown and a JSON schema preview appear. To customize the schema, select Edit in Builder.
Select the Analyze button.

View results

Generated text appears in the right panel. Select the Visual or JSON tab to switch between views.

Visual tab

The Visual tab shows the generated text.

JSON tab

The JSON tab shows the raw JSON response.

Work with results

Play a specific moment

If the generated text includes timestamps or segments, select one in the Visual tab to play it.

View the code snippet

Select View code to see the request as cURL, Python, or JavaScript. Copy and paste the snippet into your application.