Skip to main content

Long-context Outputs Demo

Navigate over to the video-transcription playground in our playground to see the long-context outputs in action.
VLM Run provides robust support for processing and extracting structured data from long-context inputs like audio and video files. This capability enables you to work with extended content that would exceed the token limits of many foundation models (typically around 8K tokens).
This feature is currently only available for our enterprise-tier customers. If you are interested in using this feature, please contact us.

What are Long-context Outputs?

Long-context outputs refer to VLM Run’s ability to process and extract structured data from extended output content like:
  • Transcripts of long-form audio or video content (12+ hours of audio or 4+ hours of video)
  • Extracted structured data from multi-page documents (>128 pages)
  • Extracted structured data from large collections of related images (>128 images)
This capability is essential for applications that deal with lengthy content such as podcast transcriptions, lecture recordings, interviews, meetings, and extended video analysis.

Using Long-context Processing

You can process long-form audio and video content using the VLM Run API with batch processing enabled:

Long-form Audio / Audio Processing (with batch processing)

For all long-form audio and video processing, we only support batch=True mode.
For batch processing, you are provided with a prediction ID and can check the status of the prediction later. We provide a polling mechanism and some convenience functions to check the status of the prediction and wait for it to complete.

Domain-specific Schemas

VLM Run provides specialized schemas for different types of long-form content:
  • audio.transcription: General-purpose audio transcription with speaker detection
  • video.transcription: General-purpose video transcription with visual scene analysis
  • video.transcription-summary: Summary of a video transcription with key points and speaker analysis
  • video.conferencing-summary: Summary of a video conference with key points and speaker analysis
  • video.tv-news-summary: Summary of a TV news broadcast with anchors, reporters, chyrons, and segments
  • video.dashcam: Analysis of a dashcam video with scene analysis and spoken language detection
Refer to the Hub Catalog for more information on the schemas supported by VLM Run.

Example: Transcription of a YC Podcast Episode

Here’s an example of a long-context output for a YC episode on How New Technology Creates New Businesses. As you can see, the output is a list of temporal segments grounded with start and end times, both audio transcription and visual understanding of the content.
YC Podcast Transcription

Use Cases

  • Content Search: Make audio/video content searchable through transcription
  • Meeting Intelligence: Extract action items and key points from meeting recordings
  • Media Monitoring: Analyze news broadcasts and identify topics and speakers
  • Educational Content: Structure course lectures with chapters and topics
  • Podcast Production: Generate show notes, summaries, and topic timestamps
By leveraging VLM Run’s long-context output capabilities, you can efficiently extract structured information from extended audio and video content that would otherwise exceed traditional token limits.

Try our Video / Audio -> JSON API today

Head over to our Video -> JSON or Audio -> JSON to start building your own video/audio processing pipelines with VLM Run. Sign-up for access on our platform.