Skip to content
Nappai Documentation

Describe Video

Describe Video analyses a video with an AI model that can see images. The component takes several frames spread across the video and sends them to the model. If you connect a transcriber, it also transcribes the audio track and the model answers taking into account both what is seen and what is said. With the Prompt Text field you decide what you want to know.

If you connect several videos, each one is analysed separately and you get one description per video.

When to use it

  • Summarise training or meeting recordings to know what they cover without watching them in full.
  • Review the videos a customer sends to assess damage or an incident.
  • Write descriptions for the product videos in your catalogue.
  • Check whether a video meets your requirements before publishing it.

Configuration

FieldWhat it does
Video Data (required)The video or videos you want to analyse. Connect the file coming from another component, a record with the video’s content or address, or a message with the video’s address.
Model (required)The AI model that analyses the video. It must be a model that can see images.
Audio TranscriberOptional. The transcriber that turns the video’s audio into text. Without it, only the frames are analysed.

Advanced options

FieldWhat it does
Prompt TextWhat you want the model to do with the video. Default: “Describe the video.”

The model receives at most eight frames per video, spread across the whole recording, together with the full audio transcript when a transcriber is connected.

Output

The Data output returns one result per video with:

  • video_description: the model’s answer.
  • name: the video file’s name.
  • url: the video’s address, if it had one.
  • mimetype: the file type.

Example

An insurer receives videos of damage to a home recorded by its customers.

  1. Connect the received video to Video Data.
  2. Connect a vision-capable model to Model and, to take the audio into account, a transcriber to Audio Transcriber.
  3. In Prompt Text, write: “List the damage visible in the home, which rooms are affected and what the customer explains about how it happened.”
  4. Send the description to the loss adjuster together with the case file.

Good practices

  • Be specific in Prompt Text. If you need an answer in a fixed format (for example, a list), ask for it explicitly.
  • Since the model only sees a few frames, brief details in long videos may be missed. For what matters, rely on what is said in the audio too.