ClairaClaira Help Desk
AI ReviewPromptingTemplatesMultimedia

Video Summary

Other languages

Video Summary

Use this template to summarize what a video shows and says, with timestamps, so reviewers can triage footage without watching it end to end.

ClairaPrompt LabQuickstartPrompt LibraryImages, audio and videoVideo Summary

Insert Template puts this prompt into the prompt box. Claira writes each answer to the destination field you choose: a Memo field, since the summary runs over several lines.

Video summaryMemoclaira-video-summary-01

What this prompt is for

This template writes a factual summary of a video under fixed headings: the setting, timestamped events, what is said, on-screen text and the recording's limitations.

Video evidence (surveillance clips, body camera footage, recorded interviews and depositions) has no extracted text to review, so every minute normally has to be watched. The summary tells a reviewer what happens and when, so they can go straight to the moments that matter, and its limitations line says what the recording cannot show before anyone relies on it.

Use it to triage a collection of recordings, or to find the clips worth watching in full. Do not use it as a transcript: speech is summarized, not written out word for word. A video with no picture worth describing, such as a recorded call, is better scanned as Audio with a prompt of its own.

Step-by-step in Claira

  1. In Prompt Lab, click Quickstart above the prompt box. Bulk Scan has the same button.
  2. Under Prompt Library, open Images, audio and video. On the Video Summary card, click Expand to read the prompt, then click Insert Template.
  3. If the prompt box already has text, Replace existing prompt? asks what to do with it. Replace keeps the prompt tied to this template; Append adds the template below your text.
  4. Set Scan as, next to Scan document, to Video, or leave it on Auto, which sends the video file whenever the document has one. Video scans need a Gemini model. See Media scans for the accepted formats, the length cap of about 40 minutes and which models can run a Video scan.
  5. Open a video in Nuix Discover and click Scan document. The answer appears under AI Response. To save it, choose a Memo field in Apply to field... and click Apply. Do this on a small sample and check a few summaries against the footage. Single scans are free.
  6. When the sample reads well, select the videos in Nuix Discover and click Start bulk scan. In Ready to start bulk scan, check the Destination field, then click Start bulk scan.

The prompt

Summarize this video recording from what can be seen and heard in it.

Report only what is visible or audible. Do not guess at intent, at who someone is, or at events outside the recording. Name a person only when the recording shows or says the name; otherwise describe them, for example the driver or a man in a red jacket. Give timestamps from the start of the file as mm:ss.

If there is no audio track or no speech, say so under SPEECH. If parts of the picture or sound cannot be made out, say so under LIMITATIONS. If the document has no video to watch, answer exactly:
NO VIDEO FOUND

Answer in exactly this form:
SETTING: one or two sentences on where the recording takes place, the date and time if the recording shows them, and who and what appears
EVENTS: the main events in order, at most 10, each on its own line starting with its mm:ss timestamp
SPEECH: who says what, in one to three sentences, or No speech and the reason
ON-SCREEN TEXT: signs, labels, captions, number plates and any date or time shown on the picture, copied as written, or None
LIMITATIONS: one or two sentences on lighting, camera movement, cuts or gaps, and missing or unclear audio, or None
  • Looking for something specific (a person, a vehicle, an exchange)? Add a sentence such as "Give the timestamp of every appearance of the white van." and keep the headings as they are.
  • Raise the limit of 10 events for long recordings such as interviews, and lower it for short clips.
  • Where the exact wording matters, change the SPEECH line to ask for the key statements as short quotes, each with its timestamp.
  • Keep NO VIDEO FOUND as written if you scan with Auto: searching for it separates the documents that had no video from the summaries.

Worked example

Input excerpt

A 90-second clip from a car park camera at night, with no audio track. The picture carries the overlay "CAM 3 LOADING BAY 2025-03-14" and a running clock from 23:41:07. At 00:05 a white van reverses into the loading bay; at 00:20 two people get out and open its rear doors; at 00:35 they unload six boxes onto a trolley; the clip ends at 01:30 while they are still at the van. In the low light the two people's faces cannot be made out, and the van's number plate cannot be read.

Expected output shape

SETTING: A fixed camera overlooking a loading bay in a car park at night, on 14 March 2025 from 23:41:07 by the picture's overlay; a white van and two people appear.
EVENTS:
00:05 A white van reverses into the loading bay.
00:20 Two people get out of the van and open its rear doors.
00:35 The two people unload six boxes onto a trolley.
01:30 The clip ends while both people are still at the van.
SPEECH: No speech; the recording has no audio track.
ON-SCREEN TEXT: CAM 3 LOADING BAY 2025-03-14, with a running clock starting at 23:41:07.
LIMITATIONS: Low light hides the two people's faces, the van's number plate cannot be read, and the clip ends before they leave.

Troubleshooting

  • The scan fails. Check the file against the accepted video formats and the cap of about 40 minutes in Media scans. Longer files need to be split before scanning.
  • NO VIDEO FOUND on a video. The scan sent text or an image instead of the video. Set Scan as to Video and scan the document again.
  • Speech is missing. The audio track may be absent or inaudible: the SPEECH and LIMITATIONS lines should say which.
  • Interpretation creeps in, such as why someone did something. Add a sentence: "Describe actions, never reasons." and rerun the sample.
  • A timestamp is a few seconds off. Timestamps are approximate. Check the footage before quoting one.

Was this page helpful?

Need more help?

Contact our support team at support@claira.to — we are here to help.

On this page