Video Content Summary
Summarise a video into a report with scene captures. I will give a public YouTube link or attach a video or audio file. Return a report that explains the video section by section with timestamps, places a capture of the actual scene next to the sections whose picture matters, and lists the numbers and claims the video states and the points to check.
Run this
This is the example task. Press Run Task to start it in a new chat — you can change the highlighted parts there before sending.
Summarise this lecture into study notes with the key slides — https://www.youtube.com/watch?v=b8u2CQLQBVU
Agent works for you
The agent runs these steps in order and does not stop to ask you questions along the way.
- Check and read the video
- Re-read key parts
- Sections and scene captures
- Report
- Answer
Get this
A sample result from a real run. Click the image to open the full output.
Explore similar agents
Video Content Summary: Extracting and analysing data
Give a public YouTube link or upload a video. The agent reads what is shown and said, explains the video section by section with timestamps, and places a capture of the actual scene next to each description — in a report you can keep.
Who this is for
- Researchers who need notes from lectures, conference talks or seminars
- Students turning a long lecture into study notes with the key slides
- UX and HCI researchers reviewing a recorded session or demo
- Anyone who must report what a video showed, at which moment
What you get
One report. Every time section has a short description with its timestamps, and the most informative sections carry a capture of that moment.
- Video content summary report: The whole video, section by section · Summary, key points with timestamps, section-by-section notes with scene captures, numbers and claims the video states, points to check · MD
How to ask
Summarise {the video} — {what you care about}. — Change the underlined parts.
- the video: a public YouTube link, or attach a video or audio file
- what you care about: e.g. the method and results, the steps of a tutorial, the answers to each question
You can attach
- Video (MP4 · MOV · WEBM · MKV · M4V · AVI) or audio: Read on our server: speech is transcribed word for word and scenes are captured in original quality
You can also ask
- Summarise this conference talk as research notes: question, method, results, limitations — https://youtu.be/…: → A research talk is organised as question → method → results
- Make study notes from the attached lecture with the key slides.: → Slides are captured at the moment they appear
How it works
Code reads the video and captures the scenes; the model only writes the descriptions from what was read and never adds what the video does not show or say.
- Check the video
What happens: Finds the YouTube link in your request or the attached video, and its title and length · Depends on the level: —
- Read the timeline
What happens: YouTube: an AI video model reads picture and speech segment by segment. Upload: speech is transcribed and key frames are described · Depends on the level: Lite 15 min · Pro 30 min · Expert 1 hour · Max 2 hours
- Re-read key parts
What happens: The densest parts of a YouTube video are read again in finer detail · Depends on the level: Expert 2 parts, Max 3 parts
- Choose sections and scenes
What happens: Splits the video into time sections and picks the moments whose picture says the most (slides, charts, demos) · Depends on the level: Sections: 5 · 10 · 16 · 24
- Capture scenes
What happens: Captures the chosen moments — original quality for uploads, YouTube preview (storyboard) quality for links · Depends on the level: Captures: 3 · 6 · 10 · 16
- Write the report
What happens: Places each capture next to its section; Max removes sentences the timeline does not support · Depends on the level: Max checks every statement
What changes with the level
- Time sections (most)
- 5
- Scene captures (most)
- 3
- Length read
- First 15 min
- Re-read key parts (YouTube)
- —
- Every statement checked against the timeline
- —
| Lite | Pro | Expert | Max | |
|---|---|---|---|---|
| Time sections (most) | 5 | 10 | 16 | 24 |
| Scene captures (most) | 3 | 6 | 10 | 16 |
| Length read | First 15 min | First 30 min | First 1 hour | First 2 hours |
| Re-read key parts (YouTube) | — | — | 2 | 3 |
| Every statement checked against the timeline | — | — | — | ✓ |
Longer videos are packed into the same number of sections, so each section covers more time. A video longer than the level's length is summarised from the start, and the chat says so and which level reads longer.
Sources
- Public YouTube link: How it is read: AI video model (Gemini) reads the link; nothing is downloaded · Scene captures: YouTube preview images (about 320×180), marked in the report
- Uploaded video: How it is read: Whisper transcript + key frames described on our server · Scene captures: Original frames (up to 960 px)
- Uploaded audio: How it is read: Whisper transcript · Scene captures: —
- Private, unlisted, age- or region-restricted and live YouTube videos cannot be read by link — upload the file instead.
FAQ
- For uploads, yes (transcript). For YouTube links the AI model writes the gist of what is said, so the report marks it as a gist.