AI Video Analyzer
Scene notes, on-screen text, transcript and summary. Works on videos with no speech.
Reads the frames: scenes and the text on screen. One minute per minute of video.
Free to try, no sign-up. Watch is one minute per minute of video. Both is two.
Have a file instead? Watching a local file runs in the command-line tool or the Mac app.
How it works
- 1
Paste a link
A public video link. For a file, use the command-line tool or the Mac app.
- 2
Scribiz watches
It groups the frames into scenes and transcribes any speech.
- 3
Open On screen
Scene notes with timestamps and the text read from the frames.
Scene notes and on-screen text
- 0:00
A spreadsheet is open with a column of dates and a column of amounts.
On screenQ3 expensesAdd row
- 0:08
The cursor selects the amount column and opens a formatting menu.
On screenFormatCurrency
- 0:15
A bar chart is inserted beside the table.
On screenInsert chart
What the video analyzer looks at
It samples a frame every second or two and groups them into scenes. Each scene gets a timestamp, a one sentence description and the text on screen.
A flash under a second can be missed. Pick Watch for a silent video, and Watch and listen when the speech is worth keeping. Watch is one minute per minute of video, and Both is two.
It describes what is visible. Faces are described, never named. Scene text is a description of the picture, not a record, so check names and numbers. If you want the speech first and the screen second, video to text returns both.
What the video analyzer returns
Use it for what a transcript cannot give you: the screen of a tutorial, the frames of an ad, the slides of a lecture. The scene list is a timeline you can scan quickly, with a time on each scene. Nothing worth noting on screen? The result says so. The run still uses its minutes.
| Video | What you get |
|---|---|
| Silent screen recording | The steps in order, with the menus and labels that were visible. |
| Product ad | The shots, the claims shown as text and the offer on screen. |
| Slide lecture | Slide titles and bullet text with timestamps. |
For people and for agents
The On screen tab is for reading. The same notes are in the Context file, which you can paste into an AI chat, and in the JSON that the Context object describes.
If you want a short answer, not scene notes, use the video summarizer. The MCP server gives an agent in Claude or Cursor the transcript, summary and chapters. It does not read scene notes without an API key.
Limits and allowances
- Public links on the web. Watching a file runs in the command-line tool or the Mac app.
- It describes faces but never names people.
- Without an account: 1 free video, up to 2 hours.
Frequently asked questions
What does the analyzer see?
Frames sampled a second or two apart, plus the soundtrack when there is speech. It returns scenes with a timestamp, a one sentence description and the text on screen.Does it work without audio?
Yes. A silent screen recording gives scene notes and on-screen text, which is the main reason to use Watch.How are scenes decided?
A new scene starts when the picture changes in a way that matters: a new slide, a new app screen, a cut. A static stretch stays in one scene.Can it make things up?
Scene text is a model's description, not a record. It can misread small or fast text, so check anything that matters. It does not identify people.How long can the video be?
2 hours without an account or with a free account, 6 hours on Pro. Watch counts as 1 minute per minute of video and Both as 2.What happens to my video?
For a YouTube link, Google's model reads the public video and our server downloads nothing. For other sites and direct media links, our server downloads the audio and, for Watch, a low-resolution copy, sends them to Google and deletes both when the job ends. The result is a private link, kept 24 hours, or 30 days with an account. Privacy and data has the full list.
Use it from Claude, Cursor, Codex or your code
Got a video? Get the transcript.
Or


