Paste a link and get the TLDR, the key points with timestamps you can click, and the topics the video covers. No signup for your first summaries, and every point links back to the second it came from so you can check it.
2 free summaries per day, no signup needed. Summarising draws on a tighter allowance than plain transcript extraction, because it spends model tokens on top of the fetch.
There is no magic reading of the video here. The tool does two things in sequence, and it is worth knowing which is which, because the second one is where the errors come from.
youtu.be short link, a Shorts link, an embed link or the bare 11-character video ID all work. Your browser never talks to YouTube and never holds a key.Nothing is stored against your name, because there is no account involved. Copy the TLDR and points as plain text with one button.
Summarising is the most expensive thing on this site to give away. Fetching a caption track costs bandwidth and a little compute; summarising it costs that plus model tokens, and long videos cost proportionally more of them. So the two tools draw on different allowances, and we would rather say that outright than let you discover it as a surprise error.
| Tool | Anonymous daily allowance | What it costs us |
|---|---|---|
| This summariser | 2 AI actions per day | Caption fetch plus model tokens |
| Transcript generator | 3 extractions per day | Caption fetch only |
| Word counter | Shares the extraction allowance | Caption fetch only — the maths happens in your browser |
The counter above always shows the number the API reports back, using its own wording for what is being counted, so what you see is the real remaining balance rather than our guess at it. A free account comes with credits, no card asked for.
The summary is only ever as good as the captions underneath it, and those are often machine-written. This matters more for a summary than for a raw transcript, because a mistranscribed word in a transcript is visibly odd, while the same mistake absorbed into a fluent summary sentence reads as fact.
Nothing here is authoritative. It is a model's reading of a caption track that may itself be wrong. Use it to decide whether a video is worth your time and where to start watching — then click through to the timestamp and confirm anything you intend to repeat.
If you would rather write your own prompt, or summarise a whole channel on a schedule, the API returns the raw transcript and leaves the summarising to you. See the API reference or the cookbook for working examples.
Paste the video link into the box above and press Summarise video. The page fetches the caption track, sends it to a model, and gives you back a TLDR paragraph, the key points with timestamps, and the topics covered. There is no signup for the first two summaries each day and nothing to install.
Summarising spends model tokens on top of the bandwidth needed to fetch the captions, so it costs real money per call in a way plain extraction does not. Anonymous visitors get two AI actions a day here, against three plain transcript extractions on the transcript tool. A free account raises both.
Yes. Every key point carries the moment in the video it was drawn from, and clicking it opens YouTube at that second in a new tab. That is the check on the summary — if a point matters to you, jump to it and hear what was actually said.
It is a summary of the captions, not of the video, and auto-generated captions get names, acronyms and overlapping speech wrong. The model can also flatten a nuance or drop a caveat. Treat the output as a map of the video, not as a source you can quote — the timestamps are there so you can verify anything you plan to rely on.
Very long transcripts are summarised from the first portion rather than the whole thing, and the result says so plainly when that happens. If you need the full text of a long video, use the transcript tool or the API and run your own summarisation over all of it.