# Five Talks Gave 41 Clips out of 153 Scored Moments

Five real Snipri jobs on conference talks, counted: 153 moments scored, 47 over the bar and 41 clips, 6 to 10 a talk. With the method and what it cannot show.

- Page: https://www.snipri.com/blog/how-many-clips-per-hour
- Author: [Cihan Düzgören](https://www.snipri.com/about), who builds and runs Snipri, and wrote this guide
- Topic: [Clipping long videos](https://www.snipri.com/blog/topics/clipping)
- Published: November 13, 2026

![Data study, clipping: five talks, 153 moments, 41 clips. 47 of 153 moments reached the bar of 7 out of 10; 6 to 10 clips a talk, since one job returns at most 10; the clips ran 21 to 60 seconds, 40 at the median. Beside it, three columns drawn to scale: 153 moments scored, 47 over the bar and 41 clips delivered.](https://www.snipri.com/blog/how-many-clips-per-hour/cover.webp)

**Short answer.** In five real Snipri jobs on conference talks of 40 to 53 minutes, run on 2 and 3 November 2026, the selection scored 153 moments, 47 reached its bar of 7 out of 10 and 41 became clips: 6 to 10 a talk, since one job returns at most 10. Moments over the bar came to 12.6 for each hour of talk. That is Snipri's own count of these five talks, not a rate for every video.

- In five real Snipri jobs run on 2 and 3 November 2026, the selection scored 153 moments in 3 hours 44 minutes of conference talks, and 47 of them reached its bar of 7 out of 10.
- In that count 41 moments became clips, 6 to 10 clips a talk. One Snipri job returns at most 10 clips, and the one talk with 16 moments over the bar left 6 of them out.
- Across the five talks, moments over the bar came to 12.6 for each hour of talk, from 8.9 in the talk with slides to 18.1 in the lecture.
- The 41 clips of the five jobs ran 21 to 60 seconds, with a median of 40 seconds. Together they hold 29 minutes, which is 12.9% of the talk time.
- Clips came from every third of every one of the five talks: 14 from the first thirds, 12 from the middle thirds and 15 from the last.
- In the five jobs, 12 of the 41 clips sit exactly on 7 on all three gates, the lowest scores a clip can have, and no clip scored a 10 on anything.

Nobody can say how many clips an hour of talk holds in general, but five talks can be counted. Every figure below comes from five real jobs Snipri ran on conference talks on 2 and 3 November 2026, counted on 13 November 2026 by a script that reads the jobs' own records. The talks, the method, the data and what the numbers cannot show are all on this page, and each job is printed whole on a page of its own.

## How many clips did five talks give?

6 to 10 clips a talk, 41 clips in all. The five talks ran 40 to 53 minutes each, 3 hours 44 minutes together. In them Snipri's selection scored 153 moments, a moment being a stretch of a talk it put forward as a possible clip. 47 of those reached the bar, which is 7 out of 10 on each of three scores, and 41 became clips:

| Talk | Kind picked in the Studio | Length | Moments scored | Over the bar | Clips delivered | Its job, whole |
| --- | --- | --- | --- | --- | --- | --- |
| Fireside chat, GodotFest 2025 | Podcast | 46 min | 30 | 7 | 7 | [Podcast clip maker](https://www.snipri.com/tools/podcast-clip-maker) |
| Talk with slides, 38C3 | Webinar | 40 min | 28 | 6 | 6 | [Webinar clip maker](https://www.snipri.com/tools/webinar-clip-maker) |
| On-stage interview, Hacker Hotel 2025 | Q&A | 41 min | 20 | 8 | 8 | [Interview clip maker](https://www.snipri.com/tools/interview-clip-maker) |
| Keynote, Hacker Hotel 2025 | Auto | 44 min | 37 | 10 | 10 | [YouTube Shorts maker](https://www.snipri.com/tools/long-video-to-shorts) |
| Lecture, WHY2025 | Academic | 53 min | 38 | 16 | 10 | [YouTube clip maker](https://www.snipri.com/tools/youtube-clip-maker) |
| All five | One of each | 3 h 44 min | 153 | 47 | 41 | This page |

The last two counts differ in one row. One Snipri job returns at most 10 clips, the strongest ones, and in the lecture 16 moments reached the bar, so 6 were left out. In the other four talks every moment over the bar became a clip. That is why this post keeps two numbers apart: moments over the bar say what the selection found, and clips delivered say what one job hands back.

![A chart titled 41 clips out of 153 scored moments. Five rows of squares, one square for each moment a talk's job scored, sorted by what became of it. Fireside chat, 46 min: 30 scored, 7 over the bar, 7 clips. Talk with slides, 40 min: 28 scored, 6 over the bar, 6 clips. On-stage interview, 41 min: 20 scored, 8 over the bar, 8 clips. Keynote, 44 min: 37 scored, 10 over the bar, 10 clips. Lecture, 53 min: 38 scored, 16 over the bar, 10 clips. Key: 41 clips delivered; 6 over the bar and left out, because one job returns at most 10 clips; 20 close calls kept on record, the four closest of each talk; 86 more under the bar, counted and not kept.](https://www.snipri.com/blog/how-many-clips-per-hour/moments-by-talk.webp)

*Every moment Snipri's selection scored in the five jobs, a square each, by what became of it. A drawing from the jobs' counts: the squares are sorted by outcome, not by where a moment sits in its talk. The talks are credited in the next section.*

## What was counted, and how?

Five talks, one job each, and the records a job leaves behind. The talks were chosen for Snipri's use-case pages: recordings whose licence allows reuse with credit, one for each kind of video those pages are about. They were not picked for how many clips they would give, and they are not a random sample of anything.

- **The day.** The jobs ran on 2 and 3 November 2026 in Snipri's own account, and the count was made on 13 November 2026.
- **The settings.** Each job ran with one kind of video picked in the Studio, the kind in the table above, and every other setting at its default: Best moments and the Auto clip length.
- **The records.** A job's log gives how many moments the selection scored and how many reached the bar. The job itself keeps every clip and up to four close calls, each with its start and end, a title, five scores and a reason in a sentence.
- **The method.** A script reads those records, adds them up, writes the two data files linked below and draws this page's pictures in one run. It stops when a figure in this text no longer matches the count.

The terms, once each:

| Term | What it means here | Where it comes from |
| --- | --- | --- |
| Moment scored | A stretch of a talk the selection put forward and gave five scores from 1 to 10: hook, payoff, standalone sense, arc and quotability | The job's log |
| The bar | 7 out of 10 on hook, payoff and standalone sense, the three scores that work as gates | Snipri's selection rules, the same for every job |
| Clip delivered | A moment over the bar that the job cut, framed vertically and captioned | The job's clips |
| Close call | One of the moments under the bar that came closest. A job keeps up to four | The job's record of what it left out |
| The ceiling | The most clips one job returns, which is 10 | Snipri's selection rules |

The talks are published under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) on media.ccc.de: [“Fireside Chat: Godot’s Next Milestones”](https://media.ccc.de/v/godotfest2025-fireside-chat-godots-next-milestones) by Ben Vehling, Matt Ellis and Rémi Verschelde, GodotFest 2025; [“Find My * 101”](https://media.ccc.de/v/38c3-find-my-101) by Henryk Plötz, 38C3; [“Ask me anything, with Huib Modderkolk”](https://media.ccc.de/v/2025-224-ask-me-anything-with-huib-modderkolk) by Chris van 't Hof and Huib Modderkolk, Hacker Hotel 2025; [“Frank talks AI, keynote style”](https://media.ccc.de/v/2025-217-frank-talks-ai-keynote-style) by Frank Breedijk, Hacker Hotel 2025; and [“Reverse Engineering Life: A teardown of the DNA source code of a whole bacterium”](https://media.ccc.de/v/why2025-34-reverse-engineering-life-a-teardown-of-the-dna-source-code-of-a-whole-bacterium) by bert hubert, WHY2025. Their pages were read again on 13 November 2026. Those pages give each talk's length in whole minutes, and the lengths here are the ones the jobs measured. The scores, titles and reasons are Snipri's. The speakers and the conferences have no connection to Snipri.

## How many clips is that for an hour of talk?

About 12.6 moments over the bar for each hour of talk, taking the five talks together, and anywhere from 8.9 to 18.1 by talk:

| Talk | Moments scored, for each hour of talk | Moments over the bar, for each hour of talk | Share of scored moments over the bar |
| --- | --- | --- | --- |
| Fireside chat | 39.2 | 9.1 | 23.3% |
| Talk with slides | 41.7 | 8.9 | 21.4% |
| On-stage interview | 29.1 | 11.6 | 40.0% |
| Keynote | 50.5 | 13.6 | 27.0% |
| Lecture | 43.1 | 18.1 | 42.1% |
| All five | 40.9 | 12.6 | 30.7% |

Clips delivered work out to 11.0 for each hour of talk, and that figure needs a warning the others do not. No talk here ran a full hour, and one job returns at most 10 clips, so a single job on a one-hour talk cannot give 11. The lecture shows the ceiling at work: 18.1 moments over the bar for each hour of talk, and 10 clips. A job also stops at 10 minutes of clips in all, a limit none of the five came near: the fullest held 7 minutes 10 seconds.

The spread says more than the average. With the same bar, and every setting but the kind of video at its default, the lecture had 42.1% of its scored moments over the bar and the talk with slides 21.4%.

## How many moments did the selection turn down?

106 of the 153, each short of 7 on at least one gate. A job keeps the four that came closest, so the five jobs hold 20 close calls, each with its scores and the reason it was left out. The other 86 left nothing but their number in a job's log.

Of the 20 close calls, 15 fell short on the hook and 14 on standalone sense, and none on payoff; 11 missed one gate and 9 missed two. [Hook ideas](https://www.snipri.com/blog/hook-ideas) goes through what those openings sounded like, so it is not repeated here, and [When a video comes back with no clips](https://www.snipri.com/docs/near-miss-clips) says how a close call can still be made into a clip.

The 6 moments the ceiling left out are a different case. They reached the bar, and nothing about them was kept but their number.

The video builds the count up in that order: every scored moment as a square, the ones that reached the bar, the ones that became clips, and then where the clips sit in each talk.

[Video: Five talks counted: scored, over the bar, clips](https://www.snipri.com/blog/how-many-clips-per-hour/five-talks-counted.mp4)

*A drawing with no footage, built from the counts of the five jobs. Each scored moment appears as a square in its talk's row, 153 in all; the 47 that reached the bar light up; the 41 that became clips turn green, and the 6 the ceiling left out stay as outlines. Then each talk becomes a bar from its first second to its last, and its clips light up where they sit as a line passes: 14 in the first thirds, 12 in the middle thirds and 15 in the last. The squares are sorted by outcome, not by where a moment sits in its talk.*

## How far over the bar did the clips land?

Not far. 12 of the 41 clips sit exactly on 7 on all three gates, and no clip scored a 10 on anything:

| Score | Clips at 7 | Clips at 8 | Clips at 9 | What it measures |
| --- | --- | --- | --- | --- |
| Hook | 28 | 13 | 0 | How hard the opening seconds grab attention |
| Payoff | 20 | 19 | 2 | Whether the moment delivers a point, a punchline or an insight by its end |
| Standalone sense | 32 | 8 | 1 | Whether it makes sense with no context around it |

The two scores that are not gates run lower: 13 clips have a 6 on arc and 8 have a 6 on quotability, and they are clips all the same. Only 5 of the 41 clips hold a 9 anywhere. The scale keeps 9 and 10 for the exceptional, and in these five talks the selection used it that way.

A clip on the bar has still passed all three checks. What the table does show is how thin the margin is: with one point less on one gate, a clip that sits on the bar would have been left out. [See why a clip was chosen](https://www.snipri.com/docs/why-a-clip-was-chosen) says what each score measures and where a clip shows them, and the [AI clip maker](https://www.snipri.com/features/ai-clip-maker) page shows them on real clips.

## How long were the clips, and how much of a talk did they use?

The 41 clips ran 21 to 60 seconds, with a median of 40 seconds. Every job ran on Auto, the Studio's default length. Clips run 15 to 60 seconds by default. Before a job you can choose 10 to 30 seconds, 30 to 60 seconds, 60 to 90 seconds or 90 seconds to 3 minutes instead. None of the 41 came out under 20 seconds, and 12 ran 50 seconds or longer.

Together the clips hold 29 minutes, which is 12.9% of the 3 hours 44 minutes of talk:

| Talk | Clips | Clip time in all | Share of the talk | Shortest and longest clip |
| --- | --- | --- | --- | --- |
| Fireside chat | 7 | 4 min 33 s | 9.9% | 21 and 52 s |
| Talk with slides | 6 | 4 min 26 s | 11.0% | 31 and 57 s |
| On-stage interview | 8 | 5 min 53 s | 14.3% | 29 and 60 s |
| Keynote | 10 | 6 min 59 s | 15.9% | 32 and 55 s |
| Lecture | 10 | 7 min 10 s | 13.5% | 27 and 58 s |
| All five | 41 | 29 min 1 s | 12.9% | 21 and 60 s |

![A chart titled Clips ran from 21 to 60 seconds. The 41 clips as squares stacked by length in steps of 5 seconds: none from 15 to 20 seconds, 2 from 20 to 25, 2 from 25 to 30, 7 from 30 to 35, 8 from 35 to 40, 5 from 40 to 45, 5 from 45 to 50, 6 from 50 to 55 and 6 from 55 to 60, with a dashed line at the median of 40 seconds. Under it a bar: 12.9% of the talk time ended up in a clip, 29 min of clips in 3 h 44 min of talk.](https://www.snipri.com/blog/how-many-clips-per-hour/clip-lengths.webp)

*The lengths of the 41 clips the five jobs delivered, and the part of the talks they hold. Every job ran on Auto, which cuts clips of 15 to 60 seconds. A drawing from the jobs' own start and end times; the talks are credited above.*

## Where in a talk did the clips sit?

In every third of every talk. Counting each clip in the third of its talk that its midpoint falls in, 14 came from the first thirds, 12 from the middle thirds and 15 from the last:

| Talk | Clips in the first third | Clips in the middle third | Clips in the last third | First clip starts at | Longest stretch with no clip |
| --- | --- | --- | --- | --- | --- |
| Fireside chat | 2 | 3 | 2 | 6:13 | 9:43 |
| Talk with slides | 1 | 3 | 2 | 6:23 | 8:49 |
| On-stage interview | 1 | 3 | 4 | 7:35 | 9:07 |
| Keynote | 5 | 1 | 4 | 0:27 | 10:09 |
| Lecture | 5 | 2 | 3 | 2:51 | 13:46 |
| All five | 14 | 12 | 15 | 0:27 to 7:35 | 8:49 to 13:46 |

No third went without a clip, and no talk spread its clips evenly. The keynote and the lecture gave half of their clips in the first third, and the on-stage interview gave half of its clips in the last. In three talks the first clip started more than six minutes in. Every talk also had one stretch of 8 minutes 49 seconds or longer with no clip in it.

![A chart titled Clips came from every third of every talk. Five bars, one for each talk from its first second to its last, with its clips in green and its close calls in grey where they sit. Clips in the first, middle and last third: fireside chat, 46 min, 2, 3 and 2; talk with slides, 40 min, 1, 3 and 2; on-stage interview, 41 min, 1, 3 and 4; keynote, 44 min, 5, 1 and 4; lecture, 53 min, 5, 2 and 3. All five, 3 h 44 min: 14, 12 and 15 clips.](https://www.snipri.com/blog/how-many-clips-per-hour/where-clips-sit.webp)

*Where the 41 clips and the 20 close calls of the five jobs sit in their talks. Each talk is drawn at one width whatever its length, and a clip counts in the third its midpoint falls in. A drawing from the jobs' own start and end times; the talks are credited above.*

## What do these numbers not show?

Most of what a reader would like them to. Five limits, plainly:

- **Not a rate.** Five talks are five talks. All are conference talks in English, 40 to 53 minutes long, chosen because their licence allows reuse and because each matched a page. Nothing here says what a two-hour podcast, a sermon, a stream or a talk in another language gives.
- **Not a repeat.** Each talk was run once, on the defaults, with one kind of video picked. Another run, another clip length or a line in Moments to look for could change the count, and none was tried.
- **Not a cause.** The lecture had the most moments over the bar and the highest share of them. The talks differ in speaker, subject, length and the kind of video picked, all at once, so a count cannot say which of those made the difference.
- **Not a measure of results.** The scores are the selection's judgement of a transcript, made before anyone watched a clip. Nothing here says how a clip does once it is posted, and this post has no viewing figures.
- **Not all 153 moments.** A job keeps its clips and its close calls. The other 92 moments, 86 under the bar and the 6 the ceiling left out, exist only as numbers in a log, so the sections on scores, lengths and places describe the 61 moments on record.

## Where is the data?

In two CSV files, free to download:

- [snipri-five-talks.csv](https://www.snipri.com/blog/how-many-clips-per-hour/snipri-five-talks.csv) has a row for each talk: its title, speakers, event, the day it was recorded, its address and licence, the kind of video picked, its length in seconds, and its counts, which are moments scored, moments over the bar, clips delivered, moments left out by the ceiling and close calls kept, with the clips' seconds in all.
- [snipri-five-talks-moments.csv](https://www.snipri.com/blog/how-many-clips-per-hour/snipri-five-talks-moments.csv) has a row for each of the 61 moments on record: its talk, whether it is a clip or a close call, its rank among its job's clips or close calls, where it starts and ends in seconds, its length, the title Snipri wrote for it, its five scores, the gate it fell short on if it did, and the reason Snipri gave.

Both are UTF-8 text with a header row, written by the script that draws this page's pictures, in the same run, from the jobs' records. The files hold no audio, video or transcript of a recording. Every title and reason in them is Snipri's wording about a moment, some of the reasons quote a few of a speaker's words, and each row names the talk it was counted from.

The count can also be checked a job at a time, on the five pages linked in the first table. Each prints its job whole: the talk drawn to scale with a mark at every clip, how many moments the selection scored in it, every clip with its place, its five scores and the reason Snipri gave, and the close calls with the score that kept each one out. The counts in this post are those five pages added up.

Nothing here is cut from a recording: the pictures and the video on this page are drawings of numbers. If you reuse the recordings themselves, credit their speakers as [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) asks.

## What can you take to your own recording?

Three things, each only as far as five talks go:

1. **Expect to turn down more than you keep.** The selection kept 41 of the 153 moments it had already found worth scoring. If every moment you mark becomes a clip, it is worth asking what your bar is.
2. **Go through the whole recording.** Clips came from every third of every talk, and the first one started anywhere between 0:27 and 7:35.
3. **Do not cut to a number.** The same bar, with every setting but the kind of video at its default, found 6 moments over the bar in one talk and 16 in another. [How to make podcast clips that stand alone](https://www.snipri.com/blog/how-to-make-podcast-clips) has five checks to run on each candidate by hand.

More on choosing moments is under [Clipping long videos into short clips](https://www.snipri.com/blog/topics/clipping).

## Common questions

### Can one Snipri job return more than ten clips?

No. One job returns at most 10 clips, the strongest of the moments that reached the bar. In this count the lecture had 16 such moments, so 6 were left out, and nothing about them was kept but their number.

### Does a longer video give more clips?

Five talks cannot say. The longest talk here, a lecture of 53 minutes, had the most moments over the bar, but the talks also differ in speaker, subject and kind, and each was run once. One job returns at most 10 clips whatever the length.

### What happens to a moment that just misses the bar?

Snipri keeps up to four of them in each job as close calls, each with its scores, the gate it missed and the reason. When a job ends with no clips it lists them, and you can make one into a clip.

### Would a second run on the same talk give the same clips?

This count cannot say, because each talk was run once. The selection is a model's judgement of a transcript, and nothing here measures how much it changes from one run to the next.

### Can I download the data behind this count?

Yes. Two CSV files on this page hold it: one row for each of the five talks and one row for each of the 61 moments on record, each moment with its five scores and the reason Snipri gave. They hold no audio, video or transcript of a recording.

### Were the speakers or the conferences involved in this count?

No. The talks are published under CC BY 4.0 on media.ccc.de, which allows reuse with credit. The scores, titles and reasons are Snipri's, and the speakers and the conferences have no connection to Snipri.

## Sources

1. [media.ccc.de: Fireside Chat: Godot’s Next Milestones, GodotFest 2025](https://media.ccc.de/v/godotfest2025-fireside-chat-godots-next-milestones), checked November 13, 2026.
2. [media.ccc.de: Find My \* 101, 38C3](https://media.ccc.de/v/38c3-find-my-101), checked November 13, 2026.
3. [media.ccc.de: Ask me anything, with Huib Modderkolk, Hacker Hotel 2025](https://media.ccc.de/v/2025-224-ask-me-anything-with-huib-modderkolk), checked November 13, 2026.
4. [media.ccc.de: Frank talks AI, keynote style, Hacker Hotel 2025](https://media.ccc.de/v/2025-217-frank-talks-ai-keynote-style), checked November 13, 2026.
5. [media.ccc.de: Reverse Engineering Life, WHY2025](https://media.ccc.de/v/why2025-34-reverse-engineering-life-a-teardown-of-the-dna-source-code-of-a-whole-bacterium), checked November 13, 2026.
6. [Creative Commons: Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/), checked November 13, 2026.

Read next: [Hook Ideas: Openings Found in Your Recording](https://www.snipri.com/blog/hook-ideas)
