11/20/2026
29 of 91: What the Numbers Say About the Clips We Used
Short answer: The Coach Grinkler premiere uses 29 distinct clips. Of the 29, 19 come from the 81 clips in Media Hub's AI folder, about one in four; the other 10 sit in a folder called Unmatched and were filed after the 91 were counted, so 29 of 91 is not a like-for-like ratio. Seventeen of the 29 carry a person's keeper mark, and the other 12 have no mark at all. None of the 31 clips marked blooper or unusable in the AI folder made the cut, but 27 of its keepers did not either. If you are budgeting for AI video, plan to generate about four clips for every one you use.

Where these numbers come from
The saved cut list for the premiere names every clip by its Media Hub key. A short script, written by Claude, read Media Hub's record for each of the 29 keys on 11/17/2026 and tallied the marks, sizes and measurements below. Every count here comes from those records, except the film's running time and the master's size, which come from the finished file.
One number could not be reproduced. The "91" comes from our earlier post about sorting the clips, published 11/06. Media Hub does not hold the 91 as one group today: its AI folder holds 81 clips, and another 20 sit in Unmatched. Ten of the 29 clips we used are in Unmatched, and the file names of those ten carry dates of 11/14 and 11/15, after the sorting post. That suggests they were not among the 91, which is why the like-for-like figure is 19 of 81. Wherever this post compares the clips we used with the clips we did not, it compares against the 81 in the AI folder, and says so.
The 91 also counts clips, not shots. The library holds more than one take of some shots, and the cut list's own notes call one clip a "runner-up" and say a repeated wrestling commentator was replaced. So "four clips for every one you use" is a count of generated clips per clip used.
The 29 at a glance
| What was measured | Clips (of 29) |
|---|---|
| Carry a person's keeper mark | 17 |
| No mark recorded | 12 |
| 1080p | 16 |
| 720p | 13 |
| Media Hub found a freeze range in the picture | 9 |
| Media Hub detected Google's sparkle mark | 3 |
| Total length of all 29 clips | 256.3 seconds |
The film runs 2:26, or 146 seconds, and the last 7 seconds are the end card. By our arithmetic that is about 139 seconds of picture, drawn from 256 seconds of clips, so even the clips that made the cut were trimmed to a little over half their length.
Who decided: the marks
A keeper mark is a person's verdict on a clip, saved in Media Hub with a note and a date. Here is how the 81 clips in the AI folder split between the 19 that were used and the 62 that were not:
| Verdict | Used (19) | Not used (62) |
|---|---|---|
| Keeper | 17 | 27 |
| Blooper | 0 | 11 |
| Unusable | 0 | 20 |
| No mark | 2 | 4 |
Two things stand out. The marks fit the cut: nothing marked blooper or unusable was used, and 17 of the 19 used clips were keepers, though 12 of the 29 clips in the film were never marked at all. And a keeper mark is not a promise of use. Of the folder's 44 keepers, 17 were used, which is fewer than two in five.
The other 12 clips of the 29 have no mark. Ten of them sit in Unmatched, and the other two are replacement takes of earlier clips. So a cut can use a clip nobody ever marked, and a missing mark has to be read as "not yet judged", never as "approved".
The file names say when each of the 12 arrived. Seven were filed on 11/14, as 1080p versions. Three were filed on 11/15, one of them the painted tablet clip. Two are the 1080p replacements Media Hub links to earlier takes, dated 11/01.
Resolution: the mix, and what the master is
| Resolution | Used, AI folder (19) | Used, Unmatched folder (10) | Not used, AI folder (62) |
|---|---|---|---|
| 1080p | 8 | 8 | 13 |
| 720p | 11 | 2 | 49 |
The clips we used lean toward 1080p. Inside the AI folder, 8 of the 19 used clips are 1080p (42 percent), against 13 of the 62 unused ones (21 percent). The 10 used clips in Unmatched add 8 more at 1080p, which is how the total reaches 16 of 29. The records do not say whether a clip was picked because it was 1080p or downloaded at 1080p because it was picked, so this post does not claim either.
What the records do show: no clip in the cut is larger than 1080p, and the cut list sets the frame at 1920 by 1080. The YouTube master is 2560 by 1440 at about 28 megabits a second, with audio at about 336 kilobits, so it was scaled up from sources of 1080p or smaller. A higher bitrate keeps the re-compression from making the picture worse, but it cannot add detail the 13 clips generated at 720p never had.
Freeze ranges: what the detector flagged
Media Hub's analysis marks stretches where the picture does not change. Nine of the 29 clips have one. The saved cut list gives each shot's start and length in the source, so we compared those windows with the flagged ranges. Nine of the cut's 32 shot placements overlap a flagged range. The comparison assumes each shot plays at normal speed from its start time. It does not count the 0.2 to 0.3 second holds on a first frame that three scenes open with on purpose.
The football scoreboard shows why this is not a list of mistakes. Its clip is flagged as unchanging for the whole 10 seconds, yet in the finished film the score reads 21 in one frame and 27 in the next. A change that small does not break a flagged range.
So the flag is a prompt to look, not a verdict. Our AI Video Checklist lists frozen frames among the things to check before you publish, and these nine placements are the ones to look at: scenes 03gym1, 04gym2, 08bow, 09brian, 17volley and its punch-in, two shots of 09burst, and the scoreboard shot of 12montage. We looked at the scoreboard frames only. We have not looked at the other eight by eye, so this post reports the overlap and does not claim the edit avoided a frozen frame.
The sparkle, and the logo bug
Media Hub detected Google's sparkle mark on 3 of the 29 clips, each time in the lower right corner, and on 24 of the 81 clips in the AI folder. A detection is a measurement with a confidence score, and a clip with no detection might still carry the mark.
The cut list also puts the Next Day Videos logo in the lower right corner at 75 percent opacity, 180 pixels wide, and its note says the logo "covers Google's AI mark." On a 1920 by 1080 frame, the detected mark's box sits inside the logo's box.
We checked the finished film's lower right corner every two seconds. Where the footage carries the sparkle, it shows above the logo's wordmark, and the logo does not hide it. The cut list's note is out of date, and the film leaves the mark visible. Our post Let the Star Show explains why we label our AI video.
Which tool made each clip
| What Media Hub recorded for the clip | Clips (of 29) |
|---|---|
| Declared as Google Flow, downloaded at 1080p on 11/02 | 2 |
| Declared as Gemini, prompt unknown | 2 |
| No tool recorded | 25 |
The two Gemini clips are the football scoreboard and the student control room, the shot that replaces the repeated wrestling commentator. For the 25 with no tool recorded, the library never wrote down which tool made them. Two of those 25 carry a detected Gemini sparkle mark. The lesson is a small one you can copy: write down the tool and the prompt when you download a clip, because six weeks later nobody remembers, and a credit line has to be true.
Our other posts about this film name both Google Flow and Gemini. The film's end card reads "Video and voices generated with AI (Google Flow)", and so does the YouTube description. As of 11/17/2026, two clips are recorded as Gemini, so the credit in the video and the credit in these posts do not match.
What to take away when you budget
- Plan for about four clips per clip used. Of the 81 in the AI folder, 19 were used (23 percent, or about four to one). Counting all 29 used clips against the 91 gives about three to one, but 10 of the 29 were filed after the 91 were counted. That is one film, and your rate will differ, but it is a better starting number than hoping every clip is a keeper.
- Mark every clip when you watch it. The marks are the only record of who judged what, and the 12 unmarked clips are the ones to double-check.
- Measure before you cut. Resolution, freeze ranges and the sparkle mark are all in Media Hub's records and cost nothing to read. The ones that disagree with the picture are your editing notes.
- Record the tool and prompt at download. It is the only way to write an honest credit later.
For how we found what was in each clip without watching all of them, see the sorting post linked above. For when to use AI footage at all, see AI video or real footage. For the edit itself, see From 91 Clips to a 2:26 Premiere, or watch the premiere.
The figures in this post are from Media Hub's records of 11/17/2026. The library's marks change as clips are judged, and marking the 12 unmarked clips would change the marked count.
Researched and drafted with AI, then reviewed and edited by the site's editors.