Nobody is
talking.It clips the game anyway.
A game broadcast runs three hours, carries no transcript worth scoring, and shows the play from a hundred feet up. Cutlist watches the frames instead of the words, tracks what is moving so the crop stays on the play, and reads only the stretch of the file that is actually the game.
Transcript-based clipping scores words. A game has almost none, and the ones it has belong to a commentator describing something that already finished.
Face detection has nothing to hold onto in a wide shot, so a centre crop hands you a vertical clip of empty grass while the play happens off-frame.
Warmups, the pre-show, halftime, the panel afterwards. Anything that reads the whole recording charges you for all of it.
A wide shotis the input.
Tell it the footage is Action and the pipeline leads with the pictures. A whistle blows, the crowd stands up and the score bug ticks over: three separate signals agree that something happened, and not one of them is a word.
- motion
- audio
- scene cut
- nominated
The two peaks are the plays. Motion spikes as it happens and the crowd arrives a beat later, which is what separates a score from a camera pan — and the commentator is often still mid-sentence about the last drive, so a transcript alone puts the clip in the wrong place.
Illustrative shape · your own curves are measured, not drawn
Action, not Talk
Talk leads with the transcript. Action leads with the pictures: sampled frames become the detector, and the crowd, the whistle and the impact are read as audio rather than thrown away. Auto reads the footage and tells you which way it went - on a game, set Action and there is nothing left to infer.
The scoreboard is context
On-screen text is read: the clock, the score bug, the graphic that names the scorer. None of that exists in a transcript, and all of it is used both to judge a moment and to title it.
Ask for the play you want
Plain words in the prompt box. "Every scoring play and every turnover." "Only the defensive stops." Selection narrows to what you described instead of to whatever was loudest.
It shows its working
Clips come back ranked by score, and each one reports what it saw, what the audio did, and why it landed where it did. Open the top four and check the reasoning rather than trust the number.
The score is a judgement about the footage, not about your season. It cannot know that the third-quarter stop mattered because of who you were playing. Sort by it, then overrule it.
The crop followsthe play.
A vertical frame keeps about a third of a wide shot. Centre it and you get the halfway line. Cutlist measures where the motion is across the frame and pans with it, so the play stays in the cut instead of leaving it.
Follow
Motion tracking, horizontal. The crop pans with whatever is moving, smoothed and speed-limited so a replay wipe cannot whip it across the frame. This is the setting that matters most on a game.
Fit
The whole picture on a blurred bed. Nothing is ever cropped out, which is the safe choice when play is spread across the width of the field.
Manual
Place the crop yourself and leave it there. For a fixed camera pointed at one end, or a sideline angle where the action stays in a known third of the frame.
Vertical, square or wide
9:16 for the feed, 1:1 for a board, 16:9 for the film session. Each rendered at full resolution rather than upscaled from a smaller cut.
Follow tracks the centre of the motion, not a chosen subject. When two things happen at opposite ends of the field the crop sits between them and holds neither cleanly. That is the footage to run on Fit.
Pay for the game,not the pre-show.
One credit is one minute of source, rounded up, at x1.8 a minute with the see and listen review or x0.7 when you supply the timestamps - plus a start-up charge for the work that happens once, which is 152 credits on a found job and 1 on an anchored one. On a game the bigger lever is the range. Set where the game starts and ends and the rest of the file is never read, never scanned and never billed.
| What you set | What gets read | Billed | Credits |
|---|---|---|---|
| Nothing. Drop the file in | All three hours: warmups, the pre-show, halftime, the panel afterwards | 180 min x 1.8 | 477 |
| Range 0:20:00 - 2:40:00 | Kickoff to the final whistle. Only the window is read, and only the window is billed | 140 min x 1.8 | 391 |
| Timestamps from the play log | Detection skipped entirely. Captions off, because the audio is a crowd | 180 min x 0.7 | 128 |
| Timestamps, captions on | Still no detection, but a transcript has to be made and that part is billed | 180 min x 0.98 | 178 |
A three-hour broadcast · game window 0:20:00 to 2:40:00 · figures computed by the code that bills you
One control, 86 credits. A weekly game wants Studio at $249 a month: 6 full games clipped that way, or 18 when a play log supplies the timestamps and the captions stay off. The estimate updates as you type the range, so the number is on screen before you spend anything.
A moment that straddles the edge of the window is trimmed to fit inside it, and dropped if the trim leaves it under the minimum clip length. Give the window a minute of slack at each end. Supplied timestamps take precedence over a range - the engine processes the whole source and the bill follows it, at the anchored rate.
Your play log beatsanything a model infers.
If a scorekeeper logged the game, you already have better ground truth than detection will ever produce. Paste the times, or have your stats platform send them. Detection is skipped entirely, the moments are never re-ranked or deduplicated, and the start-up charge falls from 152 credits to 1 - the delivery and the renders, which happen either way.
12:34 - Touchdown, ranges like 1:02:11-1:02:40, bracketed times, bare seconds. One per line. The box counts what it recognised as you type, so a mistyped line is visible before you spend.
Supplied moments are never re-scored, merged or trimmed to a clip limit. Give a start and an end where the play has a clear one and it is cut as written.
The same route exists as an API. Your product sends the timestamps and gets clips back on a signed webhook, carrying whatever metadata it sent.
Statside does this
Statside is a multi-sport live-stats platform. A scorekeeper logs every play as it happens and the broadcast worker already scores each one. When a play is worth posting, Statside sends that timestamp and its weight to Cutlist and gets back a vertical, captioned clip - without a model ever being asked what happened, because the platform already knew. If you run your league on a stats product, this is the integration to ask them for.
How the partner API worksWhoever has topost it.
Coaches
A sideline recording in, a handful of teaching clips out. Vertical for the group chat, wide for the room.
League operators
Every game in the competition through one account, cut to a shape that fits a feed without an editor in the loop.
Athletic departments
Recruiting and social from footage you already pay to record. No second crew, no edit bay time.
Team media staff
The person who has to post before the bus leaves. Set the range, run it on the drive home.
What itwill not do.
Live clipping
It works on a recording, not on a feed that is still running. Upload or link the file once the game is over.
A roster
You can ask it for a number on a shirt and it reads what is legible on screen. It keeps no player index, and it will not build a season cutup per athlete.
Stat graphics
No generated score bug, no lower thirds, no results card. Your logo and your colour, plus whatever the broadcast already burned in.
Original versus replay
The score cannot tell a play from the replay that follows it, and will sometimes rate both. When two clips of the same moment come back, delete one.
Posting
Download and post yourself. We do not publish on your behalf and we do not touch your accounts.
Generated b-roll
Deliberate, not missing. Your clip is your footage - no stock inserts, no synthetic filler.
An archive
Clips live 30 days, source uploads are removed within 2 days of a finished render, and the account holds 100 GB across every project. Download what you want to keep.
A rights answer
Whether you may clip a broadcast you did not shoot is a rights question. The tool cannot answer it for you.