Reel samples your video, looks at each frame on hardware you own, measures the loudness and the scene cuts, and hands back candidate windows by timecode — with the sampling interval, the blind spot and the margin printed beside them.
Nothing opens in this browser — this one takes a job and does the work on our own graphics cards in Oregon. The free allowance below is how you try it: 60 frames, no card and no payment, on the key you are issued.
When the top two windows are closer together than one sampled frame is worth, it says there is no first place and lists them level.
The problem
None of them has. They all sample — a frame every few seconds, because looking at every frame of a twenty-minute master would cost more than the clip is worth. The difference is that they do not tell you the interval, so you cannot tell whether the moment they missed was two seconds long or twenty. Then they print a best clip with a score to three decimal places, worked out from six frames, and the decimals do the rest.
Printed on every sheet. Sampling every five seconds means four seconds between every pair of frames that Reel did not see, and a moment shorter than that can sit inside a window it scored low.
What one frame out of six changing its verdict is worth. Any gap smaller than that is reported level rather than ranked, because which window came first was decided by a single frame.
There was no speech-to-text model on the hardware, so Reel does not claim to know what was said. It measures loudness with ffmpeg, calls it loudness, gives it a fifth of the score, and drops it entirely when the audio is too flat to separate anything.
A thirty-second window sampled every five seconds is scored on six frames. One of those six coming back unusable instead of usable moves the score by a fixed, published amount. When the top windows are inside it, Reel says so and lists them level — because you can look at two thirty-second clips and pick, and you cannot un-pick the one a false ranking talked you out of.
How it works
Nothing is uploaded. Your masters are already on your disk and a five-minute one is hundreds of megabytes; Reel reads it where it sits and never renames, moves or re-encodes it.
One frame every few seconds goes to the graphics cards. Separately, and costing nothing, ffmpeg measures how loud each second is and finds every scene cut in the file.
Overlapping windows are suppressed, so a shortlist of five is five different moments. Each one carries its coverage, its movement, its loudness and whether it starts on a measured cut.
Press the button and the window is written out as a new mp4 — re-encoded rather than stream-copied, so it starts exactly where the sheet says it starts.
What it will never do
Three of these are refusals to claim more than was looked at. The fourth is a refusal to show you the same moment five times and call it a shortlist.
Pricing
A frame is one picture looked at, so what a video costs depends on how long it is and how often you sample. A ten-minute master at the default five-second interval is 120 frames. Sixty free is a five-minute video sampled every five seconds, or a ten-minute one sampled every ten.
One short video, judged properly.
About twenty-five ten-minute masters.
A studio choosing promos off every shoot.
Per frame looked at. A frame that failed on the GPU box, or that came back without saying whether it is usable footage, is not charged — and it still counts against the windows it falls in rather than being dropped and quietly raising them. Measuring the loudness, finding the scene cuts, the window arithmetic and cutting the clip are free and always will be. Your licence key is emailed to you within two working days of payment.
Checkout is our own bitcoin server, not a card processor. The units you buy are added to your key: buy from the Buy more box inside the program and they arrive on their own once the payment confirms, or buy here and paste the invoice number from your receipt into that box.
Questions
A ten-minute video at thirty frames a second is eighteen thousand frames. At any price per frame that is not a product, and the answer from eighteen thousand frames is not meaningfully better than the answer from a hundred and twenty.
The honest thing is not to look at every frame — it is to say how often you looked, which is what the blind spot line on every sheet does.
Because the alternative is a ranking that one frame decided, and acting on it means cutting the second-best thirty seconds of your own video and never finding out.
The margin is printed, so you can see exactly how close it was and pick between two clips yourself.
That something got louder. Not what it was, and not what was said — there is no speech-to-text in the path and Reel does not pretend there is.
It is a fifth of the score, it is measured by ffmpeg rather than judged by anything, and on a video with no audio or with flat audio it is dropped from the score entirely and the sheet says so.
No. It is read where it sits on your own disk and is never renamed, moved or re-encoded. The sampled frames go to the graphics cards on your own network and nowhere else — never to a hosted vision API.
The cut is written as a new file beside the job's other outputs. Your master is not touched.
Yes, and that is why cutting takes a moment instead of being instant. A stream copy has to begin on a keyframe and would quietly move the in-point by up to a couple of seconds.
One clip that does not start where the sheet says it starts would make every timecode in the product untrustworthy, so the cost is paid in CPU instead.
A key. One line of text, issued to you and shown exactly once — only its hash is kept here, so it cannot be looked up later and we could not tell you what yours is. Every key arrives with a free allowance already on it, which is the tier at the top of the pricing above, and the work starts the first time you send a job with that key on it.
Each job is metered in the unit named in the pricing, and a unit that fails — a file that will not read, a row the model held back — is not charged and is named in the job report rather than silently skipped. When an allowance runs out the next job is refused before any work starts, with the number it used printed on the refusal.
More units are bought in bitcoin. The checkout is a bitcoin server we run ourselves. Buy from the Buy more box inside the program and the units are added to your key on their own as soon as the payment confirms; buy from the plan buttons above and paste the invoice number from your receipt into that same box. Each payment adds its units once, to one key.
Sixty frames are free. The useful outcome is not that Reel agrees with you — it is finding out how much of your video it could not see, and whether the window you picked was ever ahead by more than one frame.