Point it at the footage.
Add camera folders, or a Final Cut library, which is only ever read. When more footage turns up on Monday, Analyse New and Unfinished Clips reads only what is new, and says roughly how long it will take before it starts.
SIFTfcp, footage selects for Final Cut Pro
SIFTfcp watches every frame of every clip on your Mac and hands Final Cut the usable stretches as a selects project. It gets you about 85% there. The rest is a custom job, and it learns from your custom selects, so it gets better over time.Every usable stretch of every clip, into Final Cut. About 85% of the way there; the rest is yours, and it learns from it.
Add camera folders, or a Final Cut library, which is only ever read. When more footage turns up on Monday, Analyse New and Unfinished Clips reads only what is new, and says roughly how long it will take before it starts.
Each clip is decoded once, on your Mac. Frame by frame, SIFTfcp measures camera movement to a fraction of a pixel, zoom, focus and exposure, and notices when the camera was left filming the ground. A steady move passes. Handheld jitter does not.
Every stretch it leaves out says why: shaky, a whip pan, a quick zoom, soft, too dark, blown out, the exposure changing mid-shot, or the camera pointed at the ground. On this shoot that came to 26 minutes of 2 hours 5. The other 79% is usable footage.
The screen has moved to a day in a carving shed: hands, tools and shavings across 34 clips. The rules are about the picture, not the occasion. The Clips pane lays every select out as a picture to skim, and the vertical clips stand upright and go to Final Cut in their own shape.
Speeches, toasts and talks go to a Talking project, taken whole so nobody is cut off mid-sentence, shaky or not. Interviews are transcribed on the Mac, split at each question and titled by topic: Sam 03 How they met.
In Review/Tune every select is a yellow box on the filmstrip. Skim to the moment, set an in point, keep skimming, set an out point. It works as it does in Final Cut, even with the mouse held while you skim. A shot it left out that you want back is two keys away, and a clip set by hand stays as you set it.
Every rule is a slider: shake limit, fastest pan, quick zoom, focus, too dark, blown out. The measurements are kept, so moving one re-plans all 136 clips in about a second, without reading the footage again.
Second cameras, a Zoom recorder, lavaliers: SIFTfcp lines them up by the shape of their sound and Final Cut receives a multicam clip, with the best sound chosen for each speech.
SIFTfcp writes Final Cut XML: a selects project, a Talking project and an Interviews project, stabilisation switched on and keywords waiting in the browser. Media is referenced where it sits; nothing is copied. Footage that arrives later goes in as a project of its own.
The automatic pass gets you about 85% there. The rest is a custom job, and the clips you set by hand are what SIFTfcp learns from. Reach the last clip and it offers to train; every new shoot after that starts from what it learned, by itself, with nothing to apply.
How it fits your week
1 The day of the shoot
Add the camera folders and press Analyse. It runs in the background at low priority, and stopping and starting again loses nothing.
2 Overnight
As fast as the drive allows. From a USB hard drive, 4K reads at about the speed it plays, and a countdown says how long is left.
3 Morning
Skim what it chose. Mark in and out where you want more, or move a slider if it was strict. The pictures are kept, so this works with the drive unplugged.
4 Then
Selects, Talking and Interviews arrive as projects of their own. When the second shooter's footage turns up, only what is new goes in.
How it is meant to be used
1 The heart of it
Point SIFTfcp at a shoot and it keeps every usable stretch of every clip, with a reason written down for everything it left out. That gets you about 85% of the way there.
2 Then
Drag a select's ends to trim it, or its middle to slide it, and the picture follows frame by frame. Move a rule's slider and every clip is re-planned in about a second, without reading the footage again.
3 Or
Skim any clip and set its in and out points, even with the mouse still held. A shot the rules dropped becomes a select.
4 Over time
The clips you set by hand are what SIFTfcp learns from: what you kept counts as worth having, what you took out as not. Train it on them and every new shoot starts from what it learned, with nothing to apply. The more clips you set yourself, the closer it gets.
Pricing
Free for your first 100 clips: every feature, the export to Final Cut too. No time limit, no payment details.
Download for MacCancel any time. What you have already analysed and exported stays yours.
Questions
Yes. Analysis, browsing, review, hand picks, multicam and the export to Final Cut all run without a connection, and there is nothing to sign in to. Once you subscribe, SIFTfcp checks its licence now and then, when the paid-through date is within ten days. That check sends the key, the version of SIFTfcp and a number standing for your Mac, and nothing about your footage.
The one other exception: the first time it transcribes an interview, macOS may download Apple's speech model for your language, once.
No. Nothing leaves the Mac: no cloud, nothing to sign in to, no analytics. Your clients' footage stays on your drives. If you ever send us a problem report, it is a text file you choose to send, with folder and library names taken out and no footage, pictures or names in it.
What gets cut is measured, not guessed: camera movement to a fraction of a pixel, zoom and roll, focus from the sharpest edges, exposure from brightness and clipping. Those are measurements with limits you can see and move, not a model's opinion of your shots.
For what measurement cannot do, SIFTfcp uses Apple's own on-device frameworks, built into macOS: Sound Analysis to hear when someone is speaking, Apple's speech recogniser to transcribe interviews, and Vision to recognise what is in a frame, such as people, grass or sky. With Apple Intelligence switched on, it has one job: naming each part of an interview by its topic.
No cloud AI. No generative AI makes your edit, nothing is trained on your footage, and none of it leaves the Mac.
Never. Source files are only read. A Final Cut library is never opened: its database is copied aside and the copy is read. The names SIFTfcp gives clips live only in the XML.
No, and it is not trying to. It does the first pass, the selects: every usable stretch of every clip, with the talking and the interviews in projects of their own.
SIFTfcp gets you about 85% of the way there, and the rest is a custom job: the selects only you would make. It learns from those custom selects, so it gets closer to how you pick with every shoot. The film is still yours to cut.
Nothing is thrown away. The selects point at your original files, and Review/Tune shows every stretch that was left out and why. A shot the rules dropped can be marked back in by setting its in and out points.
The clips you set by hand are also what SIFTfcp learns from. Training tries the settings that are judgement calls across every shoot you have set by hand, and every new shoot after that starts from what it learned, with nothing to apply. The shoot you were reviewing keeps its selects as they are.
Final Cut Pro on the Mac, and only Final Cut Pro (FCPXML 1.11). Not Final Cut Pro for iPad, which does not import XML, and not Premiere or Resolve. Final Cut is not needed to analyse or review, only to open the export.
macOS 15 or later, on Apple silicon or Intel. Transcribing interviews needs macOS 26. Topic titles for interview parts use Apple Intelligence where the Mac has it, and otherwise the names and nouns said most.
Anything macOS can play: H.264, HEVC and ProRes from cameras, drones and phones, in camera folders or a Final Cut library.
Reading the footage is the slow part, and it goes as fast as the drive allows. From a USB hard drive, 4K reads at about the speed it plays. Proxies on the Mac's own drive go much quicker, and the export can still point at the camera originals.
Before it starts, the Footage pane gives a rough time, going by the length of the clips not yet measured and how fast the last analysis from that drive went, and a countdown runs beside the progress.
It runs in the background at low priority, and stopping and starting again loses nothing. Once the footage is measured, changing any setting re-plans every clip in about a second.
Browsing and choosing, yes: the pictures of every clip are kept with the analysis. Playing needs the drive.
$10 a month, or $85 a year, which saves $35 against paying monthly. Prices are in Canadian dollars. Cancel any time; what you have already analysed and exported stays yours.
Yes. The first 100 clips of your own footage are free, with every feature, the export to Final Cut included. No time limit and no payment details.
It is a clip limit rather than a day because reading a full day of footage from a USB drive takes about as long as the footage runs, and a day could go on one analysis before you had seen a single result.
Who made it
SIFTfcp is a project of the Folklore Creative Company. We have been making films for over 20 years, and for most of them we wished a tool like this existed. In the end we built it ourselves.
Our own team uses it on every shoot now, and it has done what we hoped: our edits start from its selects instead of from the raw footage.
Point SIFTfcp at the footage tonight. In the morning, Final Cut has the selects, the speeches and the interviews, each in a project of its own.