// Guide.
The application runs seven steps in order, each needing what the one before it wrote. This is what happens at each, and where your judgement is the thing that decides.
ffmpeg travels inside the application. There is nothing else to install.
Open the disk image and drag the app onto Applications.
On first launch macOS will refuse to open it. Go to System Settings › Privacy & Security, scroll to the bottom, and click Open Anyway. Once, and never again.
Control-clicking and choosing Open, which older guides suggest, no longer works: macOS 15 removed that route for applications without an Apple Developer ID.
Open Resolve first, with a project loaded, then open the assistant.
One folder for the event, with the footage of each camera in its own subfolder and the board recording in an audio folder. The application can create the structure for you — Create Project Structure on the first step.
Folder names are read for hints. A folder with PUPITRU,
BOOTH, FOH, DRONE, FPV
or MIXER in its name is recognised as that role, which saves
you assigning it by hand later.
Which camera is which. Guessing from a folder name is a convenience, not a fact — confirm it in step 2.
Reads every folder, imports the media into Resolve, and organises it into bins. Nothing is renamed and nothing is moved on disk.
That every camera you shot is in the list. A card that did not finish copying is the commonest reason one is missing, and it is far cheaper to notice here than after the sync.
Assign each camera its role — booth, FOH, roaming, drone, FPV — which is what later steps use to decide who is on screen and for how long.
The roles, and any per-camera sync correction. If a camera lands wrong in step 3, you drag its clips into place in Resolve, note how far you moved them, and type that number here.
Aligns every clip against the board recording by acoustic analysis, using file timestamps as corroboration. It builds a synced multicam timeline from the result.
Nothing it cannot verify is presented as settled. A clip with too little overlap with the recording, a camera whose clock disagrees with the audio, two clips of one camera that overlap in time — all are marked for checking rather than quietly placed. The rule the whole step rests on is that no clip may be both wrong and unflagged.
Short clips are checked three ways before being trusted. Two of them listen to the loudness of the music; the third, added in 1.0 rc 8, listens across twenty-four frequency bands at once, which makes it far less bothered by the room and by how far the camera was from the speakers. It is used to settle cases where the first two disagreed, and to place clips neither could vouch for.
Sony cameras: keep the small .XML file that
sits beside each clip when you copy a card. It holds the recording time
even when the camera wrote none into the video itself, so clips from a
camera whose clock was never set can still be lined up. They are a few
kilobytes each and worth more than they look.
The clips it flagged, and the multicam timeline generally. Drones and FPV usually have no sound and no usable clock, so there is nothing to correlate — those it will tell you to place by hand.
If you forgot a file, copy it into the folder it belongs in and press Update file list back on step 1. It takes in anything the project has never seen and keeps every role, setting and position already computed — running the ingest again would discard all of it. If what you added belongs to the mixer, run this step again as well: the mixer is the event's clock, and the mix will stop and say so until you have.
If the recorder was stopped during the
event, the log says so — the recorder was stopped and started again — and that silence is kept as part of the
night, because it is. You will see the gap on the synced timeline and it
is meant to be there. Any clip filmed inside it comes out marked whatever
its match reads: there is no board feed at that moment to check it
against, so it is placed from the camera's own spacing, which is good to
about a second.
If you would rather align it yourself. Align the material however you like — by hand in Resolve, or in a dedicated tool such as Syncaila — then press Read Sync from Timeline and name the timeline holding your alignment. Every clip's position is taken from it and you carry on from step 4. The name you type is remembered, and steps 4 to 7 work on that timeline.
This is not optional if you aligned outside the application. Steps 4, 6 and 7 read positions from the project file, not from the timeline: a clip it has no position for is left out of the edit entirely, and the reels come out wrong while looking like everything worked.
Levels and prepares the board feed that the reels will take their sound from. It reports the condition of the recording first — clipping, or a level so low that bringing it up would lift its own noise floor into what you can hear.
What it says about the recording. Neither fault affects the sync, and both affect what you deliver.
With the standard 01_FOOTAGE/CAM_A_PUPITRU layout, each
camera gets its own role automatically. Without it there is nothing to
guess from, and every folder arrives as Mobile / Roaming — a
project can end up with three cameras in one role without anyone choosing
that.
Cameras that share a role share that role's cuts between them, which is intended but rarely what you meant if it happened by accident. When the ingest says it could not guess the roles, set them in Track Mapping before you sync.
Finds the drops in the board recording and places one marker at each. Each is a point on the timeline, at the instant the drop lands, with the score and the tempo in its note.
Three things have to be true at once, and the score is the weakest of them: the low end steps up and stays up for eight bars, it had been gone for the eight bars before, and it arrives loud — not merely louder than the silence it followed. The last two are what separate a drop from everything else a set does: the kick comes back after every fill and every transition too, and a kick creeping back into a breakdown is a huge jump without being a drop.
Audition every marker. Move the ones that are close but not right, delete the ones that are not drops, and add markers of your own where it missed something. Then press Read Markers: it takes the markers as they now stand, measures the tempo around any you added, and rebuilds every reel window on the bar grid.
On Apple Silicon the tempo and the beat grid come from a
beat tracker, so the reel windows land on the music's own bar lines. On
Intel they are estimated from the energy instead — the drops are found the
same way and in the same places, the windows are a few tens of
milliseconds looser. The first line of this step says which you have:
Rhythm analysis: librosa or numpy/scipy.
Moving the threshold answers immediately: the weaker candidates are kept too, so raising and lowering the bar re-marks the list on the spot rather than asking for another pass over the recording, and the count beside it says how many you are keeping. Only the ones above the bar get a marker and a preview clip.
A drop needs two cameras rolling to be kept — one camera is a single unbroken shot rather than an edit. Drops that happened while nobody was filming are left out, and counted in the log. The board records the whole night; the cameras do not film all of it.
Detection is good and it is not you. A false positive that reaches the export is a reel of nothing much; a drop it missed is a reel you never got.
Decides which camera is on screen at each moment of each reel — on the beat grid, respecting each role's minimum shot length, and only ever cutting to a camera that was actually recording at that instant.
The proposal, once it is on a timeline in step 7. The recipe is a starting point, and re-cutting a shot in Resolve is a normal thing to do.
Builds the export timelines — 16:9, 4:5 and 9:16 — with the framing applied and the audio laid under them.
Watch the reels. All of them. Vertical framing is a mathematical centre crop, so a subject standing off to one side will be off to one side in the crop — that is the moment to nudge the framing on that clip.
Nothing is rendered without you asking. The application builds timelines; you press render.
Premiere or Final Cut: Export writes two files
into 03_EXPORTS/ — a Premiere XML and a Final Cut FCPXML — each
with the synced multicam, a marker on every drop, and every reel as its own
sequence. Import it, and every clip is where the sync put it. Keep the
footage where it was when you exported.
Everything about a project lives in a .dropcut folder inside
the event folder: the project file, cached analysis, drop previews and any
processed audio. Move the event folder and it follows — the application
reads and writes where the project actually is, not where it used to be.
Delete .dropcut/cache whenever you like; it
rebuilds. Deleting the project file starts the event again from step 1.
No subject tracking or AI reframing — vertical framing is a plain centre crop, which stays predictable from cut to cut. No hardware timecode — alignment comes from the audio and from file metadata. No colour grading — clips are prepared for Resolve's own Remote Grades workflow, so grading stays where it belongs.
And it does not decide whether the reels are good. That is the job it is named for: it assists.
Questions: hello@thedropcut.com.
Thedropcut