Plug in the GoPro. Check three points. GO.
The station identifies the guest by QR code, finds exit, canopy opening and landing, and edits them into a main video, best-of, reel and photo selection. Processing runs on a Mac mini at your dropzone. The manifest then sends the link.
- 01Voucher and booking
- 02Check-in on the iPad
- 03Load and jump
- 04Editing at the stationThis page
- 05Media page and purchase
- 06Settlement
Nine steps. Three of them are done by a person.
Before the jump the camera films a QR code, after landing it is plugged in, then a quick check. The station, NAS and manifest handle the rest, right up to the message to the guest.
GM-2417Point the GoPro at the QR code and film briefly
- 01
Film the QR code
Before boarding, the tandem instructor taps his guest on the staff terminal or at the station and films the QR code briefly with the GoPro. Forgot? Then the guest is picked from the day list at the station.
Tandem instructor - 02
Plug in, import with checksum
After landing, the GoPro goes on the USB cable or the SD card into the card reader. The station loads all clips, verifies every file and finds the QR code along the way. Guest, load, tandem instructor and booked package come from the manifest.
Tandem instructor · Station - 03
Empty the camera
Only once the copy is confirmed is the GoPro emptied. Until then it says in red: do not unplug. "Keep clips" cancels the emptying.
Station
- 04
AI markers
The analysis divides the clips into six scenes and sets exit, canopy opening and landing, each marker with a confidence score. For a typical jump this takes around 20 seconds.
Station - 05
Check and GO
Three short loops, three traffic lights. If all are green, the edit starts by itself after a 30 second countdown; any intervention stops it. Otherwise the tandem instructor presses GO and confirms with "Markers correct? YES", or first drags a marker into place.
Tandem instructor - 06
Versions are created
Main video, best-of, reel and photo selection. Whatever the guest has not booked is additionally created as a watermarked preview and is finished first.
Station
- 07
Backup to the NAS
The project folder is copied to the NAS and verified by checksum. Only then does it count as backed up.
Station · NAS - 08
Upload to cloud storage
After an adjustable delay, a service on the NAS uploads the media: the full version if it has been purchased, otherwise the watermarked preview. If the internet goes down, it waits and continues afterwards.
NAS - 09
Link to the guest
As soon as everything is available, the manifest sends the guest an email with their personal link, by SMS as well if desired.
Manifest
Simplified view with sample data
When the jump begins is in the camera's data.
Besides the picture, the GoPro also records motion data: acceleration around 200 times per second, plus rotation and gravity. Together with the audio and its own image AI, the station sets three markers: exit, canopy opening and landing.
Motion data
At the exit, acceleration briefly drops below 1 g, the opening shows a clear shock, the touchdown a sharp one. Our own AI models derive the fine values from this.
Audio
In freefall the wind roars, under the canopy it goes quiet. The audio helps divide the scenes and detects the "Landed" scene even when the touchdown was not filmed.
Image AI
A scene AI looks at five still frames per clip. Together with the sensor and audio data, it assigns each clip to one of the six scenes.
Every marker gets a confidence score. A plausibility check requires exactly one freefall, the opening after the exit and the landing after that. If something does not fit, the score drops. This becomes the traffic light in the check:
- Green confidence score 0.75 or higher without warning
- Yellow uncertain, please take a look
- Red not found or very uncertain
This is how accurate the AI markers were when re-measured.
We compared the AI's markers with manually set ones, on test sets from our own jumps. For the exit it was 169 jumps that were never seen during training.
For comparison
Our earlier audio-based detection was off by a median of 5.2 seconds at the exit. These are exactly the points you search for when editing by hand.
Why a person still checks
A median is no promise for every jump. That is why the station only starts by itself when all three traffic lights are green, and even then only after a countdown that any intervention stops. In daily use that is more the exception: usually one light is yellow. Then the tandem instructor takes a look, confirms with YES or first drags the marker into place.
Evaluation from September 2026. It is based on around 1,900 manually set markers from our own jumps (968 exits, about 540 openings, about 400 landings). Part of these were never used for training, only for measuring. The AI markers have been in daily use since September 2026.
The check fits on one screen.
After the analysis the station shows what it has found: one loop each around freefall start, freefall end and landing, in slow motion at the marker. The tandem instructor takes a look, confirms or drags a marker into place. Guest and tandem instructor stay at the top of the screen the whole time.
Simplified view with sample data
Three loops
One each around freefall start, freefall end and landing. They run continuously, slowed down at the marker. A click pauses one window and jumps to the marker; the others keep running.
Traffic light per marker
Green means "Looks good": confidence score of 0.75 or higher and no warning. Otherwise it says "Check", yellow or red.
Move a marker
Drag the marker on the timeline or move it frame by frame with the arrow keys. At the edge, the excerpt grows with it.
Start
If a traffic light is not green, the station asks again after clicking GO: "Markers correct? YES". If all are green, a countdown runs. Any intervention stops it; the edit then starts only via GO.
Starting in 30 s▶ Now
Scenes
The bar shows the division into six scenes, clip by clip. A scene can be changed or hidden, undo included. The classic timeline view remains available at any time.
The film is already being made during the check
While the tandem instructor is looking, the station pre-renders in the background using the markers it found. If he confirms without changes, it uses the finished parts. If he changes something, the pre-render starts over.
Built in since September, not yet tested in jump operations.
One jump, four versions.
From the handcam footage, the station creates the film, best-of, portrait reel and photo selection, processed on the Mac at your dropzone. Whether the guest has booked video, photos or both, the station knows from the manifest.
Main video with musical dramaturgy
Intro, the scenes in their order, outro. Song 1 carries the beginning. Song 2 starts before the exit; its climax lands exactly on the exit. Around the canopy opening only the original audio plays, then song 3 carries through to the outro. Music and original audio are leveled per scene. With a handcam, the video is created in Full HD, as H.264 at 10 Mbit/s.
You choose the songs, intro and outro.
Portrait reel
A short portrait reel at 1080 × 1920, target length 30 seconds (adjustable). It begins with the exit, slightly slowed forwards and then fast backwards, followed by the best moments. The framing is aligned to the guest's face per scene. With its own music, ready for Instagram and TikTok.
Best-of
Around 60 seconds with the best moments per scene, selected by faces and smiles. Exit and canopy opening are always included, plus song 2 with beat drop.
Photo selection
The station samples the handcam's 4K raw video at 4 frames per second, discards blurry ones and rates the rest. The result is 200 photos plus an exit series of 20 frames. The 20 best go into the gallery.
Whatever is not booked is available as a preview.
If the guest has not booked video or photos, that part is additionally created with a watermark: the logo tiled diagonally across the whole image. This way the guest sees their real jump. If they buy later, the system unlocks the full version.
This is how the Media salesSimplified view with sample data
The guest sees their jump while they are still excited.
After the edit, playback starts at the station by itself. Whatever the guest has not booked is finished first, with a watermark. That is the moment for adding on at the desk.
- Three milestones. New · in trial The station shows when the video is playable, when the photos are finished and when the next GoPro is up. It estimates the remaining time from its own recent runs.
- One playback window New · in trial for video, best-of, reel and photos, with the status of each version. If nothing is booked, it moves on from the video to the photos by itself.
- At the desk in parallel: On the media station's iPad, the guest sees their media straight from the editing station. A button registers the wish to add on; payment is taken at the desk.
Simplified view with sample data
Milestones and playback window have been built in since September and are not yet tested in jump operations.
The camera is only emptied once the copy is verifiably correct.
Every copy is verified: from the camera to the Mac and from the Mac to the NAS. Only what is verifiably backed up is deleted.
Simplified view with sample data
- When loading from the GoPro every file is checked for complete size and reloaded up to three times on errors. If a clip is still missing after that, the import aborts and everything remains on the camera.
- With SD cards an MD5 checksum is created while copying, and the copy is then verified against it. By default the card is afterwards only renamed, not erased.
- To the NAS goes the entire project folder, again with checksum. Only then does it count as backed up. New · in trial The backup runs in the background; the station is free again right away. An error stays shown in red, and one click repeats the backup.
- Disk housekeeping keeps the Mac below an adjustable fill level. It only deletes projects that are verifiably backed up, the oldest first, and never raw videos or photos from the same day.
Processed locally
Analysis and editing run on the Mac at your dropzone. The raw videos do not need to go to the internet first.
Corrections in the log
With every edit, the station records which markers were corrected and how it was started. This allows the automation to be measured against real operating data.
Outside camera is left untouched
The station never erases the card or GoPro of an outside camera flyer. Their footage is only read and backed up.
One Mac mini per station, your GoPros and a NAS.
We tell you in advance exactly what works and what does not. The station has been tested on the Mac mini M4 with GoPro cameras.
Mac mini
One Mac mini per station, tested with M4. The station uses macOS components for video encoding, camera connection and image analysis. Where it was developed, three stations run in daily use.
GoPro
Your tandem instructors' cameras. They connect to the station via USB cable or SD card, including the HERO 11 to 13, which do not show up as a drive. The AI markers need the GoPro's motion data. For photos from the video, the handcam films in 4K.
NAS and internet
A NAS on the local network backs up the projects and uploads the media for the guests to cloud storage. This requires internet at the dropzone.
Manifest system
The counterpart: this is where the day list, load, tandem instructor and booked package come from, and where the station reports its status. The manifest sends the link to the guest. To the manifest system
From you: music, intro, outro, logo
You supply the songs for the main video and reel, your intro and outro, and the logo for the watermark. We do not provide music.
Set up for your dropzone
The station is not a device you unpack and start using. It is configured for your dropzone: station names, NAS and the connection to your manifest. It does not run on Windows.
What operators want to know about the station
What happens if the AI gets it wrong?
If the AI is uncertain, the traffic light shows yellow or red and the station does not start by itself. But green is not a guarantee. That is why even with three green lights a 30 second countdown runs first. Any intervention stops it; the edit then starts only via GO. The tandem instructor drags the marker into place or moves it frame by frame with the arrow keys. Anyone who prefers working in the familiar timeline switches to the classic view.
Does it work with any camera?
Importing, yes: the station also reads SD cards from other cameras, and photos including RAW and HEIC are backed up too. The AI markers, however, need the motion data of a GoPro. Without it, the operator sets the markers by hand.
How long does an edit take?
That depends on the length of the clips and on the station, so we do not quote a fixed time. Newly built in and still in trial is a display at the station: it estimates when the video is playable, when the photos are finished and when the next GoPro is up. The estimate is based on the station's recent runs.
What about an outside camera or a separate photo camera?
Both are built in. With "OUT video to follow", the import waits for the outside camera flyer's camera; its scenes go into the main video, and additional photos are created from the outside perspective. A separate photo card is assigned by the station to the same guest, and the best images are selected from it. Where the software was developed, the case with one handcam is what mostly runs day to day.
How many stations does a dropzone need?
That depends on how many tandems jump per load. A station edits one guest after another, and while an edit is running it does not accept a new camera. Each station continuously reports its status to the manifest; in the admin you can see which station is online and what it is currently working on. Where the software was developed, three Mac mini stations run in daily use.
What music plays in the videos?
Yours. You store the songs for the main video and reel as well as the intro and outro. We do not provide music; you clear the rights to the songs yourself.
See what the station reports to the manifest.
In your own demo with the fictional dropzone "Skydive Nordlicht", you land a load and watch the editing station report import, edit and done. The station is simulated in the demo. Then you open a guest's media page, with watermark and post-sale purchase. Questions about the station itself we answer in a conversation.
