PrimeCam 4K HLG Cinema Camera

Ryabinski Labs

About this app

PrimeCam is a 4K HLG10 cinema camera for Android, built for one job: producing footage a professional editor can cut without fighting it.

IT WRITES A FILE FINAL CUT PRO OPENS

Every clip is HEVC Main 10 in a .mov file, tagged with the full Rec. 2020 HLG colour triple and using the hvc1 sample entry Apple's stack actually accepts. Drop one into a Final Cut Pro library set to automatic colour and Final Cut recognises it as Rec. 2020 HLG on its own, offers to switch the library to wide-gamut HDR, and cuts the original file. No optimised media. No proxies. No transcode.

3840 x 2160, 24 fps, AAC 48 kHz stereo.

IT HAS NO INTERNET PERMISSION

Not a privacy policy — a missing capability. PrimeCam ships without the Android INTERNET permission, so the operating system itself will refuse the app a network connection. There are no accounts, no analytics, no crash reporting, and no third-party SDKs. Subject detection runs on models bundled inside the app, not on a server.

You do not have to take that on faith. Pull the package and read its manifest: CAMERA, RECORD_AUDIO, POST_NOTIFICATIONS, the two foreground-service permissions a recording needs, and BLUETOOTH_CONNECT so you can pick a Bluetooth microphone. That is the whole list.

PRIMES, NOT A ZOOM SLIDER

Three real focal lengths named in millimetres the way a lens is — 13mm ultrawide, 24mm wide, 104mm tele. The rail tells you which glass you are on and what the stabiliser is taking off the edges, so the crop is a number you chose rather than a surprise.

EYE-DETECTION AUTOFOCUS, AND IT SAYS WHETHER IT REALLY WORKED

Eye-detection autofocus — Eye AF — runs eye, then head, then body. It is the ladder professional stills and cinema bodies focus by, and historically you bought a mirrorless or cinema body to get it. PrimeCam runs that ladder on the phone, holding the nearer eye by its own coordinate, not the middle of a face box. It follows the subject’s eye, not the operator’s gaze — eye-detection autofocus, not eye tracking.

Then it does the part the expensive cameras still do not. A reticle reading TRACKING for everything hides the moment the eye is lost and the focus point slides onto a chest. PrimeCam names the rung it is actually on: EYE, FACE, BODY — or NO SUBJECT when it is honestly holding a place rather than a person. It climbs the same ladder for animals.

EVERY TAKE REMEMBERS ITSELF

Each clip is written with a JSON sidecar recording the lens, equivalent focal length, frame rate, stabilisation mode and its measured crop, bitrate, audio source and device — and every automatic action the camera took while you were rolling, with the second it happened. When a take looks wrong in the edit, the file tells you why.

MONITORING THAT EARNS ITS PIXELS

Waveform, false colour and focus peaking. A shutter-angle and ISO readout that stays legible in a greyscale photograph of the screen. And a stabilisation ladder that greys out the rungs the available light will not support, with the reason, instead of letting you shoot an unusable take and find out later.

THIS IS AN EARLY BUILD

PrimeCam is version 0.1.0 and has been tested on the Pixel 10 and nothing else, which makes every other phone untested rather than unsupported. A take is written as a spanned clip — numbered parts of a few seconds each, each finished as it is written — so a crash costs the last few seconds, not the take. The parts share the take’s name and import as one clip. One known issue is worth stating plainly: the rate. PrimeCam targets 24 fps rather than 23.976, but measured clips average marginally under nominal, and the JSON sidecar currently records the rate you asked for rather than the one the sensor achieved. Please do not shoot anything irreplaceable on this build.

Requires Android 13 or later, for the HLG10 dynamic-range profiles the camera depends on, and a camera that can record 4K in 10-bit HLG. Where a device cannot, PrimeCam says which part it cannot do rather than guessing.

What’s new

A take now survives the app being killed.

Recording is written as a spanned clip: numbered parts of a few seconds each, each finished as it is written. If the app stops unexpectedly you lose only the last few seconds, not the whole take. The parts of a take share its name and are numbered in order - they are one take, not several, and import that way.

Subject following names what it holds - EYE, FACE or BODY - and keeps it across movement, including animals.