Live Photos on iPhone & iPad

How a Live Photo is recorded

It starts right in the Camera app. Make sure Live Photo is turned on and you are in Photo mode, then take the shot as you normally would. The options that matter are all on one screen:

Figure: The iPhone Camera controls used to capture a Live Photo - turn Live Photo on, stay in Photo mode, and tap the shutter.

The effect feels like magic, but the mechanism is simple. While Live mode is on, the camera is already running and continuously buffering recent frames and audio in memory - even before you tap the shutter. The moment you press it, two things happen:

• The frame at the shutter is kept as the full-resolution key photo - the image you see in your gallery.
• The buffered frames and audio from just before and just after are written out as the paired motion video.

Figure: Live mode buffers continuously; the shutter picks the key photo and the surrounding window becomes the motion video.

Because the window extends before the shutter, a Live Photo often captures the instant that actually mattered - a smile, a wave, a splash - even if your timing was slightly off. That same property is what makes the format so useful downstream.

The output files

A single Live Photo exports as two files that always travel together:

• The still image - usually HEIC (Apple's efficient format), or JPEG in broader-compatibility setups. This is the key photo, at full resolution.
• The motion video - a QuickTime .mov encoded with HEVC or H.264, carrying the short clip and its audio.
• A shared content identifier - the key that pairs the still and the movie. Lose it, and you have an unrelated photo and clip; keep it, and you have a Live Photo.

Figure: Two paired files - a still and a motion video - matched by one content identifier.

Both files carry metadata worth preserving for any dataset: resolution, orientation, duration, codec and frame rate. Orientation in particular matters - the motion video is stored in the sensor's native landscape with a rotation flag, so a pipeline that ignores that flag will show portrait clips sideways. Collecting Live Photos well means keeping the pair intact and keeping this metadata correct.

Why Live Photos make great AI training data

A still photo gives a model one frame. A Live Photo gives it a frame plus the motion around it - and that extra temporal context is exactly what many modern models are hungry for. From one capture you can derive several supervision signals:

Figure: One Live Photo, many supervision signals.

• Liveness & anti-spoofing: genuine motion and parallax help distinguish a live subject from a photo-of-a-photo or a screen replay - valuable for face and document verification.
• Action & gesture recognition: the short clip shows how a motion unfolds, not just its end state.
• Depth & parallax cues: the small natural camera movement across frames hints at 3D structure that a single image cannot convey.
• Deblurring & super-resolution: multiple nearby frames can be fused to sharpen or upscale a single still.
• Best-frame selection: pick the sharpest, eyes-open, least-occluded frame from the window instead of being stuck with one shutter instant.
• Micro-expressions & intent: subtle changes in a face or hand that a photo freezes out are preserved in the motion.

In short, Live Photos let a dataset carry richer, more authentic signal per capture - without asking contributors to shoot full videos.

Collecting Live Photos at scale with HaiCrowd

Knowing why Live Photos are valuable is one thing; collecting them at scale, from real devices, across the world is another. That is where HaiCrowd comes in.

HaiCrowd supports native Live Photo capture inside the data-collection app, so a global crowd of contributors can record genuine Live Photos on their own iPhone and iPad devices. The platform is built to keep the format intact end to end:

• Captured natively - both the still and the motion video are recorded, upright, exactly as Apple produces them.
• Kept paired - the still and .mov upload together and stay matched by their shared identifier, with metadata preserved.
• Quality controlled - automated and human QC verify the pairing, quality and any device requirements before data is accepted.
• Exported cleanly - delivered to the customer as paired, versioned files with their metadata.

Figure: A live_photo collection task in the HaiCrowd app - contributors tap Live Photo to capture, and progress is tracked through upload, QC and rework.

The core iOS APIs behind Live Photo capture

For the technically curious, Live Photo capture is built entirely on Apple's public frameworks - AVFoundation for capture and PhotoKit for display and export. At a high level, four pieces do the work:

Figure: The public AVFoundation / PhotoKit pipeline for a Live Photo.

• AVCaptureSession - drives the camera and microphone with the .photo preset and provides the live preview.
• AVCapturePhotoOutput - the output that produces the result; the app checks isLivePhotoCaptureSupported and turns on isLivePhotoCaptureEnabled.
• AVCapturePhotoSettings - configured per shot; setting livePhotoMovieFileURL tells the system where to write the paired movie.
• AVCapturePhotoCaptureDelegate - receives the results through two callbacks: one delivers the processed still photo, the other reports the finished Live Photo movie once it is written.
• PHLivePhoto (PhotoKit) - recombines the still and movie so the pair can be displayed and played back as a single Live Photo, or exported.

That is the whole shape of it: a session, a photo output with Live Photo enabled, per-shot settings that name the movie file, and a delegate that hands back the two paired files. Everything else - orientation handling, pairing, QC, delivery - is built on top.

Need to collect Live Photos - or any rich mobile capture - at scale? Write to info@haidata.ai to see how HaiCrowd can help.