We capture Apple Log video using AVCaptureVideoDataOutput with: AVCaptureDevice.activeColorSpace = .appleLog or .appleLog2 kCVPixelFormatType_422YpCbCr10BiPlanarVideoRange (x422) AVAssetWriter for the final ProRes/HEVC recording We need to bake spatial operations such as a centre crop and optional anamorphic desqueeze into the recorded raster. We want the result to remain Apple Log for subsequent grading; we do not want to display-transform or convert it to HLG. Core Image/Core Graphics does not appear to expose a public Apple Log or Apple Log 2 CGColorSpace. Core Video instead identifies the signal using kCVImageBufferLogTransferFunctionKey. Would this be the supported Core Image approach for spatial-only processing? let image = CIImage( cvPixelBuffer: sourceBuffer, options: [.colorSpace: NSNull()] ) let context = CIContext( mtlDevice: metalDevice, options: [ .workingColorSpace: NSNull(), .outputColorSpace: NSNull(), .cacheIntermediates: false ] ) let outputImage = image .cropped(to: cropRect) .transformed(by: spatialTransform) context.render( outputImage, to: destinationX422Buffer, bounds: outputBounds, colorSpace: nil ) We would then copy the source buffer’s Apple Log colour and Log-transfer attachments to the destination buffer and append it to an AVAssetWriterInputPixelBufferAdaptor. Could an Apple engineer clarify the following? Does CIImage(cvPixelBuffer:) natively interpret kCVImageBufferLogTransferFunctionKey, even though no public Apple Log CGColorSpace exists? When NSNull()/nil is supplied as above, does Core Image leave the Apple Log values unmanaged during crop and affine operations? When rendering x422 to x422, can Core Image preserve the Log signal apart from expected resampling/rounding, or does it internally perform a colour conversion such as YCbCr → RGB → YCbCr? For scaling or anamorphic desqueeze, is Apple’s recommended workflow to: resample the encoded Apple Log values with colour management disabled, or explicitly decode Apple Log to a linear working space, resample, and encode back using a custom Metal implementation? Is copying kCVImageBufferLogTransferFunctionKey and the related colour attachments from source to destination sufficient for AVAssetWriter, assuming the writer settings were obtained from recommendedVideoSettingsForAssetWriter(writingTo:)? Is there ever a supported reason to use HLG as an intermediate for Apple Log processing, or should that be avoided? This this part, we are concerned about preserving the captured Log signal in a recording pipeline, not applying a viewing LUT or displaying Apple Log. Next, we also support optional LUT processing with CIColorCube. Some LUTs are technical Apple Log-to-display transforms, while others expect linear or Rec.709 input. If the image is created with .colorSpace: NSNull(), we understand that the LUT receives unmanaged Apple Log code values and that the LUT itself must perform the required transfer-function and gamut conversion. Does Core Image synthesize an Apple Log-aware input color space from kCVImageBufferLogTransferFunctionKey, or must applications implement Apple Log/Apple Log 2 decoding explicitly before using ordinary Core Image filters? Is there a supported CGColorSpace, ColorSync profile, or Core Image conversion API for this? Additionally, what numeric range does CIColorCube receive when its source is an x422 video-range Apple Log buffer— normalized Log RGB values, or values requiring explicit video-range conversion?

Summary. iPhone 17 Pro Max + Blackmagic Camera ProDock, genlock BNC in from a generator confirmed at a true 30.00 fps. follow(_:videoFrameDuration:delegate:) reaches .activeSync in ~2 s. On one unit the lock then holds for 10+ minutes. On a second unit, same binary and same reference, it detaches 6 s to ~4 min after lock: .activeSync → .ready with input.externalSyncDevice == nil, no runtime error and no delegate error. Calling follow() again on the still-running session is rejected with -11800 every time, while unfollowExternalSyncDevice() plus a stopRunning()/startRunning() bounce recovers reliably. I am not looking for a fix. I want to know whether my call sequence is wrong, whether this transition is expected, and how a shipping app should be structured around it. Questions Is re-following a running session supported? Is the unfollow + bounce the intended reset sequence, or is there a lighter-weight way to clear whatever state the -11800 is keyed on? And once detached, is re-acquisition entirely the app's responsibility, or is the system expected to re-calibrate on its own while the reference is present? Is a hard detach a legal outcome for an already-calibrated input? The documentation describes .freeRunSync as the hold-over when a locked input loses sync. Is that hold-over guaranteed, or must an app also handle .activeSync → .ready with a nil externalSyncDevice? Is anything in my call sequence wrong (code in a reply below), and is polling input.externalSyncDevice the right signal to key recovery on, or is there a supported notification for detach? Setup. Video-only AVCaptureSession: one .builtInWideAngleCamera input, one AVCaptureVideoDataOutput. No multi-cam, no depth output, no synchronizer, no audio. Both frame durations set to CMTime(1, 30) before the input is created, never rewritten while a follow is live. Device A (iOS 26.0.1) holds: 601 s and 956 s runs, zero detaches. Device B (26.3.1, then 26.5.2) has 39 drops across 6 logs. The ProDock, cable and generator were swapped between units; the failure followed the phone. Caveat: n=2, and unit and OS build vary together. The detach. No AVCaptureSessionRuntimeError and no delegate error (-11892 has never been observed here, so this is not the documented frame-duration-mismatch path). The session keeps running and delivering frames. Status is .ready, not .unavailable — the ProDock stays enumerated, the reference unchanged. No confirmed drop has passed through .freeRunSync. PTS ground truth, independent of the follow state: while locked, every frame PTS sits exactly on the 1/30 grid, zero drift. At the drop there is exactly one teardown gap, 238–337 ms across 8 runs, after which the clock free-runs at ~30.013 fps (444 ppm) and never returns to the grid. On -11800. It surfaces through AVCaptureSessionRuntimeErrorNotification. The bounce recovers 3/3, with no activeFormat change. Caveat: -11800 is AVErrorUnknown and I see it in unrelated cases too, so I do not assume it is specific to retained follow state. Ruled out. Exposure duration — a run at ≤ 16.67 ms, within the recommendation in the follow() documentation, still drops. Reference drift — zero, by microsecond PTS. Accessory chain — full swap; the failure followed the phone. Another client — Final Cut Camera holds genlock on the fragile unit with the same ProDock and reference. Load and resolution — load changes time-to-drop, not whether it drops; with recording and audio off it still drops, and Device B drops at both 12 MP and 1080p. Code and supporting logs are in replies below; the length limit would not take them inline. Per-frame PTS CSVs, raw status logs, and a minimal Xcode project that still drops are available on request. Prior art read: forums/thread/799739 and thread/804594 — neither covers post-lock detach or re-acquisition.

If an app allows people to create a playlist and add more songs to that created playlist, it would make sense to guard them from accidentally adding the same song to the playlist more than once. In this code, even though it is successfully receiving the existing playlist from the request, its tracks and entries always show as nil even when there are songs in the playlist. Any suggestions for how to guard against adding duplicates to a playlist? Thank you! var request = MusicLibraryRequest<Playlist>() request.filter(matching: \.name, equalTo: "AppGeneratedPlaylist") let response = try await request.response() if let existingPlaylist = response.items.first { if let tracks = existingPlaylist.entries, tracks.contains(where: { $0.id == song.id }) { print("Song is already in the playlist, so don't add again") return } else { try await MusicLibrary.shared.add(song, to: existingPlaylist) print("Added song to existing playlist: \(existingPlaylist.name)") print("Count of tracks: \(existingPlaylist.tracks?.count)") print("Count of entries: \(existingPlaylist.entries?.count)") print("Current tracks: \(existingPlaylist.tracks?.map(\.id))") print("Current entries: \(existingPlaylist.entries?.map(\.id))") } }

Due to hysterical raisins, our Apple Developer Account (2A...FW) has five FairPlay Streaming resources under https://developer.apple.com/account/resources/certificates/list Three of these are certificates (fairplay.cer) and two of them are provisioning packages (fps-bundle.zip). The three certificates all use 1024-bit RSA keys and have creation dates of: Oct 24 23:22:09 2016 (expired Oct 25 23:22:09 2018) 3J Mar 29 19:39:15 2018 (expired Mar 29 19:39:15 2020) 2N Feb 11 00:32:06 2026 (expires Feb 1 00:22:57 2027) LD (I've included the first two characters of the Apple resource ID to help keep these straight.) The key for the first two (same key for both) is lost to the mists of time. The third is a cert I created from a new key, so I have the key for it. (The developer portal will not let us create any more 1024-bit FairPlay certs.) The two FPS bundles each contain an fps_certificate.bin which itself contains a 1024-bit cert and a 2048-bit cert. Looking at this file in each bundle, the bundles include the same 1024-bit cert that I created on Feb 11, but two different 2048-bit certs with creation dates of: Feb 11 00:32:06 2026 (expires Feb 11 00:32:05 2028) YP Feb 11 00:56:22 2026 (expires Feb 11 00:56:21 2028) 9N Both 2048-bit certs use the same key (which I have). Finally, we use a third-party as our streaming provider. With them we shared the first FPS bundle (YP). So, this is a big mess. And I'm unable to delete any of these entries from our developer account. Questions: Does the expiration date on these certs matter or is it ignored for streaming purposes? How do we delete FP certs/bundles we no longer need/use/are expired? With respect to using third-party vendors for streaming (with whom we've shared an FPS bundle): Is it okay to re-use the same FPS bundle if we change vendors? Should we ask Apple to delete an FPS bundle once we stop using a vendor?

An image file whose data is cut in half is accepted by insertOrUpdateAssets, produces no error, and is reported as an image containing zero faces. That result is indistinguishable from a photograph that genuinely has no people in it. In my measurement the intact file reported 4 faces and a copy truncated to half its bytes reported 0 faces — both without throwing. Why this is worse than an error. A client that receives "zero faces, no error" will record the asset as analysed, no people present. That is a permanent, silent, incorrect result: the file is never revisited, and the people in it are lost from the catalogue with no diagnostic anywhere. An error would have been recorded as a failure and retried later. Silence is the one outcome that cannot be recovered from. The failure mode is also inconsistent. Other kinds of damaged input do throw — a zero-byte file, a text file with a .jpg extension, and a file whose interior bytes have been destroyed all raise MediaIntelligenceError.faceGroupProcessing. Truncation is the case that silently succeeds, and truncation is precisely the damage that partial downloads, interrupted copies and failing disks produce — the most common form of corruption in a real photo archive, and the one I have to handle in 97 libraries of mixed provenance. My current workaround is to fully decode every image myself before handing it to the framework, purely to detect truncation. That means every file is decoded twice, which roughly doubles the I/O of the analysis pass. Feedback: FB24174749

If a single asset in the array passed to insertOrUpdateAssets cannot be read, the returned AsyncSequence throws and delivers zero elements — including for the valid assets that appear before the offending one in the array. In my measurement a batch of 5 — two valid photographs, one zero-byte .jpg, then two more valid photographs — delivered 0 of 5 elements and reported 0 faces. The framework's own stdout log shows it had already processed the valid photographs before failing, so the work was done and then thrown away. This is the behaviour I would expect from a function returning [Result] after processing everything, not from a streaming AsyncSequence. The whole reason to expose an async sequence is to deliver results as they are produced; here the sequence produces nothing at all, which makes the streaming shape actively misleading. Why this matters at library scale. I am cataloguing 97 photo libraries, many of them archives of scanned family photographs going back to the 1970s, where a handful of damaged files is normal rather than exceptional. As the API stands, the batch size I choose for throughput is also the amount of work a single corrupt file destroys — with a batch of 100, one bad file costs 100 images. The only safe strategy is to catch the failure and re-submit the batch one asset at a time to find the culprit, which turns a rare bad file into a full re-run of that batch and makes the worst case quadratic in the number of bad files. Note also that the error gives no indication of which asset failed (see the related enhancement request on error taxonomy), so isolating the culprit by re-submission is the only option available. Feedback: FB24174733

insertOrUpdateAssets(_:) is async throws and returns an AsyncSequence, so by the Swift Concurrency contract I expected it to observe cancellation of the enclosing Task and throw CancellationError. It does not. The call runs to completion and returns successfully, with the full result set, as though the cancellation had never happened. In my measurement the cancel was delivered at 2.7 s and the call returned successfully at 19.0 s, having processed all 150 assets and reported 550 faces. The practical consequence for an app is that a "Stop" button cannot stop work that is already in flight. The only way to get responsive cancellation is to slice the work into many small calls and check Task.checkCancellation() between them — which means the batch size, which should be tuned for throughput, ends up being dictated by how long a user is willing to wait after pressing Stop. For me that is the difference between a batch of 150 (≈19 s to react) and a batch of 25 (≈3 s). The store is left in a consistent state, which is good: the assets processed before cancellation remain, and state correctly becomes .stale. So this is specifically about the cancellation signal being ignored, not about data integrity. One detail worth knowing when reproducing this insertOrUpdateAssets is declared nonisolated(nonsending), so it executes on the caller's executor. My first attempt at this reproducer used a plain Task { } created from @MainActor top-level code — which inherits main-actor isolation — and the detection therefore ran on the main actor and starved the very code that was supposed to cancel it: a Task.sleep(2.5s) on the main actor did not resume until the 19-second call had already finished, so the cancel was not even delivered until 19.0 s. The attached reproducer uses Task.detached to avoid that confound, and the cancel is correctly delivered at 2.7 s. I mention it because anyone reproducing this from a @MainActor context will see a different and misleading timeline. Feedback: FB24174707

Constructing a second FaceGroupAnalyzer while a first one is still alive registers the framework's Core Data managed object model a second time. Core Data can then no longer resolve +[MIManagedFace entity], and the process is terminated with SIGTRAP (exit status 133). Three things make this worse than a normal API misuse: It happens with completely separate working directories. The two instances are logically independent — different directories, different stores, no shared state that the API surface exposes. Nothing in the signature of init(workingDirectory:) suggests that two of them cannot coexist, and the parameter's existence implies the opposite. There is no way to detect or prevent it from a library. FaceGroupAnalyzer is not a singleton and offers no way to ask whether an instance already exists in the process. Any two independent subsystems in an app — say, a background analysis service and a foreground preview — that each construct an analyzer will terminate the app. In my case the app must now enforce single-instance access through its own serial queue and lock, which is a constraint the framework imposes but does not state. Construction alone does not fail, so the problem surfaces later and elsewhere. Both instances construct successfully; Core Data only emits warnings at that point. The abort comes when both are alive and one of them does real work (update() and reading the grouping). So the crash lands far from its cause, in code that is individually correct. Feedback: FB24174678

MPVolumeView's showsRouteButton was deprecated (https://developer.apple.com/documentation/mediaplayer/mpvolumeview/showsroutebutton?language=objc). It's not clear how can we now hide this button without deprecation warning. The documentation is lacking. Please advise. Thank you!

visionOS 26.5, xcode26.5 - app terminated with exit code 9 then crashed and rebooted the entire device (Apple Vision Pro). I was connected to the Xcode debugger when this happened, and it didn't crash in any of our code. Memory and CPU usage was low at the time. Any idea what could be causing the issue? Some logs: << FigVideoTargetRemoteXPC >> signalled err=-15562 at <>:868 ... Call start on AVKSDockingService before making requests. <<<< FigPlayerInterstitial >>>> signalled err= 18,446,744,073,709,535,945 at <>: 10,773 <<<< FigPlayerInterstitial >>>> signalled err= 18,446,744,073,709,535,945 at <>: 10,773 << FigVideoTargetRemoteXPC >> signalled err=-15562 at <>:868 <<<< PlayerRemoteXPC >>>> signalled err= 18,446,744,073,709,538,756 at <>: 1,538 SessionCore_NotificationHandlers.mm : 73 Server returned an error:. Error Domain=NSOSStatusErrorDomain Code=-50 "Session lookup failed" UserInfo={NSLocalizedDescription=Session lookup failed} <<<< PlayerRemoteXPC >>>> signalled err= 18,446,744,073,709,538,756 at <>: 1,538 ... nw_read_request_report [C 1 ] Receive failed with error " No message available on STREAM " nw_protocol_socket_reset_linger [C1:2] setsockopt SO_LINGER failed 22 Debug session ended with code 9: Terminated due to signal 9 Program ended with exit code: 9 Thanks, bvsdev

We are testing an Osmo Mobile 8 through Apple's DockKit framework. Configuration: iPhone 15 Pro Max iOS 26.5.2 (23F84) DJI Osmo Mobile 8 DockKit model identifier: DS308 Firmware reported through DockKit: 1.0.0 Apple's official DockKit camera sample used as the test application We disabled DockKit system tracking and commanded yaw movement only: Vector3D(x: 0, y: 0.2, z: 0). The pitch component remained zero throughout the test. Short left and right movements worked and kept the reported tilt reasonably stable. We then performed a complete yaw rotation. Pan completed successfully, including the expected angle wrap from approximately +150° to -170°. However, DockKit's reported tilt changed from approximately -8° at the starting heading to approximately -37° around the rear of the rotation. It returned to approximately -7° when the gimbal completed the rotation and returned to its original heading. Selected telemetry (Pan / Reported tilt): -0.86° / -8.08° +94.48° / -21.20° +149.89° / -33.35° -169.88° / -36.96° -89.90° / -25.95° -5.67° / -8.59° +1.95° / -7.05° We also found that Apple's sample application's manual chevrons did not initially produce visible movement, although direct calls to the same setAngularVelocity API worked reliably. Questions: Should the Osmo Mobile 8 maintain its physical horizon during a DockKit yaw-only command? Does this model provide a horizon-lock or leveling mode that third-party DockKit applications need to select? Is firmware 1.0.0 the expected DockKit-reported firmware version for this model? Is the changing tilt related to the Osmo's gimbal geometry, firmware stabilization, or DockKit coordinate reporting? Is there newer firmware or a recommended DJI test application we should use for comparison? The change was also visible in the phone's physical orientation, so it does not appear to be telemetry-only.

Hello, I am developing BROXMEDIA Intercom, an iOS intercom application for live audiovisual production. The app uses a native Swift audio plugin, Google WebRTC, AVAudioSession, and a Capacitor user interface. The current TestFlight version is 2.0, build 17. Environment: iPhone 16 Pro Max iOS 26.5.2 TestFlight internal build AVAudioSession category: playAndRecord AVAudioSession mode: voiceChat Background audio capability enabled Bidirectional WebRTC audio between a web browser and the native iOS app Observed behavior: A remote web browser publishes WebRTC audio. The native iOS app receives and plays the audio correctly for approximately one hour. Wi-Fi disconnection/reconnection and airplane mode on/off initially recover correctly. After the prolonged session, the native app stops playing the remote audio. Signaling and participant presence remain connected. The remote participant is still shown as speaking. Completely closing and reopening the native app does not recover the remote audio. Safari on the same iPhone, connected to the same room and network, can still hear the same remote transmission. Restarting the remote web publication usually causes the native app to receive audio again. This suggests that the remote publication, network connection, signaling server, and device audio hardware are still operational when the native route fails. We are investigating whether: AVAudioSession or the underlying WebRTC audio unit has stopped rendering; the native RTCPeerConnection retains a stale receiver or audio track; inbound RTP has stopped even though the peer remains connected; an interruption, route change, or media-services reset has not been fully recovered. Our current recovery logic checks the peer connection state and whether a remote audio track object exists. However, we do not yet continuously verify that inbound RTP packets or bytes are increasing for each participant. Questions: Can AVAudioSession or its underlying audio unit stop rendering audio while RTCPeerConnection signaling remains connected? Which AVAudioSession or audio-unit callbacks should be monitored to distinguish an iOS audio-session failure from a WebRTC receiver or inbound-RTP failure? After AVAudioSession.mediaServicesWereResetNotification, should an app recreate the complete WebRTC audio engine, or is reactivating AVAudioSession normally sufficient? Is monitoring inbound RTP progression and audio energy the recommended way to detect a remote audio track that still exists but is no longer delivering usable audio? Are there any known considerations for prolonged bidirectional VoIP-style audio using playAndRecord, voiceChat, and background audio? We can add diagnostic logging and provide a Feedback Assistant report with sysdiagnose if the problem is reproduced again. Thank you.

So, in macos golden gate 27, there is this new live subtitles that happens in every video that i can't even turn off? (live captions is off) and also is there anyway from code we can turn this feature off, because making a background app, it exists whenever, and i can't seem to find a way to disable it from coding and the app.

I am testing SFSpeechRecognizer using the en_US locale. When the iPhone Simulator’s system language is set to Japanese, SFSpeechRecognizer.isAvailable returns false for en_US, so speech recognition is unavailable. As far as I have tested, this issue does not occur on iOS Simulator 26.2 or earlier. Is this a Simulator-specific issue, or is it a behavior change that could also occur on physical devices? Has any additional setup become necessary to use speech recognition in Simulator? I then changed the iPhone Simulator’s system language to English. After doing so, SFSpeechRecognizer.isAvailable returned true for en_US. However, starting a recognition task still failed immediately with kLSRErrorDomain Code=300, “Failed to initialize recognizer.” The following error was returned: Error Domain=kLSRErrorDomain Code=300 "Failed to initialize recognizer" UserInfo={ NSLocalizedDescription=Failed to initialize recognizer, NSUnderlyingError={ Error Domain=kLSRErrorDomain Code=300 "Failed to create recognizer from=/Users/<USERNAME>/Library/Developer/CoreSimulator/Devices/<DEVICE-UUID>/data/private/var/MobileAsset/AssetsV2/com_apple_MobileAsset_UAF_Siri_Understanding/purpose_auto/ c079bfa6b8856202dc8cb2135fef3b06229ced6e.asset/AssetData/mini.json" UserInfo={ NSLocalizedDescription=Failed to create recognizer from=/Users/<USERNAME>/Library/Developer/CoreSimulator/Devices/<DEVICE-UUID>/data/private/var/MobileAsset/AssetsV2/com_apple_MobileAsset_UAF_Siri_Understanding/purpose_auto/ c079bfa6b8856202dc8cb2135fef3b06229ced6e.asset/AssetData/mini.json } } } I have not observed either of these issues on iOS Simulator 26.2 or earlier: With the Simulator language set to Japanese, the en_US recognizer does not become unavailable. When SFSpeechRecognizer.isAvailable is true, recognition does not fail with kLSRErrorDomain Code=300. Environment Xcode: 26.6 (17F113) iOS Simulator 26.5 (23F77), iPhone 17 (A3258J/A) iOS Simulator 26.4 (23E244), iPhone 17 (A3258J/A)

Hi All,I am working on a project to turn my iPhone to silent mode via Code (Swift), Can someone ,plz, help to put on the right direction as I am very new to Xcode and Swift.Regards

Hey y'all, I'm new here and in the process of building some audio software. I'm hitting a roadblock trying to incorporate VPIO into audio playback states without it ducking what's currently playing in my DAW. I have a few questions: Can VPIO other-audio ducking be completely disabled on macOS, beyond fixed/minimum? Minimum is not enough for professional audio environments. Is attenuation of an unrelated application routed through a different Core Audio output device expected? Is split-device VPIO—built-in microphone input with Apollo/Universal Audio output—supported? Why can AVAudioEngine.start() succeed while the engine remains stopped and immediately emits a configuration-change notification? Thanks for any insight into this!

Hi! I was wondering if there is currently a method that you have been using to retrieve information on credits (i.e. mixers, producers, engineers etc.) via the Apple Music API as an endpoint. I know it is possible to retrieve performing artists but would find it very useful to export data on the "Composition & Lyrics" and "Production & Engineering" sections when you view a songs credits on the front end as seen in the attached image. If not, is this a feature that is being currently worked on? Would be so useful for developers working with large catalogues. Thanks! Spencer

Hello, I cannot find reference to any possible limitation on ShazamKit. How many calls can I make per minute? Is there any limit at all? Thanks

I'm observing an interesting behavior on iOS when uploading edited videos from the Photos library. Scenario Device: iPhone Video: 4K, 60 FPS Original video size (Photos app): 660 MB Uploaded original asset: 660 MB After applying a simple edit (e.g., a filter) in the Photos app: Photos app still shows the edited video. When my app uploads the current/edited asset, the uploaded file size becomes 167 MB. So the edited version is approximately 75% smaller than the original (660 MB → 167 MB), even though the duration appears unchanged. Questions Is this expected behavior for edited videos in the Photos library? Does iOS automatically re-encode edited videos using a lower bitrate or a different codec (e.g., HEVC) when generating the current rendition? Does the Photos app continue to display the size of the original asset rather than the size of the edited rendition? I'm aware that PhotoKit allows retrieving both the original and the current (edited) versions of a PHAsset. In this case, I'm intentionally retrieving the current version, and this behavior is observed only for the edited asset.

On iOS 26, the first ProRes RAW recording after launching the app consistently stalls for the entire take: audio records normally, but video contains only a few frames (for example, an 11.8-second clip at approximately 0.25 fps instead of 30 fps). Every subsequent recording works correctly. Before startRecording, the Bayer format is active, ProRes RAW is available, white balance is locked, and frame duration is pinned to 1/30. Reordering setOutputSettings and reasserting configuration have not resolved it; only discarding the first recording acts as a reliable warm-up. Is this a known one-time ProRes RAW encoder initialisation issue, and is there a supported way to prepare the encoder before the first user takes? I have a full working sample code if anyone needs but here is the setup function: import CoreMedia import CoreVideo @available(iOS 26.0, *) final class ProResRAWRecorder: NSObject, AVCaptureFileOutputRecordingDelegate { let session = AVCaptureSession() private let movieOutput = AVCaptureMovieFileOutput() private var camera: AVCaptureDevice! // Call on a serial capture queue. func configure() throws { session.beginConfiguration() defer { session.commitConfiguration() } session.sessionPreset = .inputPriority session.automaticallyConfiguresCaptureDeviceForWideColor = false guard let device = AVCaptureDevice.default( .builtInWideAngleCamera, for: .video, position: .back ) else { throw SampleError.noCamera } let cameraInput = try AVCaptureDeviceInput(device: device) guard session.canAddInput(cameraInput) else { throw SampleError.cannotAddInput } session.addInput(cameraInput) camera = device // Optional, but reproduces the real topology where audio remains healthy. if let microphone = AVCaptureDevice.default(for: .audio), let microphoneInput = try? AVCaptureDeviceInput(device: microphone), session.canAddInput(microphoneInput) { session.addInput(microphoneInput) } // Find a 12-bit packed Bayer format supporting 30 fps. guard let rawFormat = device.formats .filter({ CMFormatDescriptionGetMediaSubType($0.formatDescription) == kCVPixelFormatType_96VersatileBayerPacked12 && $0.videoSupportedFrameRateRanges.contains { $0.minFrameRate <= 30 && $0.maxFrameRate >= 30 } }) .max(by: { let a = CMVideoFormatDescriptionGetDimensions($0.formatDescription) let b = CMVideoFormatDescriptionGetDimensions($1.formatDescription) return Int(a.width) * Int(a.height) < Int(b.width) * Int(b.height) }) else { throw SampleError.noRAWFormat } try device.lockForConfiguration() device.activeFormat = rawFormat if rawFormat.supportedColorSpaces.contains(.appleLog) { device.activeColorSpace = .appleLog } let frameDuration = CMTime(value: 1, timescale: 30) device.activeVideoMinFrameDuration = frameDuration device.activeVideoMaxFrameDuration = frameDuration if device.isWhiteBalanceModeSupported(.locked) { device.whiteBalanceMode = .locked } device.unlockForConfiguration() guard session.canAddOutput(movieOutput) else { throw SampleError.cannotAddOutput } session.addOutput(movieOutput) session.startRunning() } // Call on the same serial capture queue after startRunning() returns. func record(to url: URL) throws { guard let connection = movieOutput.connection(with: .video) else { throw SampleError.noVideoConnection } // Reassert the required device state at the take boundary. try camera.lockForConfiguration() if camera.isWhiteBalanceModeSupported(.locked) { camera.whiteBalanceMode = .locked } let frameDuration = CMTime(value: 1, timescale: 30) camera.activeVideoMinFrameDuration = frameDuration camera.activeVideoMaxFrameDuration = frameDuration camera.unlockForConfiguration() guard movieOutput.availableVideoCodecTypes.contains(.proResRAW) else { throw SampleError.rawCodecUnavailable } movieOutput.setOutputSettings( [AVVideoCodecKey: AVVideoCodecType.proResRAW], for: connection ) movieOutput.startRecording(to: url, recordingDelegate: self) } func stop() { if movieOutput.isRecording { movieOutput.stopRecording() } } func fileOutput( _ output: AVCaptureFileOutput, didFinishRecordingTo outputFileURL: URL, from connections: [AVCaptureConnection], error: Error? ) { print("Finished:", outputFileURL, "error:", error as Any) } enum SampleError: Error { case noCamera case cannotAddInput case noRAWFormat case cannotAddOutput case noVideoConnection case rawCodecUnavailable } }