Hi everyone, I’m encountering a memory overflow issue in my visionOS app and I’d like to confirm if this is expected behavior or if I’m missing something in cleanup. App Context The app showcases apartments in real scale using AR. Apartments are heavy USDZ models (hundreds of thousands of triangles, high-resolution textures). Users can walk inside the apartments, and performance is good even close to hardware limits. Flow The app starts in a full immersive space (RealityView) for selecting the apartment. When an apartment is selected, a new ImmersiveSpace opens and the apartment scene loads. The scene includes multiple USDZ models, EnvironmentResources, and dynamic textures for skyboxes. When the user dismisses the experience, we attempt cleanup: Nulling out all entity references. Removing ModelComponents. Clearing cached textures and skyboxes. Forcing dictionaries/collections to empty. Despite this cleanup, memory usage remains very high. Problem After dismissing the ImmersiveSpace, memory does not return to baseline. Check the attached screenshot of the profiling made using Instruments: Initial state: ~30MB (main menu). After loading models sequentially: ~3.3GB. Skybox textures bring it near ~4GB. After dismissing the experience (at ~01:00 mark): memory only drops slightly (to ~2.66GB). When loading the second apartment, memory continues to increase until ~5GB, at which point the app crashes due to memory pressure. The issue is consistently visible under VM: IOSurface in Instruments. No leaks are detected. So it looks like RealityKit (or lower-level frameworks) keeps caching meshes and textures, and does not free them when RealityView is ended. But for my use case, these resources should be fully released once the ImmersiveSpace is dismissed, since new apartments will load entirely different models and textures. Cleanup Code Example Here’s a simplified version of the cleanup I’m doing: func clearAllRoomEntities() { for (entityName, entity) in entityFromMarker { entity.removeFromParent() if let modelEntity = entity as? ModelEntity { modelEntity.components.removeAll() modelEntity.children.forEach { $0.removeFromParent() } modelEntity.clearTexturesAndMaterials() } entityFromMarker[entityName] = nil removeSkyboxPortals(from: entityName) } entityFromMarker.removeAll() } extension ModelEntity { func clearTexturesAndMaterials() { guard var modelComponent = self.model else { return } for index in modelComponent.materials.indices { removeTextures(from: &modelComponent.materials[index]) } modelComponent.materials.removeAll() self.model = modelComponent self.model = nil } private func removeTextures(from material: inout any Material) { if var pbr = material as? PhysicallyBasedMaterial { pbr.baseColor.texture = nil pbr.emissiveColor.texture = nil pbr.metallic.texture = nil pbr.roughness.texture = nil pbr.normal.texture = nil pbr.ambientOcclusion.texture = nil pbr.clearcoat.texture = nil material = pbr } else if var simple = material as? SimpleMaterial { simple.color.texture = nil material = simple } } } Questions Is this expected RealityKit behavior (textures/meshes cached internally)? Is there a way to force RealityKit to release GPU resources tied to USDZ models when they’re no longer used? Should dismissing the ImmersiveSpace automatically free those IOSurfaces, or do I need to handle this differently? Any guidance, best practices, or confirmation would be hugely appreciated. Thanks in advance!

苹果12内存不够用，怎么办

We're trying to switch from using main camera access on Arkit to screen-capture with passthrough however we're facing some issues and it seems a bit complicated to debug. We have set up a broadcast Extension, set up some logs on the sample Handler but we get nothing in the console nor that the recording starts, we set up the picker as well and we can see our extension in the control center as one of the choices but clicking start, results in it stopping in less than one second after. The only message that is rather contradictory we see in the console.app is the following [INFO] -[RPRecordingManager getSystemBroadcastExtensionInfo:]_block_invoke:1333 Extension has passthrough license and just right after [INFO] -[RPRecordingManager getSystemBroadcastExtensionInfo:]_block_invoke:1336 Extension does not have passthrough license

直播过程中需同时操作 Vision Pro（拍摄）、Mac（推流）、中控台（画面切换），无统一控制入口，调节 3D 模型、校准画质等操作需在多设备间切换，易出错且效率低。 期望 针对直播场景，提供桌面端专属控制软件，支持一站式管理 Vision Pro 的拍摄参数、3D 模型切换、虚实融合效果等，实现多设备协同操作的可视化、便捷化。

画面亮度存在无规律动态波动（时亮时暗），且无手动控制入口，导致商品颜色还原失真、主播面部曝光异常（过曝 / 欠曝），严重影响直播展示效果。 期望 "· 优化直播模式的自动曝光算法，提升复杂光线环境下的亮度稳定性； · 增加 “直播模式” 专属亮度锁定功能，支持手动设定亮度参数并锁定，满足直播场景下的画质可控需求。 "

切换后两者的亮度、色彩饱和度、对比度等画质参数差距较大，导致画面视觉体验割裂，破坏直播连贯性，影响用户观看沉浸感。 期望 "· 对标常规直播单反相机的画质基准，优化 Vision Pro 的画面亮度、色彩还原能力； · 提供设备端或配套软件的画质自定义调节功能（亮度、对比度、色温等），支持直播前手动校准，确保与单反相机画面风格一致。"

传输后的直播流分辨率显著下降，画面细节丢失、清晰度不足，导致 3D 家具商品的纹理、尺寸等关键信息无法精准展示，影响用户对商品的判断。 期望 优化流传输过程中的分辨率压缩策略，减少传输过程中的画质损耗，提升 Mac 端接收的直播流清晰度，匹配 3D 商品展示的高精度需求。

佩戴者头部自然晃动时，设备拍摄的画面会出现明显抖动，导致观看直播的用户产生眩晕感，严重影响直播沉浸体验和购物决策效率。 希望 优化设备内置防抖算法，降低头部常规晃动对画面稳定性的影响，提升直播画面的流畅度。

Hello, Want to understand what's the current state for developing for Apple Vision Pro? I want to stream a video from a remote server in realtime. It is a video stream and can't download it. I want to stream a low quality stream and high res stream. The server will only send the "box" where user is looking at. Are there any API to track where the user is looking at in the experience? Thanks,

I'm trying to run a PhotogrammetrySession based on photos taken in an AVCaptureSession and stored as .heic files. When I load the files I'm always seeing the error "Sample 0 missing LiDAR point cloud!" showing up for each individual sample. Debugging shows that sample.depthDataMap is populated, also the .heic contains depth data which can be extracted using e.g. heif-convert on my Mac. Comparing the .heic I created to one of the ObjectCaptureSession which doesn't show the LiDAR warning, I noticed the only difference being the HEIC information here: So my questions are: Are these the missing information in my manual capture causing this warning? Can I somehow add these information in an AVCaptureSession? Do these information allow better photogrammetry results?

Hello, I am trying to build an AVP app for real-time "zero-latency" spatial video streaming. I am trying to figure out, on a high level, the best way to do this. Currently this is my method: Server sends stereo images via a WebRTC service (ie, livekit) The WebRTC stream is converted to a CVPixelBuffer, writes them to file, plays via AVPlayer, and applies a VideoMaterial to a plane entity. However, this is a bit hacky and it seems like this won't be compatible with Apple's spatial experinces. To my understanding, Apple supports HLS streaming for spatial experiences and APMP content. However, HLS (and even Low Latency HLS) introduces a second or more of latency, likely do to the segmentation nature of HLS. Thus, HLS will not work for us. Some other alternatives I've thought of are streaming the live stream video via webrtc from the server to a local computer in the AVP's network, and then using LL-HLS to stream from the local computer to the vision pro. Still, it seems like this would introduce latency on the order of seconds. Is my current approach the best way to implement this? Or could anyone suggest a better way, perhaps something compatible with AVP's spatial experiences

I have an open Feedback conversation with Apple on this topic, but I am curious if others have run into this, or want to try out my sample code in their set up. there are two API’s for reading controller buttons, axis, and D pads: GCPhysicalInputProfile and GCControllerLiveInput. There are inconsistencies in behaviour between the two of them. Apple recommends we use GCControllerLiveInput, however, there are some capabilities on these controllers that are only accessible through GCPhysicalInputProfile, as I’ll discuss below. PSVR2 R2/L2 buttons, a.k.a. triggers, have force input analogue values. These can only be accessed on GCPhysicalInputProfile PSVR2 thumbstick direction values are read through “axes” on GCPhysicalInputProfile, but only “dpads” on GCControllerLiveInput on both GCPhysicalInputProfile and GCControllerLiveInput, All pressed events of all buttons are fired properly using generic aliases ( Trigger, Grip ,Menu, Right Thumbstick, Left Thumbstick, Right Button A & B (Circle & Cross), Left Button A&B (Triangle and Square) ). Apple reserves the system button as the equivalent of a home button for the OS. on GCPhysicalInputProfile, touch events are fired when the button is also pressed, but not for only touches. on GCControllerLiveInput , Touch events only works for the following buttons: Left Thumbstick, Right Thumbstick, Right Button A (Circle), and Right Button B (Cross). But Right Button B touch event isn’t labelled correctly, it fires as the Right Button A event. I observed this inside ALVR which uses a polling based approach to event processing: https://github.com/alvr-org/alvr-visionos/blob/17b5968f9d894944b53e97134b39dfce0993302a/ALVRClient/WorldTracker.swift#L301 To simplify to see this on a very simple app, I used the Apple example TrackingAccessories application: https://developer.apple.com/documentation/ARKit/tracking-accessories-in-volumetric-windows I’ve attached the code that replaces the AccessoryTrackingModel class. I added code that prints out what is touched/pressed, see the trackAllConnectedSpatialControllers method: https://github.com/svrc/TrackingAccessories

I posted https://developer.apple.com/forums/thread/809481 yesterday about an issue I discovered with pushWindow in visionOS 26.2 RC, but today I discovered a second problem with pushWindow. If window A calls pushWindow to present window B, and the user pins window B to a wall, the following unexpected behaviors are observed: Window B spontaneously disappears. If the user re-launches the (still running) app from the visionOS home view, both window A and window B appear simultaneously. I assume only window B should be visible at this point, since window A pushed window B. If the user closes window B, it's now impossible to present window B again. Calls to pushWindow appear to be ignored. If the user force-quits the app and relaunches it, and pushWindow is called again, window B appears, but window A remains visible. I also noticed this surprising behavior: This broken state of pushWindow behavior now affects all other apps on the system that may call pushWindow in the future, not just the app whose pushed window was pinned above. A workaround is to reboot the device, and then the system will behave as expected until the next time the user pins a pushed window.

I recently added pushWindow to my app, and I discovered that in visionOS 26.2 RC (23N301), pushWindow followed by dismissWindow no longer works as expected. Specifically, if the user moves the pushed window, then when the pushed window is later dismissed, the parent window's position isn't aligned with the pushed window's new position. Its original position is restored instead. Curiously, the bug only happens when an app is launched from the visionOS home view, and not when an app is launched from Xcode. It also doesn't happen in the visionOS 26.2 simulator. Another interesting detail is that while the parent window is hidden, if the user long-presses the Digital Crown and then dismisses the pushed window, the parent window's position seems to be immune from the Digital Crown scene reorientation. It's restored to its original real world position. Demo video: https://youtu.be/zR3t2ON3Wz0 I've submitted feedback as FB21287011 with a sample app and detailed repro steps. Has anyone else encountered this issue already and figured out a workaround? It would be nice if I could get pushWindow to work correctly in my app. Thanks everybody! 😀

Hello, I am in the process of implementing SharePlay support in my visionOS app. Everything runs fine when I test locally, but when my app is distributed via TestFlight, calling try await activity.activate() shows the SharePlay dialog as usual, but then when I start a new FaceTime call, my ImmersiveSpace gets dismissed. This is only happening when the app is distributed via TestFlight, when I run it locally the ImmersiveSpace stays active as expected. Looking at the console on my Mac I found this log: Invalid initial client settings class: UIApplicationSceneClientSettings; expected class: MRUISharedApplicationSceneClientSettings; bundle ID: com.apple.facetime; scene ID: com.apple.facetime:SFBSystemService-DDA8C751-C0C4-487E-AD85-59EF4E6C6050 Does anyone have an idea how I can fix this? It's driving me nuts and I wasted over a day looking for a workaround but so far been unsuccessful. Thanks!

https://developer.apple.com/documentation/realitykit/videomaterial The documentation: "Video materials support transparency if the source video’s file format also supports transparency." I have a transparency video(Hand.mov, HEVC with alpha), I can show the video with transparency background correctly on Vision Pro Simulates, but on physic Device the video has a black background. I'm sure the video format is ok because I can see get the texture from video and display it on an UnlitMaterial. How can I show the transparency video correctly with the RealityKit/VideoMaterial?

After adding TextComponents to my Entities on visionOS, I have observed that visualBounds will ignore the TextComponents. Documentation states that it should render a rounded rectangle mesh. These mashes are visible on the device, but not visible in the debugger ("Capture Entity Hierarchy") and ignored by visualBounds. Am I missing something? static func makeDirection(_ direction: Direction) -> Entity { let text = Entity() text.name = direction.rawValue text.setScale(SIMD3(repeating: 5), relativeTo: nil) text.transform.rotation = direction.rotation text.components.set(direction.textComponent) return text } My workaround is to add a disabled ModelEntity and take its bounds 😬

Environment visionOS 26.1, Xcode 26.1.1 Problem When a WindowGroup opens an ImmersiveSpace and the user closes the window via X button, the async Task in .onDisappear gets cancelled before dismissImmersiveSpace() completes, leaving the ImmersiveSpace active with no way to exit. Steps WindowGroup opens ImmersiveSpace in .onAppear User clicks X to close window .onDisappear fires but async cleanup cancelled ImmersiveSpace remains active, user trapped Expected ImmersiveSpace dismissed when window closes Actual ImmersiveSpace remains active Code .onAppear { Task { await openImmersiveSpace(id: "VideoCallMainCamera") } } .onDisappear { Task { await dismissImmersiveSpace() // Gets cancelled } } What I've Tried Task in .onDisappear ❌ scenePhase monitoring ❌ High priority Task ❌ .restorationBehavior(.disabled) + .defaultLaunchBehavior(.suppressed) ✅ (prevents restoration but doesn't fix immediate cleanup) Question What's the recommended pattern for ensuring ImmersiveSpace cleanup when WindowGroup closes? Is there a way to block window closure until async cleanup completes, or should ImmersiveSpaces automatically dismiss with their parent window?

Hi all, I am currently developing a game in Unity for VisionOS and I'd prefer to use the PSVR2 controllers as a source of the raycast for menu selection instead of the default VisionOS gaze for my specific use case. Is there a way to access the IMU of PSVR2 controllers to do this instead of just using eyegaze + controller click for selection? Is there a specific configuration for GCController from within Unity maybe? Thank you!

Hi, We’ve been successfully using the RoomPlan API in our application for over two years. Recently, however, users have reported encountering persistent capture errors during their sessions. Specifically, the errors observed are: CaptureError.worldTrackingFailure CaptureError.exceedSceneSizeLimit What we have observed: Persistent Errors: The errors continue to occur even after initiating new capture sessions. Normal Usage: Our implementation adheres to typical usage patterns of the RoomPlan API without exceeding any documented room size limits. Limited Feature Usage: We are not utilizing the WorldTracking feature for the StructureBuilder functionality to stitch rooms together. Potential State Caching: Given that these errors persist across sessions, we suspect that there might be memory or state cached between sessions that is not being cleared, particularly since we are not taking advantage of StructureBuilder. Request: Could you please advise if there is any internal caching or memory retention between capture sessions that might lead to these errors? Additionally, we would appreciate guidance on how to clear or manage this state when the StructureBuilder feature is not in use. Here is a generalised version of our capture session initialization code to help diagnose the issue. struct RoomARCaptureView: UIViewRepresentable { typealias Handler = (CapturedRoom, Error?) -> Void @Binding var stop: Bool @Binding var done: Bool let completion: Handler? func makeUIView(context: Self.Context) -> RoomCaptureView { let view = RoomCaptureView(frame: .zero) view.delegate = context.coordinator view.captureSession.run(configuration: .init()) return view } func updateUIView(_ uiView: RoomCaptureView, context: Self.Context) { if stop { // Stop the session only once, multiple times causes issues with the final presentation uiView.captureSession.stop() stop = false done = true } } static func dismantleUIView(_ uiView: RoomCaptureView, coordinator: Self.Coordinator) { uiView.captureSession.stop() } func makeCoordinator() -> ARViewCoordinator { ARViewCoordinator(completion) } @objc(ARViewCoordinator) class ARViewCoordinator: NSObject, RoomCaptureViewDelegate { var completion: Handler? public required init?(coder: NSCoder) {} public func encode(with coder: NSCoder) {} public init(_ completion: Handler?) { super.init() self.completion = completion } public func captureView(shouldPresent roomDataForProcessing: CapturedRoomData, error: (Error)?) -> Bool { return true } public func captureView(didPresent processedResult: CapturedRoom, error: (Error)?) { completion?(processedResult, error) } } } Thank you for your assistance.