How to Fix Lock Screen Depth Effect Wallpaper Not Separating Subject on iPhone 13 Pro Max: Complete Diagnostic Manual
⚡ Quick 60-Second Verdict: What This Issue Means
The Depth Effect on your iPhone 13 Pro Max fails to separate the subject because the VisionKit neural engine cache—responsible for real-time person segmentation on the A15 Bionic’s Neural Engine—has become corrupted or is referencing stale depth maps from a previous iOS version. This is purely a software-level cache inconsistency, not a hardware fault. The fix path: clear the VisionKit cache via a targeted lock screen reset sequence, then force a fresh segmentation pass by reassigning the wallpaper with “Depth Effect” toggled off-then-on.
- Do not restore via DFU mode for this issue—it wipes the Secure Enclave and Neural Engine calibration data, requiring a full Apple Configurator 2 revive that can brick the A15 Bionic if interrupted over Lightning.
- Never use third-party “cache cleaner” apps claiming to fix Neural Engine caches; they cannot access the VisionKit sandbox (com.apple.VisionKit) and often corrupt the Core ML model store, breaking Face ID and Photographic Styles permanently.
- Avoid heating the device to “reset sensors”—the iPhone 13 Pro Max’s LiDAR scanner and TrueDepth module are calibrated at factory with sub-micron precision; thermal cycling delaminates the VCSEL array from the flood illuminator.
- Do not insert metal tools into the Lightning port to “clean contacts” for software issues; the port carries 9V/2.2A USB-PD negotiation lines—shorting them fries the Tristar/Hydra power IC, requiring board-level microsoldering.
1. Immediate Safe Actions (Non-Destructive User Triage)
Three straightforward, non-destructive initial triage steps tailored specifically to Depth Effect subject separation failure on iPhone 13 Pro Max:
Verify Source Image Meets Depth Effect Requirements
The A15 Neural Engine requires a Portrait-mode HEIC with an embedded depth map (stored in the auxiliary image chunk) or a Live Photo captured on iPhone 13 Pro Max itself. Screenshots, downloaded JPEGs, or images edited in third-party apps strip the depth metadata VisionKit needs. Action: Open Photos → select the wallpaper candidate → tap i icon → confirm Portrait badge and Depth Effect label appear. If missing, recapture in Portrait mode on this device.
Toggle Depth Effect Off/On with Fresh Wallpaper Assignment
This forces VisionKit to discard the cached segmentation mask and regenerate it from the source depth map. Action: Long-press Lock Screen → Customize → tap wallpaper thumbnail → Depth Effect toggle OFF → Done → wait 5 seconds → re-enter Customize → tap wallpaper → Depth Effect ON → Done. Watch for the subtle parallax shift confirming subject separation. If the subject remains flat, proceed to Step 03.
Clear VisionKit User Cache via Lock Screen Reset
iOS stores per-lock-screen VisionKit segmentation caches in /var/mobile/Library/Caches/com.apple.VisionKit/LockScreen/. These survive reboots but are purged when the lock screen layout is deleted. Action: Long-press Lock Screen → swipe up on the problematic custom lock screen → Delete (trash icon) → confirm. Create a new custom lock screen from scratch using the same Portrait image. This forces a clean VKSegmentationRequest pipeline initialization on the Neural Engine.

2. Software False Alarm Recovery & Reset Protocol
iPhone 13 Pro Max Hardware Reset Sequence
This sequence clears volatile Neural Engine SRAM registers and restarts the visionkitd and lockscreenui daemons without affecting user data. Exact key combination for iPhone 13 Pro Max (Face ID, no Home button):
- Press and release Volume Up (1 press, < 0.5s)
- Press and release Volume Down (1 press, < 0.5s)
- Press and hold Side Button (right edge) — continue holding through the “slide to power off” screen — until the screen goes completely black and the Apple logo appears (~15-20s total hold)
- Release Side Button at Apple logo; allow full boot to Lock Screen
Why this works: The A15’s Neural Engine firmware (loaded by AppleNPE driver) reinitializes its weight buffers on cold boot. A standard restart (Settings → General → Shut Down) keeps the Neural Engine in a low-power retention state, preserving the corrupted cache. This hard reset forces a full NPE_RESET vector.
3. Advanced Diagnostics & Component Triage
Hardware Telemetry & Sensor Verification
If Steps 01-04 fail, verify the TrueDepth and LiDAR subsystems—the primary depth sources for VisionKit—are reporting valid data. This requires macOS Console.app (no third-party tools).
- Connect iPhone 13 Pro Max to Mac via Apple Lightning to USB-C cable (or Lightning to USB-A). Trust the computer on iPhone.
- Open Console.app (Applications → Utilities). Select your iPhone in the left sidebar under “Devices”.
- In the search field, type:
subsystem:com.apple.VisionKit category:segmentationand press Return. - On iPhone, open Camera → Portrait mode → point at a person 1-3m away. Observe Console for
VKSegmentationRequest completedentries withconfidence > 0.85. - If Console shows
VKErrorDomain Code=1101 "No valid depth map"orLiDAR timestamp mismatch, the LiDAR scanner (rear) or TrueDepth depth module (front) has a calibration drift. This requires Apple Authorized Service Provider (AASP) diagnostics—do not attempt self-repair; the VCSEL alignment is sub-micron.
Alternative on-device check (iOS 17+):: Settings → Privacy & Security → Analytics & Improvements → Analytics Data → search for VisionKit. Look for VKSegmentationFailure entries with depthMapInvalid: true.

4. Diagnostic & Operating Parameter Matrix
| Diagnostic Indicator / Check | Expected Normal State | Fault Alert Threshold | Verified Resolution Target |
|---|---|---|---|
| Source Image Depth Metadata (Photos → Info panel) |
“Portrait” badge + “Depth Effect” label visible; HEIC container with auxiliary depth map chunk (16-bit disparity map, 640×480) |
Missing Portrait badge; JPEG/HEIC without auxiliary chunk; “Edited” badge from third-party app | Recapture in native Camera Portrait mode on iPhone 13 Pro Max; Depth Effect toggle shows subject separation immediately |
| VisionKit Segmentation Confidence (Console.app: subsystem:com.apple.VisionKit) |
VKSegmentationRequest completed confidence=0.92±0.05 within 120ms on A15 NE |
confidence < 0.70 or VKErrorDomain Code=1101 (no valid depth map) |
Confidence ≥ 0.85 sustained across 5 consecutive Portrait captures; Lock Screen Depth Effect renders subject parallax |
| Lock Screen Cache Integrity (/var/mobile/Library/Caches/com.apple.VisionKit/LockScreen/) |
Directory contains segmentation_ files < 24h old; each < 512KB |
Cache files > 7 days old; size > 2MB (corrupted multi-frame buffer); or directory missing entirely | After lock screen delete/recreate: new cache file created on first Depth Effect toggle; size 120-300KB; subject separates |
| Neural Engine Firmware Version (Settings → General → About → scroll to “Neural Engine Firmware” on iOS 17.4+) |
Matches iOS build (e.g., iOS 18.2 → NEFW 18.2.0.0.1.15) |
Version mismatch (e.g., iOS 18.2 showing NEFW 17.5.0) — indicates failed OTA firmware payload |
Update to latest iOS via Settings → General → Software Update (OTA); NE firmware updates only via full OTA, not delta |
| LiDAR / TrueDepth Depth Stream (Console.app: subsystem:com.apple.depthstream) |
DepthStreamManager: valid frame rate 30fps, timestamp delta 33.3ms ±0.5ms |
timestamp delta > 50ms or frameDropped: true > 5% of frames |
Stable 30fps depth stream for 10s continuous; Lock Screen Depth Effect works with rear-camera Portrait images |
| Available Neural Engine SRAM (Xcode Instruments → Neural Engine template, or sysctl hw.npesram via mobileconfig) |
hw.npesram: 16384 (16MB dedicated SRAM on A15); > 4MB free during segmentation |
Free SRAM < 2MB (background ML tasks starving VisionKit); triggers VKErrorDomain Code=1102 (resource exhaustion) |
After hard reset (Step 04): free SRAM ≥ 8MB; Depth Effect processes in < 150ms end-to-end |
5. Frequently Asked Questions (FAQ)
Why does Depth Effect work on some Portrait photos but not others, even from the same session?
The A15 Neural Engine segments subjects using a person-detection model trained on upper-body poses. If the subject is occluded (>30% coverage), lying down, or smaller than 15% of frame height, VisionKit returns confidence < 0.65 and silently disables Depth Effect to avoid visual artifacts. This is by design. Fix: choose a Portrait where the person stands upright, faces the camera, and occupies 20-60% of frame height. The LiDAR depth map must also have valid returns on the subject—transparent/reflective clothing (sequins, silk) can cause depth holes.
I updated from iOS 17 to 18.1 and Depth Effect broke. Is my Neural Engine damaged?
No. Major iOS upgrades (17→18) replace the Core ML model bundle (/System/Library/Frameworks/VisionKit.framework/Models/PersonSegmentation.mlmodelc) but the per-lock-screen cache in /var/mobile/Library/Caches/com.apple.VisionKit/ is not invalidated by the installer. The stale cache references the old model’s tensor layout, causing silent segmentation failures. Step 03 (delete/recreate lock screen) purges this cache. If you already did that and it persists, run Step 04 (hard reset) to clear the Neural Engine SRAM residue from the old model weights.
Can I use a Mac app like “iMazing” or “PhoneRescue” to clear the VisionKit cache without deleting my lock screen?
Absolutely not. The VisionKit cache directory is inside the mobile user sandbox (/var/mobile/...) and protected by iOS Data Protection class CompleteUntilFirstUserAuthentication. No third-party tool—even with full backup access—can modify files in this domain without the user’s passcode-derived key. Any app claiming to “clean Neural Engine caches” is either lying or attempting to exploit a sandbox escape (which Apple patches rapidly). The only supported method is the lock screen delete/recreate (Step 03) or a full device erase (Settings → General → Transfer or Reset → Erase All Content and Settings)—which is overkill for this issue.
Does the iPhone 13 Pro Max’s 120Hz ProMotion display affect Depth Effect rendering?
ProMotion (adaptive 10-120Hz) only governs the display refresh rate; it has zero involvement in VisionKit segmentation. The Depth Effect parallax animation runs at the display’s current refresh rate, but the subject separation mask is computed once at lock screen assignment (or when Depth Effect is toggled) and cached as a static alpha texture. If the mask is correct, ProMotion makes the parallax smoother; if the mask is wrong (the bug you’re fixing), ProMotion just renders the wrong mask more fluidly. No display hardware issue can cause Depth Effect to fail—this is purely VisionKit/Neural Engine software.
Share your OS version, device model, and exact error code in our independent peer Q&A rooms.
Comments