Lifting a subject out of a photo, blurring a background or placing words behind someone all need the app to understand what is in the picture. That understanding happens on your device, and nothing is sent away to make it work.
Visible customer outcome
Advanced processing remains part of a familiar Studio workflow.
On-device scene understanding provides structure for Layer Magic, 3D Pop and subject-aware text. The technical capability exists to serve a clear creative outcome.
Business purpose
Create reusable scene structure for subject-aware composition.
The capability allows the product to blur or remove a background, lift a subject through the frame, place words behind it, and reuse the same creative structure during export and recovery.
Business process
AI-01Receive photoUse the normalized imported image.
→
AI-02DiscoverRequest foreground instances from Apple Vision.
→
AI-03RefineApply the shared matting and edge rules.
→
AI-04IdentifyAssociate stable subject and layer identities.
→
AI-05PersistStore reusable mask and cutout references.
→
AI-06ConsumeUse assets in Layer Magic, 3D Pop and export.
Decision table
Usable subject
Requested outcome
Expected behaviour
Status
Yes
3D Pop
Use the subject cutout and permit overflow into reserved matte clearance.
Supported
Yes
Text behind subject
Compose behind text below the subject and front content above it.
Supported
Yes
Portrait Blur
Keep the subject sharp and process the background plate.
Supported
No
Subject-dependent effect
Do not fabricate a cutout. Explain or safely disable the unavailable outcome.
Fallback
Not required
Original background
Continue through the original-image path without unnecessary segmentation.
Bypass
Business rules
Rules that protect the experience
Subject separation runs on device.
A clear subject is a precondition, not a guaranteed outcome.
Reusable visible AI pixels require persisted asset references.
The live canvas must not own full-resolution source pixels.
Quality attributes
Non-functional expectations
UI work remains responsive during analysis.
AI assets survive document restoration.
Preview and export preserve subject identity and ordering.
Unsupported Simulator inference is treated as an environment constraint.
Acceptance criteria
Given an eligible photo with a usable foreground subject, when subject-aware processing completes, then the document retains recoverable mask or cutout references for that subject.
Given a document with persisted AI assets, when the project is restored, then subject-dependent composition can be reconstructed without relying on a process-global full-resolution image.
Given a photo with no usable subject, when a subject-dependent outcome is requested, then the application provides a safe fallback rather than presenting a false successful result.
Given the same synchronized document and profile, when preview and export render the subject stack, then behind content, subject cutout and front content retain the same semantic order.
Given a look-only change such as a filter or frame colour, when the document updates, then structural AI discovery is not repeated unless the change invalidates the relevant artifact.
Common questions
How do I cut out the subject of a photo on iPhone?
Open the photo in Studio and turn on a subject effect such as 3D Pop or a background mode. The app finds the foreground subject on the device and creates a reusable cutout you can pop out of the frame or place text behind.
Does the AI work without an internet connection?
Yes. Scene understanding and subject separation run locally using Apple Vision, so they work offline and nothing about your photo leaves the device.
What if the photo has no clear subject?
The app tells you rather than inventing a result. Subject-dependent effects explain themselves or switch off safely, and the rest of your composition continues to work.
