Media Intelligence

import MediaIntelligence is a new iOS 27 framework that runs two on-device media-analysis engines over photo and video assets you supply by URL. Everything stays on-device — the media never leaves it, and you need no Vision or ML background.

When to Use

Use this skill when you're:

• Building a faces / People view in a photo manager — clustering faces into persistent people (entities) across a library you manage
• Auto-picking a representative thumbnail or building a highlight reel from a video
• Persisting and querying face↔person associations across a large asset collection

For detecting where a face is in a single image (bounding box, landmarks), or for OCR / segmentation, use the Vision framework instead — MediaIntelligence clusters identities across many assets; it does not replace per-image detection.

Example Prompts

• "Group faces into people across my photo library on-device in iOS 27."
• "Cluster faces into persons and persist the index across launches."
• "Find the highlight moments in a video and their intensity."
• "Pick the best representative thumbnail frame for a clip."
• "Fetch all the faces belonging to one person in my library."

What This Skill Provides

• FaceGroupAnalyzer – a persistent, Sendable analyzer backed by a working directory you own; clusters faces into entities (people) across image assets
• Lifecycle & state – insertOrUpdateAssets, update(), deleteAssets/deleteAllAssets/purge, and a State (.ready / .stale / .updating) telling you when to recompute
• Query surface – allEntities, allFaces, allAssetIDs, plus fetchFaces(for:) / fetchFaces(in:) / fetchAssetIDs(for:), all as AsyncSequences; Face is Codable
• VideoAnalyzer – a shared analyzer whose variadic analyze(_:for:) runs typed requests: HighlightAnalysisRequest (notable ranges + intensity levels) and KeyFrameAnalysisRequest (representative-frame timestamp)
• MediaIntelligenceError – LocalizedError cases for working-directory, media-processing, face-grouping, and result-fetching failures

Related

• Vision – per-image face detection, OCR, and segmentation, a different task from identity clustering
• Photo Library – PHPicker / PhotosPicker and the asset access MediaIntelligence consumes
• Music Understanding – the on-device audio analysis counterpart