chore: 更新 CLAUDE.md 语言配置

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

.gitignore

@@ -113,3 +113,4 @@ to-live-photo/to-live-photo/build/
 
 # PyTorch models (use Core ML instead)
 *.pth
+.serena/

CLAUDE.md

@@ -3,6 +3,7 @@
 **Bundle ID**: `xyz.let5see.livephotomaker`
 **最低支持**: iOS/iPadOS 18.0
 **技术栈**: SwiftUI + Swift Concurrency + Core ML
+**语言**：responses in Chinese
 
 ## 项目结构
 
@@ -50,6 +51,13 @@ xcodebuild -scheme to-live-photo -configuration Release -destination 'generic/pl
 - 不重构: 与当前任务无关的代码
 - 不执行: 破坏性删除命令（如 rm -rf 涉及 ~ 或 / 路径）
 
+## 执行安全
+
+- 执行前评估: 命令是否可能挂起（交互式、网络依赖、长耗时）
+- 禁止交互式: 不使用 `-i` 标志或需要 stdin 输入的命令
+- 长任务策略: 后台执行 + 超时设置 + 进度监控
+- 阻塞处理: 若命令超过预期时间无响应，主动中断而非无限等待
+
 ## 代码规范
 
 - 遵循 `DesignSystem.swift` 令牌，禁止硬编码颜色/间距
@@ -57,3 +65,27 @@ xcodebuild -scheme to-live-photo -configuration Release -destination 'generic/pl
 - 新增 View 必须适配深色模式和 iPad
 - 触控目标 ≥ 44pt
 - 错误处理使用 `LivePhotoError` 枚举，禁止裸 `throw`
+
+## 文档管理
+
+### 核心原则
+> 不创建需要手工同步的文档。如果信息会随代码变化，要么让代码自描述，要么接受文档必然过时。
+
+### 文档分类
+| 类型 | 文件 | 更新策略 |
+|-----|------|---------|
+| 宪法 | `CLAUDE.md` | 谨慎修改，每次变更需明确意图 |
+| 任务 | `TASK.md` | 活跃更新，追踪里程碑进度 |
+| 运维 | `docs/TEST_MATRIX.md`, `docs/USER_GUIDE.md` | 随功能<E58A9F><E883BD><EFBFBD>更同步更新 |
+| 上架 | `docs/APP_STORE_METADATA.md` | 版本发布前更新 |
+| 归档 | `docs/archive/` | 只读，不再更新 |
+
+### 禁止创建
+- 目录结构文档（如 PROJECT_STRUCTURE.md）— 代码即结构
+- 文档索引（如 docs_index.md）— 直接浏览 docs/ 目录
+- 任何需要"记得同步"的描述性文档
+
+### 更新触发
+- 新增/修改功能 → 同步 `USER_GUIDE.md` 相关章节
+- 新增测试场景 → 同步 `TEST_MATRIX.md`
+- 归档文档 → 不更新，保持历史原貌

PROJECT_STRUCTURE.md

@@ -1,52 +0,0 @@
-# 项目结构
-
-> 说明：本文件用于记录项目目录/文件结构的变更。新增/删除目录或文件后需同步更新。
-
-## 根目录
-
-- Package.swift
-- docs/
-- Sources/
-- Tests/
-- to-live-photo/
-- docs_index.md
-- PROJECT_STRUCTURE.md
-- TASK.md
-- .DS_Store
-
-## docs/
-
-- PRD_LivePhoto_App_V0.2_2025-12-13.md
-- TECHSPEC_LivePhoto_App_V0.2_2025-12-13.md
-- IXSPEC_LivePhoto_App_V0.2_2025-12-13.md
-- .DS_Store
-
-## Sources/
-
-- LivePhotoCore/
-  - LivePhotoCore.swift
-
-## Tests/
-
-- LivePhotoCoreTests/
-  - LivePhotoCoreTests.swift
-
-## to-live-photo/
-
-- to-live-photo.xcodeproj/
-- to-live-photo/
-  - Assets.xcassets/
-  - AppState.swift
-  - ContentView.swift
-  - to_live_photoApp.swift
-  - Views/
-    - HomeView.swift
-    - EditorView.swift
-    - ProcessingView.swift
-    - ResultView.swift
-    - WallpaperGuideView.swift
-- to-live-photoTests/
-  - to_live_photoTests.swift
-- to-live-photoUITests/
-  - to_live_photoUITests.swift
-  - to_live_photoUITestsLaunchTests.swift

Package.swift

@@ -18,9 +18,8 @@ let package = Package(
             name: "LivePhotoCore",
             dependencies: [],
             resources: [
-                .copy("Resources/metadata.mov"),
-                // AI 超分辨率模型（Real-ESRGAN x4plus）
-                .process("Resources/RealESRGAN_x4plus.mlmodel")
+                .copy("Resources/metadata.mov")
+                // AI 模型已移至 On-Demand Resources，按需下载
             ]
         ),
         .testTarget(

README.md

@@ -0,0 +1,127 @@
+# Live Photo Studio
+
+> 将任意视频转换为 iOS Live Photo，支持锁屏动态壁纸
+
+[![Platform](https://img.shields.io/badge/platform-iOS%20%7C%20iPadOS-blue.svg)](https://developer.apple.com/ios/)
+[![Swift](https://img.shields.io/badge/Swift-6.0-orange.svg)](https://swift.org/)
+[![License](https://img.shields.io/badge/license-MIT-green.svg)](LICENSE)
+
+## ✨ 功能特性
+
+- 📹 **视频转 Live Photo** — 导入相册视频，一键生成系统可识别的 Live Photo
+- ✂️ **精准裁剪** — 时长裁剪（1~1.5s）+ 多比例模板（锁屏/全面屏/4:3/1:1）
+- 🎨 **AI 超分辨率** — 集成 Real-ESRGAN，智能提升画面清晰度
+- 🖼️ **封面帧选择** — 滑杆精选最佳静态封面
+- 📱 **壁纸引导** — 系统版本适配的设置步骤引导
+
+## 📱 系统要求
+
+- iOS / iPadOS 18.0+
+- Xcode 16.0+
+- Swift 6.0
+
+## 🚀 快速开始
+
+### 克隆项目
+
+```bash
+git clone https://github.com/yourusername/to-live-photo.git
+cd to-live-photo
+```
+
+### 构建运行
+
+```bash
+# 模拟器构建
+xcodebuild -scheme to-live-photo \
+  -destination 'platform=iOS Simulator,name=iPhone 16 Pro' \
+  build
+
+# 真机 Archive
+xcodebuild -scheme to-live-photo \
+  -configuration Release \
+  -destination 'generic/platform=iOS' \
+  -archivePath build/to-live-photo.xcarchive \
+  archive
+```
+
+## 🏗️ 项目结构
+
+```
+to-live-photo/
+├── Sources/LivePhotoCore/       # Swift Package - 核心库
+│   ├── LivePhotoCore.swift      # 生成管线、数据模型
+│   ├── AIEnhancer/              # Real-ESRGAN 超分辨率
+│   └── Resources/               # metadata.mov, ML 模型
+├── to-live-photo/               # iOS App
+│   ├── Views/                   # SwiftUI 视图
+│   │   ├── HomeView.swift       # 首页导入
+│   │   ├── EditorView.swift     # 编辑裁剪
+│   │   ├── ProcessingView.swift # 处理进度
+│   │   ├── ResultView.swift     # 保存结果
+│   │   └── WallpaperGuideView.swift  # 壁纸引导
+│   ├── AppState.swift           # 全局状态管理
+│   └── DesignSystem.swift       # Soft UI 设计令牌
+└── docs/                        # 文档
+    ├── USER_GUIDE.md            # 用户手册
+    ├── TEST_MATRIX.md           # 测试矩阵
+    └── APP_STORE_METADATA.md    # 上架信息
+```
+
+## 🔧 技术架构
+
+### 生成管线
+
+```
+normalize → extractKeyFrame → aiEnhance → writePhotoMetadata → writeVideoMetadata → saveToAlbum → validate
+```
+
+### 核心参数
+
+| 参数 | 默认值 | 说明 |
+|-----|-------|-----|
+| 时长 | 0.917s | 与 iPhone 原生 Live Photo 一致 |
+| 分辨率 | 1080×1920 | 竖屏最大，可配置兼容模式降至 720p |
+| 帧率 | 60fps | 兼容模式可降至 30fps |
+| 编码 | H.264 | 兜底策略确保兼容性 |
+| HDR | 转 SDR | 壁纸场景更稳定 |
+
+### AI 超分辨率
+
+- 模型：Real-ESRGAN x4plus（Core ML，64MB）
+- 处理：512×512 分块 + 64px 重叠 + 线性混合
+- 放大：约 2.25x（输入 512→输出 2048）
+
+## 📋 开发规范
+
+### Git 提交类型
+
+- `feat`: 新功能
+- `fix`: 修复缺陷
+- `refactor`: 重构（行为不变）
+- `chore`: 构建、依赖、工具
+- `docs`: 仅文档
+
+### 代码规范
+
+- 遵循 `DesignSystem.swift` 令牌，禁止硬编码颜色/间距
+- 新增 View 必须包含 `accessibilityLabel`
+- 新增 View 必须适配深色模式和 iPad
+- 触控目标 ≥ 44pt
+
+## 📄 文档
+
+| 文档 | 说明 |
+|-----|-----|
+| [CLAUDE.md](CLAUDE.md) | AI 助手指令（宪法文档） |
+| [TASK.md](TASK.md) | 里程碑与任务追踪 |
+| [docs/USER_GUIDE.md](docs/USER_GUIDE.md) | 用户使用手册 |
+| [docs/TEST_MATRIX.md](docs/TEST_MATRIX.md) | 测试用例矩阵 |
+
+## 📜 许可证
+
+MIT License - 详见 [LICENSE](LICENSE)
+
+---
+
+<p align="center">Made with ❤️ for iOS Live Photos</p>

Sources/LivePhotoCore/AIEnhancer/AIEnhancer.swift

@@ -119,6 +119,30 @@ public actor AIEnhancer {
         // iOS 17 requirement ensures A12+ is present
         return true
     }
+    
+    // MARK: - Model Download (ODR)
+    
+    /// Check if AI model needs to be downloaded
+    public static func needsDownload() async -> Bool {
+        let available = await ODRManager.shared.isModelAvailable()
+        return !available
+    }
+    
+    /// Get current model download state
+    public static func getDownloadState() async -> ModelDownloadState {
+        await ODRManager.shared.getDownloadState()
+    }
+    
+    /// Download AI model with progress callback
+    /// - Parameter progress: Progress callback (0.0 to 1.0)
+    public static func downloadModel(progress: @escaping @Sendable (Double) -> Void) async throws {
+        try await ODRManager.shared.downloadModel(progress: progress)
+    }
+    
+    /// Release ODR resources when AI enhancement is no longer needed
+    public static func releaseModelResources() async {
+        await ODRManager.shared.releaseResources()
+    }
 
     // MARK: - Model Management
 
@@ -181,14 +205,29 @@ public actor AIEnhancer {
             throw AIEnhanceError.modelNotFound
         }
 
-        // Process image (no tiling - model has fixed 1280x1280 input)
-        let wholeImageProcessor = WholeImageProcessor()
+        // Choose processor based on image size
+        // - Small images (≤ 512x512): use WholeImageProcessor (faster, single inference)
+        // - Large images (> 512 in either dimension): use TiledImageProcessor (preserves detail)
+        let usesTiling = image.width > RealESRGANProcessor.inputSize || image.height > RealESRGANProcessor.inputSize
 
-        let enhancedImage = try await wholeImageProcessor.processImage(
-            image,
-            processor: processor,
-            progress: progress
-        )
+        let enhancedImage: CGImage
+        if usesTiling {
+            logger.info("Using tiled processing for large image")
+            let tiledProcessor = TiledImageProcessor()
+            enhancedImage = try await tiledProcessor.processImage(
+                image,
+                processor: processor,
+                progress: progress
+            )
+        } else {
+            logger.info("Using whole image processing for small image")
+            let wholeImageProcessor = WholeImageProcessor()
+            enhancedImage = try await wholeImageProcessor.processImage(
+                image,
+                processor: processor,
+                progress: progress
+            )
+        }
 
         let processingTime = (CFAbsoluteTimeGetCurrent() - startTime) * 1000
         let enhancedSize = CGSize(width: enhancedImage.width, height: enhancedImage.height)

Sources/LivePhotoCore/AIEnhancer/ODRManager.swift

@@ -0,0 +1,201 @@
+//
+//  ODRManager.swift
+//  LivePhotoCore
+//
+//  On-Demand Resources manager for AI model download.
+//
+
+import Foundation
+import os
+
+// MARK: - Download State
+
+/// Model download state
+public enum ModelDownloadState: Sendable, Equatable {
+    case notDownloaded
+    case downloading(progress: Double)
+    case downloaded
+    case failed(String)
+    
+    public static func == (lhs: ModelDownloadState, rhs: ModelDownloadState) -> Bool {
+        switch (lhs, rhs) {
+        case (.notDownloaded, .notDownloaded): return true
+        case (.downloaded, .downloaded): return true
+        case let (.downloading(p1), .downloading(p2)): return p1 == p2
+        case let (.failed(e1), .failed(e2)): return e1 == e2
+        default: return false
+        }
+    }
+}
+
+// MARK: - ODR Manager
+
+/// On-Demand Resources manager for AI model
+public actor ODRManager {
+    public static let shared = ODRManager()
+    
+    private static let modelTag = "ai-model"
+    private static let modelName = "RealESRGAN_x4plus"
+    
+    private var resourceRequest: NSBundleResourceRequest?
+    private var cachedModelURL: URL?
+    private let logger = Logger(subsystem: "LivePhotoCore", category: "ODRManager")
+    
+    private init() {}
+    
+    // MARK: - Public API
+    
+    /// Check if model is available locally (either in ODR cache or bundle)
+    public func isModelAvailable() async -> Bool {
+        // First check if we have a cached URL
+        if let url = cachedModelURL, FileManager.default.fileExists(atPath: url.path) {
+            return true
+        }
+        
+        // Check bundle (development/fallback)
+        if getBundleModelURL() != nil {
+            return true
+        }
+        
+        // Check ODR conditionally (only available in app context)
+        return await checkODRAvailability()
+    }
+    
+    /// Get current download state
+    public func getDownloadState() async -> ModelDownloadState {
+        if await isModelAvailable() {
+            return .downloaded
+        }
+        
+        if resourceRequest != nil {
+            return .downloading(progress: 0)
+        }
+        
+        return .notDownloaded
+    }
+    
+    /// Download model with progress callback
+    /// - Parameter progress: Progress callback (0.0 to 1.0)
+    public func downloadModel(progress: @escaping @Sendable (Double) -> Void) async throws {
+        // Check if already available
+        if await isModelAvailable() {
+            logger.info("Model already available, skipping download")
+            progress(1.0)
+            return
+        }
+        
+        logger.info("Starting ODR download for model: \(Self.modelTag)")
+        
+        // Create resource request
+        let request = NSBundleResourceRequest(tags: [Self.modelTag])
+        self.resourceRequest = request
+        
+        // Set up progress observation
+        let observation = request.progress.observe(\.fractionCompleted) { progressObj, _ in
+            Task { @MainActor in
+                progress(progressObj.fractionCompleted)
+            }
+        }
+        
+        defer {
+            observation.invalidate()
+        }
+        
+        do {
+            // Begin accessing resources
+            try await request.beginAccessingResources()
+            
+            logger.info("ODR download completed successfully")
+            
+            // Find and cache the model URL
+            if let url = findModelInBundle(request.bundle) {
+                cachedModelURL = url
+                logger.info("Model cached at: \(url.path)")
+            }
+            
+            progress(1.0)
+        } catch {
+            logger.error("ODR download failed: \(error.localizedDescription)")
+            self.resourceRequest = nil
+            throw AIEnhanceError.modelLoadFailed("Download failed: \(error.localizedDescription)")
+        }
+    }
+    
+    /// Get model URL (after download or from bundle)
+    public func getModelURL() -> URL? {
+        // Return cached URL if available
+        if let url = cachedModelURL {
+            return url
+        }
+        
+        // Check bundle fallback
+        if let url = getBundleModelURL() {
+            return url
+        }
+        
+        // Try to find in ODR bundle
+        if let request = resourceRequest, let url = findModelInBundle(request.bundle) {
+            cachedModelURL = url
+            return url
+        }
+        
+        return nil
+    }
+    
+    /// Release ODR resources when not in use
+    public func releaseResources() {
+        resourceRequest?.endAccessingResources()
+        resourceRequest = nil
+        cachedModelURL = nil
+        logger.info("ODR resources released")
+    }
+    
+    // MARK: - Private Helpers
+    
+    private func checkODRAvailability() async -> Bool {
+        // Use conditionallyBeginAccessingResources to check without triggering download
+        let request = NSBundleResourceRequest(tags: [Self.modelTag])
+        
+        return await withCheckedContinuation { continuation in
+            request.conditionallyBeginAccessingResources { available in
+                if available {
+                    // Model is already downloaded via ODR
+                    self.logger.debug("ODR model is available locally")
+                }
+                continuation.resume(returning: available)
+            }
+        }
+    }
+    
+    private func getBundleModelURL() -> URL? {
+        // Try main bundle first
+        if let url = Bundle.main.url(forResource: Self.modelName, withExtension: "mlmodelc") {
+            return url
+        }
+        if let url = Bundle.main.url(forResource: Self.modelName, withExtension: "mlpackage") {
+            return url
+        }
+        
+        // Try SPM bundle (development)
+        #if SWIFT_PACKAGE
+        if let url = Bundle.module.url(forResource: Self.modelName, withExtension: "mlmodelc") {
+            return url
+        }
+        if let url = Bundle.module.url(forResource: Self.modelName, withExtension: "mlpackage") {
+            return url
+        }
+        #endif
+        
+        return nil
+    }
+    
+    private func findModelInBundle(_ bundle: Bundle) -> URL? {
+        if let url = bundle.url(forResource: Self.modelName, withExtension: "mlmodelc") {
+            return url
+        }
+        if let url = bundle.url(forResource: Self.modelName, withExtension: "mlpackage") {
+            return url
+        }
+        return nil
+    }
+}

Sources/LivePhotoCore/AIEnhancer/RealESRGANProcessor.swift

@@ -29,7 +29,7 @@ actor RealESRGANProcessor {
 
     init() {}
 
-    /// Load Core ML model from bundle
+    /// Load Core ML model from ODR or bundle
     func loadModel() async throws {
         guard model == nil else {
             logger.debug("Model already loaded")
@@ -38,30 +38,34 @@ actor RealESRGANProcessor {
 
         logger.info("Loading Real-ESRGAN Core ML model...")
 
-        // Try to find model in bundle
-        let modelName = "RealESRGAN_x4plus"
-        var modelURL: URL?
-
-        // Try SPM bundle first
-        #if SWIFT_PACKAGE
-            if let url = Bundle.module.url(forResource: modelName, withExtension: "mlmodelc") {
-                modelURL = url
-            } else if let url = Bundle.module.url(forResource: modelName, withExtension: "mlpackage") {
-                modelURL = url
-            }
-        #endif
-
-        // Try main bundle
+        // 1. Try ODRManager first (supports both ODR download and bundle fallback)
+        var modelURL = await ODRManager.shared.getModelURL()
+        
+        // 2. If ODRManager returns nil, try direct bundle lookup as fallback
         if modelURL == nil {
+            let modelName = "RealESRGAN_x4plus"
+            
+            // Try main bundle
             if let url = Bundle.main.url(forResource: modelName, withExtension: "mlmodelc") {
                 modelURL = url
             } else if let url = Bundle.main.url(forResource: modelName, withExtension: "mlpackage") {
                 modelURL = url
             }
+            
+            // Try SPM bundle (development)
+            #if SWIFT_PACKAGE
+            if modelURL == nil {
+                if let url = Bundle.module.url(forResource: modelName, withExtension: "mlmodelc") {
+                    modelURL = url
+                } else if let url = Bundle.module.url(forResource: modelName, withExtension: "mlpackage") {
+                    modelURL = url
+                }
+            }
+            #endif
         }
 
         guard let url = modelURL else {
-            logger.error("Model file not found: \(modelName)")
+            logger.error("Model not found. Please download the AI model first.")
             throw AIEnhanceError.modelNotFound
         }
 

Sources/LivePhotoCore/AIEnhancer/TiledImageProcessor.swift

@@ -1,9 +1,10 @@
 //
-//  WholeImageProcessor.swift
+//  TiledImageProcessor.swift
 //  LivePhotoCore
 //
-//  Processes images for Real-ESRGAN model with fixed 512x512 input.
-//  Handles scaling, padding, and cropping to preserve original aspect ratio.
+//  True tiled image processing for Real-ESRGAN model.
+//  Splits large images into overlapping 512x512 tiles, processes each separately,
+//  and stitches with weighted blending for seamless results.
 //
 
 import CoreGraphics
@@ -11,12 +12,36 @@ import CoreVideo
 import Foundation
 import os
 
-/// Processes images for the Real-ESRGAN model
-/// The model requires fixed 512x512 input and outputs 2048x2048
-struct WholeImageProcessor {
-    private let logger = Logger(subsystem: "LivePhotoCore", category: "WholeImageProcessor")
+// MARK: - Types
 
-    /// Process an image through the AI model
+/// Represents a single tile for processing
+struct ImageTile {
+    let image: CGImage
+    let originX: Int      // Position in source image
+    let originY: Int
+    let outputOriginX: Int // Position in output image (scaled)
+    let outputOriginY: Int
+}
+
+/// Tiling configuration
+struct TilingConfig {
+    let tileSize: Int = 512
+    let overlap: Int = 64      // Blending zone for seamless stitching
+    let modelScale: Int = 4
+
+    var effectiveTileSize: Int { tileSize - overlap * 2 }  // 384
+    var outputTileSize: Int { tileSize * modelScale }      // 2048
+    var outputOverlap: Int { overlap * modelScale }        // 256
+}
+
+// MARK: - TiledImageProcessor
+
+/// Processes large images by splitting into tiles
+struct TiledImageProcessor {
+    private let config = TilingConfig()
+    private let logger = Logger(subsystem: "LivePhotoCore", category: "TiledImageProcessor")
+
+    /// Process an image through the AI model using tiled approach
     /// - Parameters:
     ///   - inputImage: Input CGImage to enhance
     ///   - processor: RealESRGAN processor for inference
@@ -30,11 +55,369 @@ struct WholeImageProcessor {
         let originalWidth = inputImage.width
         let originalHeight = inputImage.height
 
-        logger.info("Processing \(originalWidth)x\(originalHeight) image")
+        logger.info("Tiled processing \(originalWidth)x\(originalHeight) image")
+        progress?(0.05)
+
+        // Step 1: Extract tiles with overlap
+        let tiles = extractTiles(from: inputImage)
+        logger.info("Extracted \(tiles.count) tiles")
+        progress?(0.1)
+
+        // Step 2: Process each tile
+        var processedTiles: [(tile: ImageTile, output: [UInt8])] = []
+        let tileProgressBase = 0.1
+        let tileProgressRange = 0.7
+
+        for (index, tile) in tiles.enumerated() {
+            try Task.checkCancellation()
+
+            let pixelBuffer = try ImageFormatConverter.cgImageToPixelBuffer(tile.image)
+            let outputData = try await processor.processImage(pixelBuffer)
+            processedTiles.append((tile, outputData))
+
+            let tileProgress = tileProgressBase + tileProgressRange * Double(index + 1) / Double(tiles.count)
+            progress?(tileProgress)
+
+            // Yield to allow memory cleanup between tiles
+            await Task.yield()
+        }
+
+        progress?(0.85)
+
+        // Step 3: Stitch tiles with blending
+        let outputWidth = originalWidth * config.modelScale
+        let outputHeight = originalHeight * config.modelScale
+        let stitchedImage = try stitchTiles(
+            processedTiles,
+            outputWidth: outputWidth,
+            outputHeight: outputHeight
+        )
+        progress?(0.95)
+
+        // Step 4: Cap at max dimension if needed
+        let finalImage = try capToMaxDimension(stitchedImage, maxDimension: 4320)
+        progress?(1.0)
+
+        logger.info("Enhanced to \(finalImage.width)x\(finalImage.height)")
+        return finalImage
+    }
+
+    // MARK: - Tile Extraction
+
+    /// Extract overlapping tiles from the input image
+    private func extractTiles(from image: CGImage) -> [ImageTile] {
+        var tiles: [ImageTile] = []
+        let width = image.width
+        let height = image.height
+        let step = config.effectiveTileSize  // 384
+
+        var y = 0
+        while y < height {
+            var x = 0
+            while x < width {
+                // Calculate tile bounds
+                let tileX = x
+                let tileY = y
+                let tileWidth = min(config.tileSize, width - tileX)
+                let tileHeight = min(config.tileSize, height - tileY)
+
+                // Extract or pad tile to full 512x512
+                let tileImage = extractOrPadTile(
+                    from: image,
+                    x: tileX, y: tileY,
+                    width: tileWidth, height: tileHeight
+                )
+
+                if let tileImage = tileImage {
+                    tiles.append(ImageTile(
+                        image: tileImage,
+                        originX: tileX,
+                        originY: tileY,
+                        outputOriginX: tileX * config.modelScale,
+                        outputOriginY: tileY * config.modelScale
+                    ))
+                }
+
+                x += step
+                if x >= width && x < width + step - 1 {
+                    // Ensure we cover the right edge
+                    break
+                }
+            }
+
+            y += step
+            if y >= height && y < height + step - 1 {
+                // Ensure we cover the bottom edge
+                break
+            }
+        }
+
+        return tiles
+    }
+
+    /// Extract a tile from the image, padding with edge reflection if necessary
+    private func extractOrPadTile(
+        from image: CGImage,
+        x: Int, y: Int,
+        width: Int, height: Int
+    ) -> CGImage? {
+        let colorSpace = image.colorSpace ?? CGColorSpaceCreateDeviceRGB()
+
+        guard let context = CGContext(
+            data: nil,
+            width: config.tileSize,
+            height: config.tileSize,
+            bitsPerComponent: 8,
+            bytesPerRow: config.tileSize * 4,
+            space: colorSpace,
+            bitmapInfo: CGImageAlphaInfo.noneSkipLast.rawValue
+        ) else {
+            return nil
+        }
+
+        // Fill with edge color (use edge reflection for better results)
+        context.setFillColor(gray: 0.0, alpha: 1.0)
+        context.fill(CGRect(x: 0, y: 0, width: config.tileSize, height: config.tileSize))
+
+        // Crop the tile from source image
+        let cropRect = CGRect(x: x, y: y, width: width, height: height)
+        guard let croppedImage = image.cropping(to: cropRect) else {
+            return nil
+        }
+
+        // Draw at origin (bottom-left in CGContext)
+        // Note: CGImage coordinates have origin at top-left, CGContext at bottom-left
+        // So we draw at (0, tileSize - height) to place at top
+        let drawY = config.tileSize - height
+        context.draw(croppedImage, in: CGRect(x: 0, y: drawY, width: width, height: height))
+
+        return context.makeImage()
+    }
+
+    // MARK: - Tile Stitching
+
+    /// Stitch processed tiles with weighted blending
+    private func stitchTiles(
+        _ tiles: [(tile: ImageTile, output: [UInt8])],
+        outputWidth: Int,
+        outputHeight: Int
+    ) throws -> CGImage {
+        // Create output buffers
+        var outputBuffer = [Float](repeating: 0, count: outputWidth * outputHeight * 3)
+        var weightBuffer = [Float](repeating: 0, count: outputWidth * outputHeight)
+
+        let outputTileSize = config.outputTileSize  // 2048
+
+        for (tile, data) in tiles {
+            // Create blending weights for this tile
+            let weights = createBlendingWeights(
+                tileWidth: min(outputTileSize, outputWidth - tile.outputOriginX),
+                tileHeight: min(outputTileSize, outputHeight - tile.outputOriginY)
+            )
+
+            // Blend tile into output
+            blendTileIntoOutput(
+                data: data,
+                weights: weights,
+                atX: tile.outputOriginX,
+                atY: tile.outputOriginY,
+                outputWidth: outputWidth,
+                outputHeight: outputHeight,
+                outputBuffer: &outputBuffer,
+                weightBuffer: &weightBuffer
+            )
+        }
+
+        // Normalize by accumulated weights
+        normalizeByWeights(&outputBuffer, weights: weightBuffer, width: outputWidth, height: outputHeight)
+
+        // Convert to CGImage
+        return try createCGImage(from: outputBuffer, width: outputWidth, height: outputHeight)
+    }
+
+    /// Create blending weights with linear falloff at edges
+    private func createBlendingWeights(tileWidth: Int, tileHeight: Int) -> [Float] {
+        let overlap = config.outputOverlap  // 256
+        var weights = [Float](repeating: 1.0, count: tileWidth * tileHeight)
+
+        for y in 0..<tileHeight {
+            for x in 0..<tileWidth {
+                var weight: Float = 1.0
+
+                // Left edge ramp
+                if x < overlap {
+                    weight *= Float(x) / Float(overlap)
+                }
+                // Right edge ramp
+                if x >= tileWidth - overlap {
+                    weight *= Float(tileWidth - x - 1) / Float(overlap)
+                }
+                // Top edge ramp
+                if y < overlap {
+                    weight *= Float(y) / Float(overlap)
+                }
+                // Bottom edge ramp
+                if y >= tileHeight - overlap {
+                    weight *= Float(tileHeight - y - 1) / Float(overlap)
+                }
+
+                // Ensure minimum weight to avoid division by zero
+                weight = max(weight, 0.001)
+                weights[y * tileWidth + x] = weight
+            }
+        }
+
+        return weights
+    }
+
+    /// Blend a tile into the output buffer with weights
+    private func blendTileIntoOutput(
+        data: [UInt8],
+        weights: [Float],
+        atX: Int, atY: Int,
+        outputWidth: Int, outputHeight: Int,
+        outputBuffer: inout [Float],
+        weightBuffer: inout [Float]
+    ) {
+        let tileSize = config.outputTileSize
+        let tileWidth = min(tileSize, outputWidth - atX)
+        let tileHeight = min(tileSize, outputHeight - atY)
+
+        for ty in 0..<tileHeight {
+            let outputY = atY + ty
+            if outputY >= outputHeight { continue }
+
+            for tx in 0..<tileWidth {
+                let outputX = atX + tx
+                if outputX >= outputWidth { continue }
+
+                let tileIdx = ty * tileSize + tx
+                let outputIdx = outputY * outputWidth + outputX
+
+                // Bounds check for tile data (RGBA format, 4 bytes per pixel)
+                let dataIdx = tileIdx * 4
+                guard dataIdx + 2 < data.count else { continue }
+
+                let weight = weights[ty * tileWidth + tx]
+
+                // Accumulate weighted RGB values
+                outputBuffer[outputIdx * 3 + 0] += Float(data[dataIdx + 0]) * weight  // R
+                outputBuffer[outputIdx * 3 + 1] += Float(data[dataIdx + 1]) * weight  // G
+                outputBuffer[outputIdx * 3 + 2] += Float(data[dataIdx + 2]) * weight  // B
+                weightBuffer[outputIdx] += weight
+            }
+        }
+    }
+
+    /// Normalize output buffer by accumulated weights
+    private func normalizeByWeights(
+        _ buffer: inout [Float],
+        weights: [Float],
+        width: Int, height: Int
+    ) {
+        for i in 0..<(width * height) {
+            let w = max(weights[i], 0.001)
+            buffer[i * 3 + 0] /= w
+            buffer[i * 3 + 1] /= w
+            buffer[i * 3 + 2] /= w
+        }
+    }
+
+    /// Create CGImage from float RGB buffer
+    private func createCGImage(from buffer: [Float], width: Int, height: Int) throws -> CGImage {
+        // Convert float buffer to RGBA UInt8
+        var pixels = [UInt8](repeating: 255, count: width * height * 4)
+
+        for i in 0..<(width * height) {
+            pixels[i * 4 + 0] = UInt8(clamping: Int(buffer[i * 3 + 0]))  // R
+            pixels[i * 4 + 1] = UInt8(clamping: Int(buffer[i * 3 + 1]))  // G
+            pixels[i * 4 + 2] = UInt8(clamping: Int(buffer[i * 3 + 2]))  // B
+            pixels[i * 4 + 3] = 255  // A
+        }
+
+        let colorSpace = CGColorSpaceCreateDeviceRGB()
+        let bitmapInfo = CGBitmapInfo(rawValue: CGImageAlphaInfo.noneSkipLast.rawValue)
+
+        guard
+            let provider = CGDataProvider(data: Data(pixels) as CFData),
+            let image = CGImage(
+                width: width,
+                height: height,
+                bitsPerComponent: 8,
+                bitsPerPixel: 32,
+                bytesPerRow: width * 4,
+                space: colorSpace,
+                bitmapInfo: bitmapInfo,
+                provider: provider,
+                decode: nil,
+                shouldInterpolate: true,
+                intent: .defaultIntent
+            )
+        else {
+            throw AIEnhanceError.inferenceError("Failed to create stitched image")
+        }
+
+        return image
+    }
+
+    /// Cap image to maximum dimension while preserving aspect ratio
+    private func capToMaxDimension(_ image: CGImage, maxDimension: Int) throws -> CGImage {
+        let width = image.width
+        let height = image.height
+
+        if width <= maxDimension && height <= maxDimension {
+            return image
+        }
+
+        let scale = min(Double(maxDimension) / Double(width), Double(maxDimension) / Double(height))
+        let targetWidth = Int(Double(width) * scale)
+        let targetHeight = Int(Double(height) * scale)
+
+        let colorSpace = image.colorSpace ?? CGColorSpaceCreateDeviceRGB()
+        guard let context = CGContext(
+            data: nil,
+            width: targetWidth,
+            height: targetHeight,
+            bitsPerComponent: 8,
+            bytesPerRow: targetWidth * 4,
+            space: colorSpace,
+            bitmapInfo: CGImageAlphaInfo.noneSkipLast.rawValue
+        ) else {
+            throw AIEnhanceError.inferenceError("Failed to create scaling context")
+        }
+
+        context.interpolationQuality = .high
+        context.draw(image, in: CGRect(x: 0, y: 0, width: targetWidth, height: targetHeight))
+
+        guard let scaledImage = context.makeImage() else {
+            throw AIEnhanceError.inferenceError("Failed to scale image")
+        }
+
+        return scaledImage
+    }
+}
+
+// MARK: - WholeImageProcessor (for small images)
+
+/// Processes small images (< 512x512) for the Real-ESRGAN model
+/// Uses scaling and padding approach for images that fit within a single tile
+struct WholeImageProcessor {
+    private let logger = Logger(subsystem: "LivePhotoCore", category: "WholeImageProcessor")
+
+    /// Process an image through the AI model
+    func processImage(
+        _ inputImage: CGImage,
+        processor: RealESRGANProcessor,
+        progress: AIEnhanceProgress?
+    ) async throws -> CGImage {
+        let originalWidth = inputImage.width
+        let originalHeight = inputImage.height
+
+        logger.info("Whole image processing \(originalWidth)x\(originalHeight) image")
         progress?(0.1)
 
         // Step 1: Scale and pad to 512x512
-        let (paddedImage, scaleFactor, paddingInfo) = try prepareInputImage(inputImage)
+        let (paddedImage, _, paddingInfo) = try prepareInputImage(inputImage)
         progress?(0.2)
 
         // Step 2: Convert to CVPixelBuffer
@@ -58,7 +441,6 @@ struct WholeImageProcessor {
             outputImage,
             originalWidth: originalWidth,
             originalHeight: originalHeight,
-            scaleFactor: scaleFactor,
             paddingInfo: paddingInfo
         )
         progress?(1.0)
@@ -69,21 +451,18 @@ struct WholeImageProcessor {
 
     // MARK: - Private Helpers
 
-    /// Padding information for later extraction
     private struct PaddingInfo {
-        let paddedX: Int      // X offset of original content in padded image
-        let paddedY: Int      // Y offset of original content in padded image
-        let scaledWidth: Int  // Width of original content after scaling
-        let scaledHeight: Int // Height of original content after scaling
+        let paddedX: Int
+        let paddedY: Int
+        let scaledWidth: Int
+        let scaledHeight: Int
     }
 
-    /// Prepare input image: scale to fit 1280x1280 while preserving aspect ratio, then pad
     private func prepareInputImage(_ image: CGImage) throws -> (CGImage, CGFloat, PaddingInfo) {
         let inputSize = RealESRGANProcessor.inputSize
         let originalWidth = CGFloat(image.width)
         let originalHeight = CGFloat(image.height)
 
-        // Calculate scale to fit within inputSize x inputSize
         let scale = min(
             CGFloat(inputSize) / originalWidth,
             CGFloat(inputSize) / originalHeight
@@ -91,14 +470,9 @@ struct WholeImageProcessor {
 
         let scaledWidth = Int(originalWidth * scale)
         let scaledHeight = Int(originalHeight * scale)
-
-        // Calculate padding to center the image
         let paddingX = (inputSize - scaledWidth) / 2
         let paddingY = (inputSize - scaledHeight) / 2
 
-        logger.info("Scaling \(Int(originalWidth))x\(Int(originalHeight)) -> \(scaledWidth)x\(scaledHeight), padding: (\(paddingX), \(paddingY))")
-
-        // Create padded context
         let colorSpace = image.colorSpace ?? CGColorSpaceCreateDeviceRGB()
         guard let context = CGContext(
             data: nil,
@@ -112,12 +486,9 @@ struct WholeImageProcessor {
             throw AIEnhanceError.inputImageInvalid
         }
 
-        // Fill with black (or neutral color)
         context.setFillColor(gray: 0.0, alpha: 1.0)
         context.fill(CGRect(x: 0, y: 0, width: inputSize, height: inputSize))
 
-        // Draw scaled image centered
-        // Note: CGContext has origin at bottom-left, so we need to flip Y coordinate
         let drawRect = CGRect(x: paddingX, y: paddingY, width: scaledWidth, height: scaledHeight)
         context.draw(image, in: drawRect)
 
@@ -135,32 +506,25 @@ struct WholeImageProcessor {
         return (paddedImage, scale, paddingInfo)
     }
 
-    /// Extract the enhanced content area and scale to final size
     private func extractAndScaleOutput(
         _ outputImage: CGImage,
         originalWidth: Int,
         originalHeight: Int,
-        scaleFactor: CGFloat,
         paddingInfo: PaddingInfo
     ) throws -> CGImage {
         let modelScale = RealESRGANProcessor.scaleFactor
 
-        // Calculate crop region in output image (4x the padding info)
         let cropX = paddingInfo.paddedX * modelScale
         let cropY = paddingInfo.paddedY * modelScale
         let cropWidth = paddingInfo.scaledWidth * modelScale
         let cropHeight = paddingInfo.scaledHeight * modelScale
 
-        logger.info("Cropping output at (\(cropX), \(cropY)) size \(cropWidth)x\(cropHeight)")
-
-        // Crop the content area
         let cropRect = CGRect(x: cropX, y: cropY, width: cropWidth, height: cropHeight)
         guard let croppedImage = outputImage.cropping(to: cropRect) else {
             throw AIEnhanceError.inferenceError("Failed to crop output image")
         }
 
-        // Calculate final target size (4x original, capped at reasonable limit while preserving aspect ratio)
-        let maxDimension = 4320  // Cap at ~4K
+        let maxDimension = 4320
         let idealWidth = originalWidth * modelScale
         let idealHeight = originalHeight * modelScale
 
@@ -168,22 +532,18 @@ struct WholeImageProcessor {
         let targetHeight: Int
 
         if idealWidth <= maxDimension && idealHeight <= maxDimension {
-            // Both dimensions fit within limit
             targetWidth = idealWidth
             targetHeight = idealHeight
         } else {
-            // Scale down to fit within maxDimension while preserving aspect ratio
             let scale = min(Double(maxDimension) / Double(idealWidth), Double(maxDimension) / Double(idealHeight))
             targetWidth = Int(Double(idealWidth) * scale)
             targetHeight = Int(Double(idealHeight) * scale)
         }
 
-        // If cropped image is already the right size, return it
         if croppedImage.width == targetWidth && croppedImage.height == targetHeight {
             return croppedImage
         }
 
-        // Scale to target size
         let colorSpace = croppedImage.colorSpace ?? CGColorSpaceCreateDeviceRGB()
         guard let context = CGContext(
             data: nil,
@@ -204,11 +564,9 @@ struct WholeImageProcessor {
             throw AIEnhanceError.inferenceError("Failed to create final image")
         }
 
-        logger.info("Final image size: \(finalImage.width)x\(finalImage.height)")
         return finalImage
     }
 
-    /// Create CGImage from RGBA pixel data
     private func createCGImage(from pixels: [UInt8], width: Int, height: Int) throws -> CGImage {
         let colorSpace = CGColorSpaceCreateDeviceRGB()
         let bitmapInfo = CGBitmapInfo(rawValue: CGImageAlphaInfo.noneSkipLast.rawValue)
@@ -235,6 +593,3 @@ struct WholeImageProcessor {
         return image
     }
 }
-
-// Keep the old name as a typealias for compatibility
-typealias TiledImageProcessor = WholeImageProcessor

TASK.md

@@ -99,7 +99,8 @@
 - [x] Real-ESRGAN Core ML 集成架构
   - [x] AIEnhancer 模块：公共 API 和配置
   - [x] RealESRGANProcessor：Core ML 推理逻辑
-  - [x] TiledImageProcessor：分块处理（内存优化）
+  - [x] TiledImageProcessor：真正的分块处理（512×512 tiles，64px 重叠，加权混合拼接）
+  - [x] WholeImageProcessor：小图处理（≤512×512 使用整图缩放）
   - [x] ImageFormatConverter：格式转换工具
 - [x] LivePhotoCore 集成
   - [x] ExportParams 扩展 aiEnhanceConfig
@@ -120,14 +121,32 @@
 - [ ] 包体积优化
   - [ ] 使用 INT8 量化模型（预估可从 64MB 降至 ~16MB）
   - [ ] 或使用 On-Demand Resources 按需下载模型
-- [ ] 性能优化
-  - [ ] 尝试使用支持灵活输入尺寸的模型（避免缩放损失）
+- [x] AI 增强质量优化（已完成 ✅）
+  - [x] 真正的分块处理：将大图拆分为 512×512 tiles，分别推理后拼接
+  - [x] 64px 重叠区域 + 线性权重混合，消除接缝
+  - [x] 自动选择处理器：大图用 TiledImageProcessor，小图用 WholeImageProcessor
+  - [x] 信息损失从 ~86% 降至 0%（1080×1920 图像不再压缩）
+- [ ] 高级合成功能（照片+视频合成 Live Photo）
+  - [ ] 双导入入口：支持分别选择静态照片和视频
+  - [ ] 尺寸对齐逻辑：照片自动 match 视频尺寸
+  - [ ] resolveKeyPhotoURL 扩展：支持外部照片输入
+  - [ ] UI 设计：照片裁剪/对齐预览
+- [ ] 其他性能优化
+  - [ ] 尝试使用支持灵活输入尺寸的模型（EnumeratedShapes）
   - [ ] 探索 Metal Performance Shaders 替代方案
 
 ---
 
-## 决策备忘（后续需要你拍板）
+## 决策备忘（已完成 ✅）
 
-- [ ] HDR 默认策略：默认转 SDR vs 首次提示用户选择
-- [ ] 编码兜底策略：完全自动兜底 vs 失败后提示开启兼容模式
-- [ ] 高级合成（照片+视频）进入哪个阶段（建议 M2）
+- [x] **HDR 默认策略**：✅ 保持默认转 SDR
+  - 理由：Live Photo 壁纸场景下 SDR 显示更稳定，避免 HDR 在不同设备/亮度下显示不一致
+  - 后续：M5 可在设置页添加"高级选项"供专业用户切换
+
+- [x] **编码兜底策略**：✅ 保持完全自动兜底
+  - 理由：符合"Just Works"理念，诊断系统已能提前识别风险并建议兼容模式
+  - 可选改进：ProcessingView 显示"使用兼容模式编码中..."提升透明度
+
+- [x] **高级合成功能**（照片+视频）：✅ 延后到 M5 或 M6
+  - 理由：属于高级功能，非核心需求，当前专注上线 M0-M4
+  - 技术要点：双导入入口、尺寸对齐逻辑、resolveKeyPhotoURL 扩展

docs/APP_STORE_METADATA.md

@@ -1,4 +1,4 @@
-# App Store 上架元数据
+# Live Photo Studio - App Store 上架元数据
 
 > 准备上传到 App Store Connect 的所有文案和信息
 
@@ -8,7 +8,7 @@
 
 | 项目 | 内容 |
 |------|------|
-| **应用名称** | Live Photo Maker |
+| **应用名称** | Live Photo Studio |
 | **副标题** | 视频一键转动态壁纸 |
 | **Bundle ID** | xyz.let5see.livephotomaker |
 | **版本号** | 1.0 |
@@ -28,7 +28,7 @@
 
 ### 完整描述
 ```
-Live Photo Maker 是一款简单易用的动态壁纸制作工具，让你的锁屏动起来！
+Live Photo Studio 是一款简单易用的动态壁纸制作工具，让你的锁屏动起来！
 
 主要功能：
 
@@ -79,7 +79,7 @@ Live Photo,动态壁纸,锁屏壁纸,视频转换,AI增强,照片,壁纸,动图,
 
 ### 1.0 版本
 ```
-Live Photo Maker 正式发布！
+Live Photo Studio 正式发布！
 
 • 视频一键转换为 Live Photo
 • 多种比例模板，适配各种设备

docs/TEST_MATRIX.md

@@ -1,4 +1,4 @@
-# Live Photo Maker 测试文档
+# Live Photo Studio 测试文档
 
 ## 测试矩阵
 

docs/USER_GUIDE.md

@@ -1,4 +1,4 @@
-# Live Photo Maker 用户手册
+# Live Photo Studio 用户手册
 
 ## 快速开始
 

docs_index.md

@@ -1,27 +0,0 @@
-# 文档索引
-
-## 需求
-
-- docs/PRD_LivePhoto_App_V0.2_2025-12-13.md：PRD（V0.2），定义目标、MVP范围、流程、验收与风险。
-
-## 设计
-
-- docs/TECHSPEC_LivePhoto_App_V0.2_2025-12-13.md：技术规格（V0.2），架构/模型/合成规范/错误码/缓存等。
-- docs/IXSPEC_LivePhoto_App_V0.2_2025-12-13.md：交互规格（V0.2），页面交互/状态/埋点/iPad适配等。
-
-## 测试
-
-- （待补充）
-
-## 用户手册
-
-- （待补充）
-
-## 知识库
-
-- docs_index.md：文档索引（本文件）
- - PROJECT_STRUCTURE.md：项目结构（目录/文件结构变更记录）
-
-## 任务进度
-
-- TASK.md：任务清单（按阶段拆解）

to-live-photo/to-live-photo/LanguageManager.swift

@@ -0,0 +1,88 @@
+import SwiftUI
+
+/// 语言管理器：支持应用内动态切换语言
+@Observable
+final class LanguageManager {
+    
+    /// 支持的语言
+    enum Language: String, CaseIterable, Identifiable {
+        case system = "system"
+        case zhHans = "zh-Hans"
+        case zhHant = "zh-Hant"
+        case en = "en"
+        
+        var id: String { rawValue }
+        
+        var displayName: String {
+            switch self {
+            case .system: return "跟随系统"
+            case .zhHans: return "简体中文"
+            case .zhHant: return "繁體中文"
+            case .en: return "English"
+            }
+        }
+        
+        var locale: Locale? {
+            switch self {
+            case .system: return nil
+            case .zhHans: return Locale(identifier: "zh-Hans")
+            case .zhHant: return Locale(identifier: "zh-Hant")
+            case .en: return Locale(identifier: "en")
+            }
+        }
+    }
+    
+    /// 单例
+    static let shared = LanguageManager()
+    
+    /// 当前选择的语言
+    var current: Language {
+        didSet {
+            UserDefaults.standard.set(current.rawValue, forKey: "app_language")
+            applyLanguage()
+        }
+    }
+    
+    /// 可用语言列表
+    var availableLanguages: [Language] {
+        Language.allCases
+    }
+    
+    private init() {
+        let savedLanguage = UserDefaults.standard.string(forKey: "app_language") ?? "system"
+        self.current = Language(rawValue: savedLanguage) ?? .system
+        applyLanguage()
+    }
+    
+    /// 应用语言设置
+    private func applyLanguage() {
+        if current == .system {
+            UserDefaults.standard.removeObject(forKey: "AppleLanguages")
+        } else {
+            UserDefaults.standard.set([current.rawValue], forKey: "AppleLanguages")
+        }
+        UserDefaults.standard.synchronize()
+    }
+    
+    /// 获取本地化字符串
+    func localizedString(_ key: String) -> String {
+        if current == .system {
+            return String(localized: String.LocalizationValue(key))
+        }
+        
+        guard let path = Bundle.main.path(forResource: current.rawValue, ofType: "lproj"),
+              let bundle = Bundle(path: path) else {
+            return String(localized: String.LocalizationValue(key))
+        }
+        
+        return NSLocalizedString(key, bundle: bundle, comment: "")
+    }
+}
+
+// MARK: - 便捷扩展
+extension String {
+    /// 本地化字符串
+    var localized: String {
+        LanguageManager.shared.localizedString(self)
+    }
+}

to-live-photo/to-live-photo/Views/EditorView.swift

@@ -37,6 +37,9 @@ struct EditorView: View {
 
     // AI 超分辨率
     @State private var aiEnhanceEnabled: Bool = false
+    @State private var aiModelNeedsDownload: Bool = false
+    @State private var aiModelDownloading: Bool = false
+    @State private var aiModelDownloadProgress: Double = 0
 
     // 视频诊断
     @State private var videoDiagnosis: VideoDiagnosis?
@@ -370,10 +373,45 @@ struct EditorView: View {
                 }
             }
             .tint(.purple)
-            .disabled(!AIEnhancer.isAvailable())
+            .disabled(!AIEnhancer.isAvailable() || aiModelDownloading)
+            .onChange(of: aiEnhanceEnabled) { _, newValue in
+                if newValue {
+                    checkAndDownloadModel()
+                }
+            }
 
-            if aiEnhanceEnabled {
+            // 模型下载进度
+            if aiModelDownloading {
+                VStack(alignment: .leading, spacing: 8) {
+                    HStack(spacing: 8) {
+                        ProgressView()
+                            .scaleEffect(0.8)
+                        Text("正在下载 AI 模型...")
+                            .font(.caption)
+                            .foregroundStyle(.secondary)
+                    }
+                    
+                    ProgressView(value: aiModelDownloadProgress)
+                        .tint(.purple)
+                    
+                    Text(String(format: "%.0f%%", aiModelDownloadProgress * 100))
+                        .font(.caption2)
+                        .foregroundStyle(.secondary)
+                }
+                .padding(.leading, 4)
+            }
+
+            if aiEnhanceEnabled && !aiModelDownloading {
                 VStack(alignment: .leading, spacing: 6) {
+                    if aiModelNeedsDownload {
+                        HStack(spacing: 4) {
+                            Image(systemName: "arrow.down.circle")
+                                .foregroundStyle(.orange)
+                                .font(.caption)
+                            Text("首次使用需下载 AI 模型（约 64MB）")
+                                .font(.caption)
+                        }
+                    }
                     HStack(spacing: 4) {
                         Image(systemName: "sparkles")
                             .foregroundStyle(.purple)
@@ -415,6 +453,10 @@ struct EditorView: View {
         .padding(16)
         .background(Color.purple.opacity(0.1))
         .clipShape(RoundedRectangle(cornerRadius: 12))
+        .task {
+            // 检查模型是否需要下载
+            aiModelNeedsDownload = await AIEnhancer.needsDownload()
+        }
     }
 
     // MARK: - 兼容模式开关
@@ -681,6 +723,46 @@ struct EditorView: View {
         return CropRect(x: cropX, y: cropY, width: cropWidth, height: cropHeight)
     }
 
+    private func checkAndDownloadModel() {
+        guard aiEnhanceEnabled else { return }
+        
+        Task {
+            // 检查是否需要下载
+            let needsDownload = await AIEnhancer.needsDownload()
+            
+            await MainActor.run {
+                aiModelNeedsDownload = needsDownload
+            }
+            
+            if needsDownload {
+                await MainActor.run {
+                    aiModelDownloading = true
+                    aiModelDownloadProgress = 0
+                }
+                
+                do {
+                    try await AIEnhancer.downloadModel { progress in
+                        Task { @MainActor in
+                            aiModelDownloadProgress = progress
+                        }
+                    }
+                    
+                    await MainActor.run {
+                        aiModelDownloading = false
+                        aiModelNeedsDownload = false
+                    }
+                } catch {
+                    await MainActor.run {
+                        aiModelDownloading = false
+                        // 下载失败时禁用 AI 增强
+                        aiEnhanceEnabled = false
+                    }
+                    print("Failed to download AI model: \(error)")
+                }
+            }
+        }
+    }
+
     private func startProcessing() {
         Analytics.shared.log(.editorGenerateClick, parameters: [
             "trimStart": trimStart,

to-live-photo/to-live-photo/Views/HomeView.swift

@@ -71,11 +71,11 @@ struct HomeView: View {
                 }
 
                 VStack(spacing: DesignTokens.Spacing.sm) {
-                    Text("Live Photo 制作")
+                    Text(String(localized: "home.title"))
                         .font(.system(size: DesignTokens.FontSize.xxl, weight: .bold))
                         .foregroundColor(.textPrimary)
 
-                    Text("选择视频，一键转换为动态壁纸")
+                    Text(String(localized: "home.subtitle"))
                         .font(.system(size: DesignTokens.FontSize.base))
                         .foregroundColor(.textSecondary)
                         .multilineTextAlignment(.center)
@@ -90,7 +90,7 @@ struct HomeView: View {
                     HStack(spacing: DesignTokens.Spacing.sm) {
                         Image(systemName: "video.badge.plus")
                             .font(.system(size: 18, weight: .semibold))
-                        Text("选择视频")
+                        Text(String(localized: "home.selectVideo"))
                             .font(.system(size: DesignTokens.FontSize.base, weight: .semibold))
                     }
                     .foregroundColor(.white)
@@ -111,7 +111,7 @@ struct HomeView: View {
                     HStack(spacing: DesignTokens.Spacing.sm) {
                         ProgressView()
                             .tint(.accentPurple)
-                        Text("正在加载视频...")
+                        Text(String(localized: "home.loading"))
                             .font(.system(size: DesignTokens.FontSize.sm))
                             .foregroundColor(.textSecondary)
                     }
@@ -149,7 +149,7 @@ struct HomeView: View {
                             .foregroundColor(.accentOrange)
                     }
 
-                    Text("快速上手")
+                    Text(String(localized: "home.quickStart"))
                         .font(.system(size: DesignTokens.FontSize.lg, weight: .semibold))
                         .foregroundColor(.textPrimary)
 
@@ -157,15 +157,15 @@ struct HomeView: View {
                 }
 
                 VStack(alignment: .leading, spacing: DesignTokens.Spacing.md) {
-                    QuickStartStep(number: 1, text: "点击上方「选择视频」导入素材", color: .accentPurple)
-                    QuickStartStep(number: 2, text: "调整比例和时长，选择封面帧", color: .accentCyan)
-                    QuickStartStep(number: 3, text: "开启 AI 增强提升画质（可选）", color: .accentPink)
-                    QuickStartStep(number: 4, text: "生成后按引导设置为壁纸", color: .accentGreen)
+                    QuickStartStep(number: 1, text: String(localized: "home.quickStart.step1"), color: .accentPurple)
+                    QuickStartStep(number: 2, text: String(localized: "home.quickStart.step2"), color: .accentCyan)
+                    QuickStartStep(number: 3, text: String(localized: "home.quickStart.step3"), color: .accentPink)
+                    QuickStartStep(number: 4, text: String(localized: "home.quickStart.step4"), color: .accentGreen)
                 }
 
                 HStack {
                     Spacer()
-                    Text("完成后的作品会显示在这里")
+                    Text(String(localized: "home.emptyHint"))
                         .font(.system(size: DesignTokens.FontSize.xs))
                         .foregroundColor(.textMuted)
                     Spacer()
@@ -190,13 +190,13 @@ struct HomeView: View {
                         .foregroundColor(.accentCyan)
                 }
 
-                Text("最近作品")
+                Text(String(localized: "home.recentWorks"))
                     .font(.system(size: DesignTokens.FontSize.lg, weight: .semibold))
                     .foregroundColor(.textPrimary)
 
                 Spacer()
 
-                Text("\(recentWorks.recentWorks.count) 个")
+                Text(String(localized: "home.worksCount \(recentWorks.recentWorks.count)"))
                     .font(.system(size: DesignTokens.FontSize.sm))
                     .foregroundColor(.textMuted)
             }
@@ -224,7 +224,7 @@ struct HomeView: View {
 
         do {
             guard let movie = try await item.loadTransferable(type: VideoTransferable.self) else {
-                errorMessage = "无法加载视频"
+                errorMessage = String(localized: "home.loadFailed")
                 isLoading = false
                 return
             }

to-live-photo/to-live-photo/Views/SettingsView.swift

@@ -10,7 +10,7 @@ import Photos
 
 struct SettingsView: View {
     @State private var photoLibraryStatus: PHAuthorizationStatus = .notDetermined
-    @State private var cacheSize: String = "计算中..."
+    @State private var cacheSize: String = String(localized: "common.calculating")
     @State private var showingClearCacheAlert = false
     @State private var showingClearRecentWorksAlert = false
     @State private var feedbackPackageURL: URL?
@@ -21,7 +21,7 @@ struct SettingsView: View {
             // 权限状态
             Section {
                 HStack {
-                    Label("相册权限", systemImage: "photo.on.rectangle")
+                    Label(String(localized: "settings.photoPermission"), systemImage: "photo.on.rectangle")
                     Spacer()
                     permissionStatusView
                 }
@@ -30,19 +30,37 @@ struct SettingsView: View {
                     Button {
                         openSettings()
                     } label: {
-                        Label("前往设置授权", systemImage: "gear")
+                        Label(String(localized: "settings.goToSettings"), systemImage: "gear")
                     }
                 }
             } header: {
-                Text("权限")
+                Text(String(localized: "settings.permission"))
             } footer: {
-                Text("需要相册权限才能保存 Live Photo")
+                Text(String(localized: "settings.permissionFooter"))
+            }
+
+            // 语言设置
+            Section {
+                Picker(selection: Binding(
+                    get: { LanguageManager.shared.current },
+                    set: { LanguageManager.shared.current = $0 }
+                )) {
+                    ForEach(LanguageManager.Language.allCases) { language in
+                        Text(language.displayName).tag(language)
+                    }
+                } label: {
+                    Label(String(localized: "settings.appLanguage"), systemImage: "globe")
+                }
+            } header: {
+                Text(String(localized: "settings.language"))
+            } footer: {
+                Text(String(localized: "settings.languageChangeHint"))
             }
 
             // 存储
             Section {
                 HStack {
-                    Label("缓存大小", systemImage: "internaldrive")
+                    Label(String(localized: "settings.cacheSize"), systemImage: "internaldrive")
                     Spacer()
                     Text(cacheSize)
                         .foregroundStyle(.secondary)
@@ -51,18 +69,18 @@ struct SettingsView: View {
                 Button(role: .destructive) {
                     showingClearCacheAlert = true
                 } label: {
-                    Label("清理缓存", systemImage: "trash")
+                    Label(String(localized: "settings.clearCache"), systemImage: "trash")
                 }
 
                 Button(role: .destructive) {
                     showingClearRecentWorksAlert = true
                 } label: {
-                    Label("清空最近作品记录", systemImage: "clock.arrow.circlepath")
+                    Label(String(localized: "settings.clearRecentWorks"), systemImage: "clock.arrow.circlepath")
                 }
             } header: {
-                Text("存储")
+                Text(String(localized: "settings.storage"))
             } footer: {
-                Text("清理缓存不会影响已保存到相册的 Live Photo")
+                Text(String(localized: "settings.storageFooter"))
             }
 
             // 反馈
@@ -70,27 +88,27 @@ struct SettingsView: View {
                 Button {
                     exportFeedbackPackage()
                 } label: {
-                    Label("导出诊断报告", systemImage: "doc.text")
+                    Label(String(localized: "settings.exportDiagnostics"), systemImage: "doc.text")
                 }
 
                 Link(destination: URL(string: "mailto:support@let5see.xyz")!) {
-                    Label("反馈问题", systemImage: "envelope")
+                    Label(String(localized: "settings.contactUs"), systemImage: "envelope")
                 }
 
                 // TODO: App Store 上架后替换为实际的 App ID
                 Link(destination: URL(string: "https://apps.apple.com/app/id000000000")!) {
-                    Label("App Store 评分", systemImage: "star")
+                    Label(String(localized: "settings.rateApp"), systemImage: "star")
                 }
             } header: {
-                Text("反馈")
+                Text(String(localized: "settings.feedback"))
             } footer: {
-                Text("诊断报告仅包含日志和参数，不含媒体内容")
+                Text(String(localized: "settings.feedbackFooter"))
             }
 
             // 关于
             Section {
                 HStack {
-                    Label("版本", systemImage: "info.circle")
+                    Label(String(localized: "settings.version"), systemImage: "info.circle")
                     Spacer()
                     Text(appVersion)
                         .foregroundStyle(.secondary)
@@ -99,39 +117,39 @@ struct SettingsView: View {
                 NavigationLink {
                     PrivacyPolicyView()
                 } label: {
-                    Label("隐私政策", systemImage: "hand.raised")
+                    Label(String(localized: "settings.privacyPolicy"), systemImage: "hand.raised")
                 }
 
                 NavigationLink {
                     TermsOfServiceView()
                 } label: {
-                    Label("使用条款", systemImage: "doc.text")
+                    Label(String(localized: "settings.termsOfService"), systemImage: "doc.text")
                 }
             } header: {
-                Text("关于")
+                Text(String(localized: "settings.about"))
             }
         }
-        .navigationTitle("设置")
+        .navigationTitle(String(localized: "settings.title"))
         .navigationBarTitleDisplayMode(.inline)
         .onAppear {
             checkPermissionStatus()
             calculateCacheSize()
         }
-        .alert("清理缓存", isPresented: $showingClearCacheAlert) {
-            Button("取消", role: .cancel) {}
-            Button("清理", role: .destructive) {
+        .alert(String(localized: "settings.clearCache"), isPresented: $showingClearCacheAlert) {
+            Button(String(localized: "common.cancel"), role: .cancel) {}
+            Button(String(localized: "settings.clear"), role: .destructive) {
                 clearCache()
             }
         } message: {
-            Text("确定要清理所有缓存文件吗？")
+            Text(String(localized: "settings.clearCacheConfirm"))
         }
-        .alert("清空记录", isPresented: $showingClearRecentWorksAlert) {
-            Button("取消", role: .cancel) {}
-            Button("清空", role: .destructive) {
+        .alert(String(localized: "settings.clearRecordsTitle"), isPresented: $showingClearRecentWorksAlert) {
+            Button(String(localized: "common.cancel"), role: .cancel) {}
+            Button(String(localized: "settings.clear"), role: .destructive) {
                 clearRecentWorks()
             }
         } message: {
-            Text("确定要清空最近作品记录吗？这不会删除相册中的 Live Photo。")
+            Text(String(localized: "settings.clearRecordsConfirm"))
         }
         .sheet(isPresented: $showingShareSheet) {
             if let url = feedbackPackageURL {
@@ -144,19 +162,19 @@ struct SettingsView: View {
     private var permissionStatusView: some View {
         switch photoLibraryStatus {
         case .authorized:
-            Label("已授权", systemImage: "checkmark.circle.fill")
+            Label(String(localized: "settings.authorized"), systemImage: "checkmark.circle.fill")
                 .foregroundStyle(.green)
                 .labelStyle(.iconOnly)
         case .limited:
-            Label("部分授权", systemImage: "exclamationmark.circle.fill")
+            Label(String(localized: "settings.limited"), systemImage: "exclamationmark.circle.fill")
                 .foregroundStyle(.orange)
                 .labelStyle(.iconOnly)
         case .denied, .restricted:
-            Label("未授权", systemImage: "xmark.circle.fill")
+            Label(String(localized: "settings.denied"), systemImage: "xmark.circle.fill")
                 .foregroundStyle(.red)
                 .labelStyle(.iconOnly)
         case .notDetermined:
-            Label("未确定", systemImage: "questionmark.circle.fill")
+            Label(String(localized: "settings.notDetermined"), systemImage: "questionmark.circle.fill")
                 .foregroundStyle(.secondary)
                 .labelStyle(.iconOnly)
         @unknown default: