Latest Changes in Palmier Pro: AI Editing, Metal Rendering, and Vector Search

Palmier Pro's latest main branch update introduces AI-driven editing menus, Metal-accelerated video effects, and a vector-based search index, all built with Swift 6.2 and covered by extensive tests.

Palmier Pro is an AI-native video editor for macOS built with Swift 6.2, SwiftUI, and AVFoundation. The latest changes in Palmier Pro (commit ab0ca99…) deliver high-impact updates that tighten the AI-editing workflow, modernize the rendering pipeline with custom Metal kernels, and extend the search system to support both visual and textual embeddings.

AI-Enhanced Editing Menu in TimelineView

The TimelineView now exposes AI-driven actions through new SwiftUI extensions. Users can trigger operations like "Generate Clip" and "Replace Audio" directly from the timeline interface.

Implementation resides in Sources/PalmierPro/Timeline/TimelineView+AIEditMenu.swift. This file registers menu items, forwards requests to the central Editor object, and inserts resulting operations into the undo stack via Editor.commit calls.

// Trigger AI generation from a SwiftUI view
Button("Generate Clip") {
    timelineView.performAIEdits(.generateClip) { result in
        switch result {
        case .success(let newClip):
            print("Created clip \(newClip.id)")
        case .failure(let err):
            print("AI generation failed: \(err)")
        }
    }
}

Metal-Accelerated Rendering Pipeline

The rendering pipeline now utilizes custom Metal kernels for real-time visual effects without CPU overhead. The plugin loader in Plugins/MetalCIKernelPlugin/MetalCIKernelPlugin.swift registers these kernels with Core Image, exposing them as CIKernel objects to Swift.

Key additions include:

• Metal/Levels.metal – Tone-mapping and exposure controls
• Metal/LUTTetra.metal – 3-D LUT application for color grading
• Metal/Glow.metal – Optimized bloom and glow effects

import CoreImage

let ciImage = CIImage(contentsOf: videoFrameURL)!
let lutKernel = try! LUTTetraKernel.load()
let processed = lutKernel.apply(to: ciImage, parameters: ["intensity": 0.85])

Vector-Based Search Index Refactor

The search subsystem has been rearchitected to support visual and text embeddings through a new coordinator pattern. This enables semantic search across video content using machine learning models stored under the models/ directory.

Core components include:

• Sources/PalmierPro/Search/SearchIndexCoordinator.swift – Orchestrates async indexing, updates, and query handling via searchVisual(image:completion:)
• Sources/PalmierPro/Search/SearchIndexConfig.swift – Defines model paths and hyper-parameters for embedding generation

let queryImage = UIImage(named: "example.png")!
SearchIndexCoordinator.shared.searchVisual(image: queryImage) { results in
    for hit in results {
        print("Found clip \(hit.clipId) with similarity \(hit.score)")
    }
}

Tests in Tests/PalmierProTests/Search/VisualSearchTests.swift verify end-to-end indexing, retrieval, and on-the-fly model downloads.

Timeline Snap Engine and Geometry Improvements

The SnapEngine now supports magnetic snapping for clip boundaries and keyframes, improving precision during drag-and-drop edits. Updated geometry calculations live in Sources/PalmierPro/Timeline/TimelineGeometry.swift, while visual snap hints render through Sources/PalmierPro/Timeline/SnapIndicatorOverlay.swift.

Unified Design System with AppTheme

All UI components now strictly reference Sources/PalmierPro/UI/AppTheme.swift for design tokens. This eliminates hard-coded values for spacing, fonts, colors, and radii, enabling consistent theming across the SwiftUI and AppKit hybrid interface.

Text("Hello, Palmier")
    .font(.system(size: AppTheme.FontSize.display))
    .foregroundColor(AppTheme.Text.primary)
    .padding(AppTheme.Spacing.m)
    .background(AppTheme.Background.secondary)
    .cornerRadius(AppTheme.Radius.l)

Additional Architecture Improvements

Drag-and-Drop Fixes – A documented macOS 26 issue with nested SwiftUI onDrop modifiers is resolved in Sources/PalmierPro/MediaPanel/MediaTab/MediaPanelDropArea.swift by enforcing native AppKit drop targets for parent areas while leaf nodes continue using SwiftUI drops.

Audio Synchronization – Faster correlation algorithms via AudioSyncCorrelator, tested in Tests/PalmierProTests/Audio/AudioSyncCorrelatorTests.swift.

Caption Generation – Automated text-to-speech and subtitle generation pipeline in Sources/PalmierPro/Captions/CaptionBuilder.swift.

Project I/O – Round-trip serialization improvements through Sources/PalmierPro/Media/ProjectDocumentIO.swift.

Export Updates – Lottie and XML exporters refreshed in Sources/PalmierPro/Export/LottieExport.swift for newer AVFoundation specifications.

Concurrency Utilities – Added AsyncSemaphore in Sources/PalmierPro/Utilities/AsyncSemaphore.swift for async task throttling.

Summary

• AI Editing: New TimelineView+AIEditMenu.swift provides SwiftUI extensions for AI-driven clip generation and audio replacement, integrated with the undo system
• Metal Rendering: Custom kernels (Levels.metal, LUTTetra.metal, Glow.metal) registered via MetalCIKernelPlugin.swift deliver GPU-accelerated effects
• Vector Search: SearchIndexCoordinator.swift enables visual and text-based semantic search with configurable embedding models
• Timeline UX: Magnetic snapping through SnapEngine and unified design tokens via AppTheme.swift improve editing precision and UI consistency

Frequently Asked Questions

What version of Swift does Palmier Pro require?

The latest changes require Swift 6.2 and target macOS exclusively, utilizing modern concurrency features and SwiftUI/AppKit interoperability.

How does Palmier Pro handle AI-generated content undo operations?

AI edits route through the central Editor object, which commits operations to the undo stack. The TimelineView+AIEditMenu.swift file implements this via performAIEdits(_:completion:), ensuring all AI-generated clips support standard undo/redo functionality.

Can third-party developers extend the Metal rendering pipeline?

Yes. The MetalCIKernelPlugin.swift loader dynamically registers Metal kernels as Core Image filters. Developers can add custom .metal files following the pattern established in Levels.metal or LUTTetra.metal, then wrap them in Swift classes conforming to the kernel protocol defined in Sources/PalmierPro/Rendering/EffectKernels.swift.

Where are the embedding models stored for the search feature?

Configuration in SearchIndexConfig.swift references models located under the models/ directory at the repository root. The system supports on-the-fly downloads and versioned model updates, with end-to-end functionality verified by VisualSearchTests.swift.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}