Live Camera. Per-Frame. Image.
A complete on-device face swap SDK. 512px resolution across all modes. CoreML acceleration, reusable face embeddings, and drop-in camera views.
What face swap modes does Amigo AI SDK support?
Live camera, per-frame processing for custom pipelines, and static image swap — all at 512x512 on-device.
Live Camera (SwiftUI & UIKit)
Drop-in AmigoLiveCameraView for SwiftUI or AmigoLiveViewController for UIKit. 30fps on iPhone 12, 45fps on iPhone 15 Pro. The SDK handles AVCaptureSession, face detection, CoreML inference, and Metal rendering.
Per-Frame API (WebRTC / Video)
Feed any CVPixelBuffer to processFrame(_:using:) and get a CIImage back synchronously. Designed for WebRTC, video playback, screen recording, or custom camera pipelines. Sustained 30fps on a serial queue.
Static Image Swap
Call swapFace(in:using:) with a UIImage and FaceLatent. Full detection → warp → inference → blend pipeline runs asynchronously. Configurable LipMode for natural results.
How does the on-device processing work?
CoreML inference on the device GPU. Models download from CDN at runtime — not bundled in your app binary.
CoreML + Neural Engine
Optimized CoreML models with Metal rendering. Automatically uses Neural Engine on A14+ devices for maximum throughput. Each frame processed in under 33ms.
CDN Model Delivery
ML models download from CDN on first use via downloadModelsIfNeeded() — free, no license charge. Only emap.bin (1MB) is bundled. Keeps your app binary small.
512px Output Resolution
All modes output at 512x512 pixels. Consistent quality across live camera, per-frame, and image swap on every supported device.
What developer features are included?
Reusable face embeddings, runtime face switching, lip blending control, and background replacement.
FaceLatent: Enroll Once, Swap Forever
enrollFace(from:) returns a 512-dimensional FaceLatent embedding. Cache it, reuse it, switch targets at runtime with zero latency. Also accepts pre-computed server-side embeddings.
Background Replacement
Set session.backgroundImage on AmigoLiveSession to replace the person's background with any image in real-time. Set nil to disable.
LipMode Control
Three lip blending modes: .none (source lips), .outerLips, .innerLips (default, most natural). Fine-tune per use case — photo editing vs live camera vs content creation.
How is user privacy protected?
All processing on-device via CoreML. No images, no face data, no biometric information ever leaves the device.
Zero Data Collection
No images, face data, or biometric information is collected or transmitted. All CoreML inference runs locally on the device GPU. Only the API key validation call touches the network.
GDPR & CCPA Compliant by Architecture
Compliant by design, not just policy. No DPA needed since no personal data is processed server-side. No privacy policy changes required to integrate.
iOS 16+, Swift Package Manager
Requires iOS 16 and Swift 5.9. Distributed via Swift Package Manager. Static library — no dynamic framework overhead. Android support coming soon.