I need a cross-platform desktop application (Windows 10+ and macOS 12+) that can capture a live webcam feed and deliver professional-quality face swaps with virtually no perceptible delay. The tool must lock on to faces accurately, track expressions, match lighting and color, and render the composite at a stable frame rate suitable for streaming or studio recording. The core pipeline should include high-resolution face detection, landmark tracking, real-time inference with a modern GAN or transformer model, and seamless blending. Feel free to lean on CUDA-accelerated TensorFlow or PyTorch, OpenCV for image I/O, and any efficient post-processing libraries you trust—what matters is rock-solid performance and visual fidelity. I want the interface to be simple: a preview window, a slot to load or capture the target face, quick toggles to enable/disable tracking, and an option to record or pipe the output to a virtual camera device. Please keep the build truly native (Qt, Swift, or similar) so the app looks and feels right on both operating systems. Deliverables • Executables for Windows and macOS • Source code with clear build instructions • Pre-trained model plus scripts or notebooks for fine-tuning new identities • Brief user guide and technical documentation Acceptance will be based on: • Swap latency under 100 ms on an RTX-level GPU • Consistent 30 fps or higher preview without dropped frames • Natural expression transfer and minimal edge artifacts in varied lighting • Compilable source verified on both target platforms