Polite offline face swaps that fit in a 10 MB installer

What it does

MagicMirror is a desktop app for dragging and dropping photos to swap faces, try hairstyles, or preview outfits. It bundles existing open-source face manipulation engines—FaceFusion, InsightFace, and TinyFace—behind a Tauri-based interface written in TypeScript. All processing happens locally on your CPU; images never leave the device.

The interesting bit

The novelty here is packaging, not algorithms. By compiling the Python backend with Nuitka and wrapping it in a lightweight Tauri frontend, the author turns a typically GPU-heavy workflow into something that runs on standard laptops with a sub-10 MB installer and under 1 GB of models.

Key highlights

• Completely offline; no cloud upload or network dependency
• Runs on standard CPUs without a dedicated GPU
• Installer under 10 MB, model files under 1 GB
• Drag-and-drop interface with no complex settings
• Supports macOS 13+ and Windows 10+

Caveats

• It is fundamentally integration code gluing together existing projects like FaceFusion and InsightFace
• The README prohibits commercial use even though the repository carries an MIT license, which creates an ambiguous legal stance
• Linux is not supported

Verdict

A good fit for casual users who want to experiment with face swaps privately on a laptop. Developers seeking a new algorithm, a server API, or a clear commercial license should look elsewhere.