Private, offline-first face search for your personal photo library.
Visage is a desktop app that scans local photo folders, detects every face, and matches them against a reference selfie — surfacing every photo of a specific person across your entire library. Nothing leaves your machine.
- Import — Point Visage at a folder of photos. It indexes the album and generates thumbnails.
- Reference — Provide one or more selfies as the target face. Visage extracts a 512-dimensional embedding for each.
- Scan — A worker pool processes every photo in parallel: detect faces (RetinaFace), align them (Umeyama transform), and compute embeddings (ArcFace).
- Match — Cosine similarity compares reference embeddings against every detected face. Results appear in under 100ms for 10K faces.
- Review & Export — Browse matches in a grid, adjust the similarity threshold (0.3–0.7), and export confirmed photos to a clean folder.
- Fully offline — No cloud uploads, no network calls, no accounts. Everything runs locally.
- Fast — Multi-threaded worker pool processes ~500 photos in under 2 minutes on CPU, under 30 seconds with GPU acceleration.
- GPU-accelerated — CoreML on Apple Silicon Macs, DirectML on Windows, with automatic CPU fallback.
- Adjustable threshold — Tune match sensitivity in real-time to balance precision and recall.
- Bulk export — Copy all confirmed matches to a destination folder in one click.
- Wide format support — JPEG, PNG, WebP, HEIC, HEIF, TIFF, BMP, and AVIF.
- Cross-platform — macOS (Apple Silicon + Intel) and Windows installers.
| Layer | Technology |
|---|---|
| Shell | Electron, React 19, TypeScript, Vite |
| Styling | Tailwind CSS 4, shadcn/ui |
| State | Zustand |
| ML Runtime | ONNX Runtime (CoreML / DirectML / CPU) |
| Detection | RetinaFace (det_10g.onnx) |
| Embedding | ArcFace (w600k_r50.onnx) — 512-dim vectors |
| Image Processing | sharp |
| Database | better-sqlite3 (WAL mode) |
| Parallelism | worker_threads pool |
| Packaging | electron-builder (DMG, NSIS) |
Grab the latest installer for your platform from the Visage website.
Copyright (c) 2025 Mizan Ali. All rights reserved.
The source code is provided for reference purposes only. No permission is granted to use, copy, modify, merge, publish, distribute, sublicense, or sell copies of this software.