Real-time object detection → emoji overlay, 100% on-device
Point your camera at anything and watch matching Apple emojis appear instantly.
Powered by YOLO26 and CoreML on the Apple Neural Engine.
Pet detection | Room scan | Kitchen | Workspace | Debug mode
| Feature | Description |
|---|---|
| 80+ Object Categories | Detects people, animals, vehicles, food, electronics, and more (COCO dataset) |
| 1,000+ Classifications | ImageNet classification per detected object for precise emoji matching |
| Real-Time Performance | Interactive frame rates on the Apple Neural Engine, Kalman-filtered for stability |
| Ultrawide Camera | Maximum field of view to detect more objects simultaneously |
| Privacy First | 100% on-device — no internet, no data collection, no tracking, no ads |
| Debug Mode | Toggle bounding boxes, classification labels, and confidence scores |
| Adjustable Settings | Confidence threshold, emoji scale, label priority, smoothing mode |
| Live Reload | Unload and reload ML models on-the-fly without restarting |
CameraService (AVCaptureSession, ultrawide camera)
| CVPixelBuffer
v
VisionService (YOLO26m detection → per-object YOLO26m-cls classification)
| [DetectionResult]
v
EmojiOverlayService (Kalman filter smoothing, overlap resolution)
| [EmojiOverlay]
v
SwiftUI Overlay (EmojiOverlayView — positioned emoji renders)
All ML inference runs on a dedicated DispatchQueue at .userInteractive QoS. Results publish back to the main actor via Combine for SwiftUI rendering.
VisionEmoji does not collect any data. All processing runs entirely on-device.
See the full Privacy Policy.
MIT © 2026 Aristides Lintzeris
Built with CoreML + SwiftUI
