Free Online RetinaFace Detector — High-Accuracy Face Detection (96.7% mAP)
Easily detect faces in your photos with RetinaFace, one of the most accurate face detection models available — scoring 96.7% mAP on the WIDER FACE benchmark (Easy set). Powered by an ncnn WebAssembly engine running entirely in your browser, it automatically finds all faces in your images and draws bounding boxes with confidence scores. The WASM model file is served from our own server — zero external requests, zero CORS issues, zero data leaving your device. Perfect for checking group photos, preparing images for face-specific editing, verifying face visibility before printing, or counting faces in photos.
Upload one or multiple images by dragging them onto the upload area or clicking to browse your files. Batch upload supported.
Each image is processed automatically by RetinaFace — faces are detected with bounding boxes and confidence scores. The results appear in the list instantly.
Download each image individually, or click Download All to save everything as a single ZIP archive.
Tips for Best Face Detection Results
- RetinaFace works well on front-facing, angled, and profile faces. It scored 96.7% on easy faces, 95.7% on medium, and 90.5% on hard cases in the WIDER FACE benchmark.
- The RetinaFace WASM model (2.8 MB) is served from our own server — no external CDN requests needed. It loads once and runs locally via WebAssembly.
- RetinaFace also detects 5 facial landmarks (eyes, nose, mouth corners) for each face, which can be used for face alignment and further processing.
Frequently Asked Questions — RetinaFace Detection
Is the RetinaFace detection tool really free?
Yes, completely free. No registration, no hidden charges, no watermarks, and no file size limits. RetinaFace runs 100% in your browser via WebAssembly — your images never leave your device.
How does RetinaFace work?
RetinaFace is a state-of-the-art face detection model developed by the InsightFace team. It uses a multi-task learning approach that simultaneously predicts face bounding boxes, 5 facial landmarks, and confidence scores. Our implementation compiles the model through ncnn into WebAssembly, running directly in your browser with no server-side processing required. The WASM engine provides near-native inference performance.
Which browsers support RetinaFace?
RetinaFace via WebAssembly works in all modern browsers including Chrome, Firefox, Edge, Safari, and Opera. It uses WebAssembly SIMD when available for faster performance, with a basic WASM fallback for browsers without SIMD support.
What image formats are supported?
JPG, PNG, WebP, BMP, AVIF, GIF, TIFF, SVG, HEIC and more. The face detection result is saved in your original upload format.
Can RetinaFace recognize who someone is?
No. Our tool uses RetinaFace for face detection only — it finds the location of faces in an image and draws bounding boxes. It does NOT identify or recognize specific individuals (that would require facial recognition with a separate embedding model, which is a different technology entirely).
How accurate is RetinaFace compared to other face detectors?
RetinaFace is among the top face detection models. On the standard WIDER FACE benchmark it scores 96.7% (easy), 95.7% (medium), and 90.5% (hard) — significantly higher than alternatives like MTCNN (91.0/88.8/78.0) or BlazeFace (89.8/87.7/71.8). It excels especially at detecting small, occluded, or challenging faces.
How large of an image can I process?
Our tool handles images up to 50 megapixels (50,000,000 pixels total). If an individual image exceeds this limit, you'll see a warning and can try a smaller version.