Label images for object detection in minutes, not days
Load your model and AutoLabel draws the boxes. You check them, fix the few it gets wrong, and export labels in YOLO format, the one-text-file-per-image layout most detector training tools read.
The demo loads ten free photos and a small public model (YOLOX-Nano, 3.7 MB), then labels them in front of you.
What it does
Private by design
Your images, models and labels stay in your browser. Nothing is uploaded.
Auto-label with your model
Load a YOLOv5, YOLOv8, YOLO11 or YOLOX model saved as ONNX (a standard file format for trained models) and it draws the boxes.
Fix boxes fast
Draw, move and resize boxes. Select a group to rotate, flip, or paste it onto every image.
Label video too
Turn an MP4, WebM or MOV clip into still frames and label them like photos.
Review as you go
Mark each image checked or flagged. Every change can be undone.
Clean YOLO export
One text file per image plus classes.txt, checked for missing classes before download.
What AutoLabel saves
Only a few numbers, so your progress shows up next time: how many images and boxes you labeled, boxes per class, time spent and the model's file name. They're kept under an anonymous session, with no account, and you can delete them any time with Clear my saved stats. Your images, models and label files never leave this device. Privacy details
Best on a computer
Labeling is precise work, so AutoLabel is built for a mouse and keyboard. On a phone you can still try the demo, draw and fix boxes with a finger, and export. Keyboard shortcuts need a keyboard, and big jobs are easier on a bigger screen.