Draw bounding boxes online

No install, no account, no upload — open an image, drag boxes, label them, copy the coordinates. Regionly runs entirely in your browser, so the image never touches a server, and the export is real data: exact pixels plus normalized 0–1 values in versioned JSON.

Three steps

1. Drop, browse, or paste (Ctrl+V) a PNG, JPG, WebP, or GIF into the editor.
2. Press R and drag boxes over your targets — a label prompt appears the moment you release. Same label twice? The boxes group in the export. Need more than boxes: polygons, circles, freeform outlines, arrows, and point markers are right there.
3. Hit Done — copy JSON for code, or the plain-text tab for an AI prompt.

The output

{
  "label": "car",
  "color": "#f0443c",
  "regions": [
    {
      "type": "rect",
      "pixels":     { "x": 412, "y": 288, "width": 684, "height": 252 },
      "normalized": { "x": 0.21458, "y": 0.26667,
                      "width": 0.35625, "height": 0.23333 },
      "boundingBox": { "...": "same rect, explicit" }
    }
  ]
}

Every region carries both coordinate systems and an explicit bounding box, so it drops into vision pipelines, cropping scripts, or prompt contexts without conversion. The schema is versioned and documented; additive changes only.

FAQ

Is there a limit on boxes or images?

No limit on regions. One image at a time in the free editor; save it as a project file and open the next.

Can I edit a box after drawing it?

Yes — select it, drag the corner and edge handles, move it, rename it, recolor it. Undo covers everything.

Does it export YOLO or COCO?

Not natively — the JSON is deliberately simple to convert (both pixel and normalized values are already there). If your whole workflow is dataset training formats, a dedicated labeling platform may fit better; Regionly is built for speed and for AI-prompt use.