Always Free · No Signup

See what's in your photo

Bounding boxes, labels and confidence scores for 80 everyday object classes. Adjust the threshold, hide classes you don't care about, and export the annotated image.

Runs entirely in your browser. Your files never leave your device.

The first run takes a few seconds to get ready. Your browser remembers it, so later runs start instantly.

How to detect objects in an image

  1. 1

    Upload a photo

    Everything stays on your device. The first run takes a few seconds to get ready; after that it starts instantly.

  2. 2

    Tune the confidence

    Start at 50%. Raise it to keep only the confident matches, lower it to surface faint ones. Filtering is instant — nothing is detected again.

  3. 3

    Hide classes and export

    Tap a class chip to remove that class from the overlay, then download a full-resolution PNG with the boxes burned in.

Object detection questions

It knows 80 everyday classes: people, cats, dogs, cars, bicycles, traffic lights, bottles, laptops, chairs, food items and so on. Anything outside those 80 is either ignored or forced into the nearest one — a fox will usually come back as a dog. It is not a general-purpose recogniser and cannot identify brands, plant species or individual people.

This tool is deliberately built small so it starts quickly, and the trade-off lands exactly on small, distant and heavily overlapping objects: a crowd scene will get the foreground people and skip the ones further back. Cropping into the region you care about and re-running usually finds them, because the object then occupies far more pixels.

No. It detects that a person is present and draws a box around them — "person" is one of the 80 everyday classes it knows. It has no concept of identity: no face matching, no database, no name attached to a box. If you need face detection or recognition, this is the wrong tool, and deliberately so.

Every box carries a score between 0 and 1 for how certain the match is. Detection runs once at a very low threshold and everything found is kept, so the slider only filters what is already there and redraws — it never starts over. Around 0.5 is a sensible default; below 0.2 you will start seeing boxes that are essentially guesses.

The first run has to get ready, which usually takes a few seconds on a good connection. Every run after that is remembered and finishes in well under a second on a capable machine, or a few seconds on a slower one.

Not yet — the export is a PNG with the boxes drawn on top, not a coordinate file. If you need annotation files for a training set, use a dedicated labelling tool. This one is built for quickly seeing and showing what is in an image, and its box edges are approximate enough that they should not be used as ground truth.

More free tools

Background Remover

Upload a photo and get a clean transparent PNG in seconds. No signup, no watermark, no quality loss.

Try Now

AI Image Upscaler

Enlarge any photo 4× with AI. Rebuilds edges and texture instead of just stretching pixels.

Enhance

Image to Text (OCR)

Extract printed text from screenshots, scans, and photos. Copy as plain text or download as a file.

Extract

Depth Map Generator

Turn any photo into a depth map. Useful for design, 3D parallax effects, and creative compositing.

Generate
Beyond Detection

Change What's in the Frame

Detection tells you what is in a photo. Changing it is a different job — upload the same image in the studio and say what should be removed, replaced, or added, in plain words.

  • Remove or replace an object by describing it
  • Add something that was never in the shot
  • Generate a clean version of the same scene
Explore Studio Features
A four panel comic in which the same dog is added, moved, and removed between frames