CanDoYa
All tools
Saved toolsYour favorites

Add to favorites

Saved in this browser only.
Private browser utility / Social & Creator

Free AI Pose Detection

Runs entirely in your browser - no upload, no sign-up.

Live workspaceLocal processing

Choose a photo to map body landmarks on your device.

Share this tool
pose detection / browser utility
01 / Overview

What is pose detection?

Pose detection estimates key body locations in an image, such as shoulders, elbows, wrists, hips, knees, and ankles. This tool maps 33 landmarks for up to three people, draws a connected skeleton, estimates four joint angles, and exports the result. The image is processed on your device.

02

How to use

  1. 01
    Choose a photo

    Drop in a JPEG, PNG, or WebP up to 20 MB and 16 megapixels, or load the sample photo to explore the workflow.

  2. 02
    Detect poses

    Start detection. The browser worker downloads the public runtime and model on the first run, then maps up to three people locally.

  3. 03
    Review and export

    Adjust landmark visibility, inspect elbow and knee angle estimates, then copy or download JSON and save the annotated PNG.

03

Who it's for

  • Fitness and sports prototypes that need quick visual feedback on joint positions before a full motion-analysis workflow is built.
  • Computer vision students exploring normalized landmarks, skeleton connections, visibility scores, and model-relative depth.
  • Designers and animators who want a fast reference skeleton from a still image without installing a desktop package.
  • Private photo workflows where uploading a client, athlete, patient, or unreleased image to a cloud API is not appropriate.

This detector uses Google's MediaPipe Pose Landmarker Lite model through the browser-only MediaPipe Tasks Vision runtime. It tries the GPU delegate first and falls back to WebAssembly on the CPU when GPU setup fails. Runtime and model files download separately, while the selected image remains inside your browser worker.

The output uses normalized image coordinates: x and y run from 0 to 1 across the image, while z is model-relative depth. The visibility slider only changes which points, connections, and angle estimates are shown. It does not rerun the model or improve uncertain source data.

FAQ

Is my image uploaded to a server?

No. The selected image is decoded, analyzed, annotated, and exported inside your browser. The worker downloads public MediaPipe runtime and model files from external hosts, but those requests do not include your image. CanDoYa does not receive or store the source photo or generated result.

Is this pose detection tool free?

Yes. There is no account, payment, credit limit, or watermark. Your device supplies the computing power. The first detection needs an internet connection to download the public runtime and compact model, and those files may be cached by your browser for later use.

What image limits apply?

The tool accepts one JPEG, PNG, or WebP image up to 20 MB and 16 megapixels. Each image side must be at least 32 pixels. These limits reduce browser memory spikes while retaining enough detail for typical photos, screenshots, and full-body reference images.

How many poses and landmarks can it detect?

The detector looks for up to three people and returns 33 landmarks for each detected pose. The set covers the face outline, shoulders, elbows, wrists, hands, hips, knees, ankles, heels, and feet. Crowded scenes and heavily overlapping people can reduce detection quality.

How accurate are the joint angles?

The elbow and knee angles are geometric estimates from three 2D image landmarks. They are useful for visual inspection and prototyping, but perspective and uncertain or hidden joints can shift them. Do not use these results for diagnosis, rehabilitation decisions, coaching safety, or precise biomechanical measurement.

What does landmark visibility mean?

Visibility is the model's estimate that a landmark is present and not hidden in the image. Raising the threshold removes uncertain points and connected lines from the overlay. A high score is not proof that the anatomical position is exact, and a low score can result from cropping or occlusion.

Does pose detection work without a GPU?

Yes, on browsers that support WebAssembly, workers, image decoding, and canvas export. The worker first tries MediaPipe's GPU delegate, then retries with CPU and WebAssembly if GPU setup is unavailable or fails. CPU detection stays local but may take longer on older devices.

Can I use pose detection offline?

A later run may work offline if the browser has retained the runtime and model files, but offline availability is not guaranteed. Storage cleanup, cache eviction, private browsing, or a runtime update can require another download. The selected image itself never needs to be uploaded.