KitJade
Images & DesignRuns in your browserNew

AI Background Remover

Erase a photo background locally with an AI model.

Use the AI Background Remover tool

Runs in your browserUpdated Report a problem

What this ai background remover does

AI Background Remover sends an image through the MODNet segmentation model and returns a transparent PNG. The model runs in this tab with WebAssembly, so the photo itself is never uploaded to a server and no account is required.

The first run downloads the quantised model (about 7 MB) from Hugging Face and the WebAssembly runtime from jsDelivr; the browser caches both for later visits. MODNet is tuned for people and a clear foreground subject, so hair, fur and very fine edges can still need manual cleanup.

Why use it

  • Make a transparent product or profile photo without opening a design app.
  • Prepare an image for a slide, poster or listing where the original background distracts.
  • Keep a personal photo on your own device instead of uploading it to a removal service.

How to use it

  1. 1Drop a PNG, JPEG or WebP image up to 15 MB.
  2. 2Choose Remove background and wait for the model to load on the first run.
  3. 3Check the cut-out edges on the checkerboard preview.
  4. 4Download the transparent PNG.

Worked examples

Portrait for a profile

Input

A 1200 × 1600 JPEG of one person against a wall

Output

A transparent PNG with the person retained and the wall removed.

Product cut-out

Input

A well-lit object on a plain background

Output

A transparent PNG that can be placed on a new background.

Common mistakes

  • Expecting a perfect result on busy scenes with several subjects; a model trained for foreground matting can drop small objects.
  • Testing a very large image on a low-memory phone and assuming the delay is a fault; the model runs on the device.
  • Uploading a photo with sensitive content to a random website; this tool keeps the pixels in the tab.
  • Forgetting that a transparent PNG may look white in an app that does not display alpha; check it on a dark background.

Frequently asked questions

Is my image uploaded?
No. The image is decoded and processed in this browser tab. The first run downloads the model and runtime, but your photo is not part of that request.
Why is the first run slower?
The browser downloads the model (about 7 MB) and a WebAssembly runtime, then compiles them. Both are cached, so later runs skip most of that work. Inference time still depends on the device.
What kind of images work best?
People and clear foreground subjects with a reasonably distinct background. MODNet is a portrait-matting model, so complex scenes, transparent objects and very fine detail may need retouching.
Does it add a watermark?
No. The downloaded PNG is the full-resolution cut-out with no watermark and no account requirement.
Can I use the result commercially?
MODNet and the runtime are open-source Apache-2.0/MIT components. You are responsible for the rights to the original photo and for checking that your intended use is permitted.

Related tools

All tools

Passport Photo Maker

Crop to 35×45 mm, 2×2 in and more, then print a 6×4 sheet.

Runs locallyNew

Image Compressor

Shrink JPG, PNG and WebP with a live size comparison.

Runs locally

Image Resizer

Resize to exact pixels or percentage with aspect control.

Runs locally

Image Metadata Remover

Remove EXIF and comments from a JPEG.

Runs locallyNew