Last Updated: September 24, 2026

AI Background Remover

Instantly remove the background from portraits and photos of people, in your photo's full original resolution. Objects and products may come out with rougher edges

Free & unlimited • No sign-up Runs in your browser: what you enter stays on your device

Click to upload or drag and drop your image here

The AI model loads once per visit; the first image may take a little longer to process

PNG JPG WEBP up to 25 MB

1 Upload a PNG, JPG, or WEBP photo, up to 25 MB

2 The AI model loads in your browser (only once) and removes the background

3 Download the result as a transparent PNG, ready to use

Private by design: The AI model runs entirely inside your device's browser using WebAssembly. Your images are never uploaded to our servers.

Share this tool

Why local processing matters

Most free background removal tools limit your exports to tiny resolutions (like 0.25 MP) because they process your images on expensive cloud GPUs and want you to pay for full size.

How we do it for free: the AI runs right in your browser. Using your own device’s hardware via WebAssembly, the tool processes and exports your image at its full native resolution with clean PNG transparency, with no server time to pay for.

How to use this tool

Click the upload area or drag a PNG, JPG or WEBP photo (up to 25 MB) into it.
Wait while the AI model loads in your browser; this happens once per visit, so the first image takes a little longer.
Check the preview, then download the result as a transparent PNG.
Use “Try another image” to process a different photo.

Worked example

Take a portrait of a person against a cluttered room, upload it and download the transparent PNG. Because the output keeps your photo's original resolution, you can place the cut-out on a plain color or a new scene in any editor, for example to make an ID-style or profile picture. Hair and loose clothing edges are the hardest parts, so zoom in on them. Products and objects may come out with rougher edges than people, and very busy backgrounds or subjects of a similar color can leave small leftovers.

Methodology

This tool uses MODNet (Apache-2.0), an open portrait-matting neural network, run with Transformers.js and ONNX Runtime Web compiled to WebAssembly. Everything runs on your own device, and no image data ever leaves it.

On-Device AI: The neural network and its runtime are downloaded once (roughly 50 MB) and cached locally, so every subsequent image you process runs instantly without a new download. MODNet comes from the paper "MODNet: Real-Time Trimap-Free Portrait Matting via Objective Decomposition" (Ke et al., AAAI 2022). Instead of a hand-drawn trimap, it predicts a soft alpha matte, a value from 0 (background) to 1 (person) for every pixel, so hair and semi-transparent edges can keep partial opacity.

Full Resolution, No Limits: Because there is no server-side processing cost to us, there is no incentive to downscale your image: the matte is computed on a reduced-size copy inside the model and then scaled back up, and the output keeps your photo's original pixel dimensions with a true alpha transparency channel. Limits: the model was trained for portraits, so it works best on photos of people with a clear subject. Very fine detail such as single hair strands, glass, smoke, or a subject whose colors match the background can come out soft or incomplete, so check the edges against the preview before using the result.

When a cut-out goes wrong, and the habits behind it

The model does one job well, separating a person from their surroundings. Most disappointing cut-outs come from asking it for something else.

Feeding it a product, a pet or a plant. The model is a portrait matting network, so objects can keep scraps of background or lose edge detail. Use people photos here and expect to retouch anything else.
Shooting dark hair against a dark wall. The model predicts a soft 0-to-1 opacity per pixel; when subject and background share a color, that value hovers in the middle and the edge looks smudged. A lighter or clearly different backdrop helps more than a bigger photo.
Re-saving the result as a JPG. The download is always a PNG with a real alpha channel. JPG has no transparency, so saving it that way fills the cut-out area with a solid color again. Keep the PNG, or place it on a new background in an editor.
Uploading the wrong kind of file. Only PNG, JPG and WEBP up to 25 MB are accepted. A phone format such as HEIC, or a very large export, is refused until you convert or shrink it.
Judging the cut at thumbnail size. The checkerboard preview hides halos. Zoom in on hair and loose clothing before you download, and test the PNG on both a light and a dark background.

Frequently asked questions

Is my photo uploaded to any server? ▼

No. The entire background removal process happens locally inside your browser using WebAssembly. Your image is never sent anywhere — not to SoSimple.tools, not to any third party. Only the AI model files themselves are downloaded to your browser from Hugging Face's servers, which can see a standard download request from your browser.

Why does the first image take longer to process? ▼

The first time you use the tool in a browsing session, it downloads the AI model (roughly 50 MB) so it can run entirely on your device. Your browser caches this model, so every image after the first one processes much faster.

Why is my download a PNG even though I uploaded a JPG? ▼

JPG has no transparency, so a cut-out photo cannot be saved in that format without a solid background. The tool exports a PNG with a real alpha channel, which keeps the removed area transparent. It accepts PNG, JPG and WEBP images up to 25 MB.

Why are the edges rough when I cut out a product or an object? ▼

MODNet is a portrait-matting model, trained to separate people from their surroundings, so it is not tuned for products, pets or objects. On those subjects it may keep bits of background or eat into the subject. For the best result, use a photo where the person is well lit and in focus, with a background that differs in color from clothes and hair, and zoom in on the preview before downloading the PNG.