Images · Free tool
Background Remover
The background remover cuts the main subject, such as a product or a person, out of a photo and gives you a transparent PNG. It uses U²-NetP, a small open-source AI model, which runs entirely in your browser: your photo is never uploaded. Open a photo, press Remove background, refine the edges if needed and download the result, with a transparent or solid-colour background.
Free, no sign-up Runs in your browser
Background Remover
The first time, your browser downloads about 19 MB: the AI model (U²-NetP) and the software that runs it. Your photo is processed on this device and is never uploaded.
Press “Remove background” to see the result here.
Refine the cut-out
Higher removes more of the faint and uncertain areas; lower keeps more of them.
100 keeps the model's soft edges; 0 makes a hard cut-out with no partly transparent pixels.
Blurs the edge slightly to hide jagged outlines.
Uses U²-NetP, the small version of the U²-Net salient object detection model by Xuebin Qin and colleagues (Apache License 2.0; licence), run with ONNX Runtime Web. Quality is moderate: hair, fur, fine edges, glass and subjects that blend into the background are hard for it.
Your photo is processed by an AI model running in your browser. It is never uploaded to our server or to anyone else.
How to use the Background Remover
- 1 Open a JPEG, PNG or WebP photo with one clear subject.
- 2 Press Remove background. The first time, your browser downloads about 19 MB (the model and the software that runs it).
- 3 Compare the original and the result on the checkerboard, which shows the transparent areas.
- 4 Adjust the mask threshold, edge softness and feather if the edges need work, and choose a transparent or solid background.
- 5 Download the PNG.
What it does
- Runs on your device with the U²-NetP model and ONNX Runtime Web: nothing is uploaded
- Transparent PNG at the photo's full size where your browser allows it
- Threshold, softness and feather controls to clean up the edges
- Optional solid background colour
- Side-by-side before and after preview on a checkerboard
Useful for small shops photographing products, people making social media posts and flyers and anyone who needs a quick cut-out without uploading a photo.
Limitations
- Quality is moderate. It works best with one clear subject against a contrasting background; hair, fur, fine edges, glass and subjects that blend into the background are hard for it.
- The model looks at a 320 × 320 pixel copy of your photo, so the edges of the cut-out are smooth rather than precise to the pixel.
- The first use downloads about 19 MB, which can take a while on a slow or metered connection.
- Photos are limited to 25 MB and 25 megapixels, and older phones may run out of memory on large photos.
How it works
- Your photo is read in the browser. Nothing is uploaded. The tool makes a 320 × 320 pixel copy, the input size the model uses.
- The model predicts the subject. U²-NetP is a salient object detection model: for every pixel it estimates how likely it is to belong to the most prominent object in the picture. The result is a grey-scale mask, white for the subject and black for the background.
- The mask becomes transparency. The mask is scaled back up to the size of your photo and used as its alpha channel, so the background becomes transparent and the edges fade smoothly.
The model and its runtime are downloaded only when you press Remove background, and run on your device’s processor with WebAssembly.
Getting a good result
- One clear subject: a product on a table, a person against a wall. Group shots and busy scenes confuse the model.
- Contrast: a dark object on a light background (or the reverse) separates best. A white cup on a white tablecloth is hard.
- Sharp, well-lit photos: blur and deep shadows blur the edge of the mask too.
- Fill the frame: the model sees a 320 × 320 version, so a small subject in a large photo gets few pixels.
Refining the edges
| Control | What it does | When to change it |
|---|---|---|
| Mask threshold | Moves the line between kept and removed | Raise it to remove a halo of background; lower it if parts of the subject disappear |
| Edge softness | How gradually the edge fades | Lower it for a crisp cut-out (logos, products); keep it high for soft subjects |
| Edge feather | Blurs the edge slightly | Add a little to hide jagged outlines |
Check the result on the checkerboard: grey and white squares show through wherever the image is transparent. Then download the PNG, which keeps the transparency. JPEG cannot store transparency, so if you need a JPEG, choose a solid background first or convert the PNG with our image compressor.
For more control, such as cropping or adjusting brightness after the cut-out, use the photo editor. If you need consistent, professional product imagery or graphics for a campaign, our branding and design team can help.
Frequently asked questions
Is my photo uploaded anywhere?
No. The AI model is downloaded to your browser and runs on your device. Your photo is read, processed and saved locally; it is never sent to our server or to a third party.
Why does it download 19 MB the first time?
To run on your device, your browser needs the AI model (about 4.6 MB) and ONNX Runtime, the software that runs it (about 14 MB). They are downloaded only when you press Remove background, and your browser can keep a copy so later uses start faster.
Which AI model does it use?
U²-NetP, the small version of U²-Net, a salient object detection model published by Xuebin Qin and colleagues in Pattern Recognition (2020) and released under the Apache License 2.0. It predicts which parts of the photo belong to the most prominent object, and that prediction becomes the transparency of the PNG.
Why are hair and edges not perfect?
The model works on a 320 × 320 pixel version of your photo, and the small model trades detail for speed and size. Raise the mask threshold to remove a halo of background, lower the edge softness for a crisper cut-out, or add a little feather to hide jagged outlines.
Can I put the cut-out on a new background?
Yes. Choose Solid colour to fill the background before you download, or keep it transparent and place the PNG on any background in your design software, a presentation or a social media post.
Does it work on a phone?
It runs in current mobile browsers that support WebAssembly. Processing large photos needs a lot of memory, so if your phone struggles, close other apps or use a smaller photo.
Sources
- Qin et al. (2020), U²-Net: Going Deeper with Nested U-Structure for Salient Object Detection, Pattern Recognition (arXiv:2005.09007)
- U²-Net source code and pre-trained models (GitHub, Apache License 2.0)
- rembg: the ONNX export of U²-NetP and its pre- and post-processing
- ONNX Runtime: add machine learning to your web application (ONNX Runtime Web)
Phoot Technologies
Need product photos and graphics that sell?
Our branding and design team prepares product images, social media graphics and brand assets for businesses in Ghana and beyond.
Related tools
Photo Editor
Crop, rotate, resize and brighten a photo in your browser, then download it as JPG, PNG or WebP.
Images · Runs in your browser
Image Compressor
Make JPG, PNG and WebP images smaller, resize them or convert between formats, without uploading them.
Images · Runs in your browser
Social Media Image Resizer
Crop and resize one image to each platform's profile, cover, post and advert sizes, with the source of every size.
Images · Runs in your browser