What does AI Background Remover do?
Remove image backgrounds with an AI model that runs in your browser. Transparent PNG or solid colour background, full resolution. Photos are never uploaded.
- Price
- Free
- Account
- Not required
- Processing
- Entirely in your browser; your data and files are not uploaded
- Works on
- Any modern browser on desktop, tablet or phone
- Category
- Image Tools
How to use the AI Background Remover
- Drop a photo onto the drop zone or click to choose one (JPG, PNG or WebP, up to 25 MB).
- Choose a model: "Best quality" for any subject, or "Fast" for portraits of people.
- Click Remove background. The first time, the model downloads and is cached by your browser.
- Keep the transparent background or choose a solid colour, compare with the slider, and download a PNG (or a JPG with a colour background).
Worked example
A 4000 x 3000 product photo
The model analyses a copy scaled to 1024 x 768 pixels (the "Best quality" model then resizes it to its 1024 x 1024 input). Its mask is scaled back up to 4000 x 3000 with bilinear interpolation and applied to the original pixels, so the PNG you download is still 4000 x 3000. A pixel the model scores at 50% becomes half transparent, which keeps soft edges soft.
How it works
The photo is analysed by an image segmentation model running in your browser with transformers.js 4.3.0 (loaded from the jsDelivr CDN), on your device's processor with WebAssembly. Best quality uses ORMBG, Open Remove Background (Apache-2.0 license, 8-bit quantized weights of about 42 MB). Fast uses MODNet (Apache-2.0 license, 8-bit quantized weights of about 6.3 MB). The model outputs a foreground probability for each pixel of its low resolution input. That mask is resized to the original size with bilinear interpolation and multiplied into the alpha channel of the original image. For a solid colour, each pixel is blended with the colour by its alpha.
Assumptions
- The model and library are downloaded from Hugging Face and the jsDelivr CDN on first use and cached by your browser. Your photo is not uploaded.
- The mask is an estimate by an AI model and can be wrong.
Frequently asked questions
Are my files uploaded?
No. Images are processed in your browser on your own device and are never sent to our server. The page keeps working if you go offline after it loads.
Which model should I choose?
"Best quality" handles people, products, animals and objects and is the default. It is a larger download and takes longer per photo. "Fast" is small and quick, and is trained for portraits of people; it does poorly on objects.
Why is the first run slow?
The model has to download once (about 42 MB for best quality or 6.3 MB for fast). Your browser caches it, so later runs start without downloading. The model then runs on your processor: in our tests the best quality model took about 9 seconds per photo on a recent desktop computer, and older phones are slower.
Why are some edges wrong?
Fine hair, fur, transparent objects, low contrast between subject and background and busy scenes are hard for any automatic model. Try the other model, or a photo with a plainer background.
Can I use the result commercially?
Both models are released under the Apache-2.0 licence, which allows commercial use. You still need the rights to the photo you edit.
Limitations
- Images up to 25 MB and 8000 pixels per side (about 16.7 megapixels on iPhones and iPads).
- Accuracy varies with fine hair, low contrast and busy scenes.
- The first run downloads the model; an internet connection is needed for that.
- Runs on the processor rather than the graphics chip, so it takes several seconds per photo, more on older phones, which may also run out of memory.
- One image at a time.