How-to guides
How to Remove the Background From an Image With AI (In Your Browser)
The background remover runs a neural network in your browser. That is why your photo never leaves your device - and why the first run is slow.
Zahid Sheeraz4 min readRemoving the background from a photo used to mean one of two things: opening Photoshop and spending an hour with a pen tool, or paying remove.bg to do it on their server in two seconds. There is now a third option, and it is the one that matters most for the photos you would not hand to a stranger.
What the tool actually does
The Background Remover runs a neural network inside your browser. The model is a segmentation network trained to tell foreground from background, and it runs on your own machine with WebAssembly and WebGPU - no server, no upload, no account.
The pipeline is:
- Your image is decoded in the browser.
- The model builds a matte - a greyscale mask of what is subject and what is background.
- The matte is applied to the image, and the result is exported as a PNG with transparency.
The whole thing happens on your device. The image is never transmitted, and the model is fetched once and cached.
Why the first run is slow
The first time you use it, the model has to be downloaded - roughly 40 to 45 MB of weights plus the runtime that runs them. That is a real download, and on a slow connection it is a real wait.
After that, the model is in your browser cache and the run is fast. If you ever clear your browser cache, you pay the download once more. The tool does not hide this: it says so on the page.
What the model gets wrong
This is the part that matters, because an AI background remover that is silent about its failures is worse than no tool at all.
The model is good at:
- Solid subjects against a clean background
- People, products and objects with clear edges
- High-contrast scenes
It struggles with:
- Flyaway hair - the fine, individual strands become a soft halo or get cut off entirely.
- Translucent fabric, glass, water - things that are partly background.
- Fur and feathers - the edges are a texture, and the model flattens them.
- A subject that matches the background - a white shirt against a white wall.
- Fine detail at the edge of a small image - the model runs at a fixed resolution, and small images have less to work with.
The honest rule: look at the preview before you download. If the edge looks wrong, the result is wrong, and no amount of hoping will fix it. The output is a standard PNG, so you can open it in any image editor and refine the mask by hand - the AI gets you 95% of the way and you finish the last 5%.
Why "no upload" matters for this specific tool
Background removal is usually the step where a photo of a person, a product or a document gets handed to a third party. The whole point of the operation is that the image is sensitive enough to need cleaning up - a product photo you do not want on a competitor's site, a person's portrait, a scan of a document with numbers on it.
A server-side remover asks you to trust that they process it and delete it. The browser-based one asks you to trust nothing, because there is nothing to trust - the image never leaves the device.
That is not a privacy policy. It is the architecture.
When to use it, and when not to
| Your image | Use this tool? | Why |
|---|---|---|
| A product photo for a shop | Yes | Clean subject, simple background, fast |
| A portrait for a profile | Yes, then check | Hair edges are the weak point |
| A logo on a white background | Yes | The model handles hard edges well |
| A photo where the subject matches the background | No | It will cut the wrong thing |
| A scan of a document | No, use a mask tool | Text and edges are not what the model was trained on |
| A photo of a person with flyaway hair | Maybe, then hand-finish | The AI gets most of it; the last 5% is manual |
A short checklist
- Look at the preview before you download. The edge is where it fails.
- The first run is slow - the model downloads once and is cached afterwards.
- It always outputs a PNG with transparency, whatever you put in.
- Refine by hand if the edge is wrong - it is a standard PNG, any editor can finish it.
- It runs in your browser and the image never leaves your device.
- For a subject that matches the background, use a manual mask instead.
- Keep the original - the result is a new file, and you may need the source.
- Check the result at full size before you publish it.