Tools
Wanting a white background on a profile photo, or the background gone from a product shot, comes up more often than you would expect, and since firing up an image editor just to erase a background is overkill, most people go looking for a free tool online.
What stopped me there was that every tool I found worked by uploading the image to a server. Handing my own face, or a product that has not been announced yet, to somebody else’s server is not a decision to make lightly just because it costs nothing, and plenty of them come with caps on how many images you can do, or drop the resolution unless you pay.
This article walks through the background removal in Media Tools, which I built, and how to make a background transparent without sending the image anywhere. Everything runs inside the browser, so the photo never leaves your device.
- What you need: the image, and a browser
- What you do not need: an account, a payment, an upload, or a cap on how many images you process
What changes when nothing is uploaded
Online background removers almost always process on the server. That is faster, since the server’s hardware does the work, but a set of constraints comes along with it.
Typical cloud service | Processed in the browser (this article) | |
|---|---|---|
Image transfer | Uploaded to a server | Never leaves the device |
Number of images | Usually a daily cap | None |
Resolution | Often reduced on free tiers | Unchanged |
First-run wait | Almost none | The AI model has to be fetched |
Speed | Fast | Depends on your machine |
Processing in the browser has its own weaknesses. The AI model has to be downloaded the first time, and the speed tracks whatever hardware you are on, so it is a trade: slower, in exchange for never handing the image over.
I chose this approach because that trade is decisive whenever the material cannot leave the company, or the photo belongs to a client.
Choosing between the two models
Two AI models are available. Try "People" first and switch to "General" if it does not come out cleanly — that is the order that works in practice.
People | General | |
|---|---|---|
Model | MODNet (25 MB) | BiRefNet_lite (100 MB) |
WebGPU | Not required | Required |
Measured on a 900×1350 photo | 16.4 s | 46.5 s |
Suits | People, and subjects with fur or hair | Hard outlines such as cars and buildings |
As the name says, "People" was built for people, but it holds up surprisingly well on other subjects. Trying it on a photo of a dog sitting in grass for this article, it cut the fur out cleanly down to the outline. Running the same photo through "General" left part of the white fur on the chest missing, and took nearly three times as long.
So "anything that is not a person needs General" does not hold. I start with "People", which is faster and does not need WebGPU, and only reach for "General" when the result does not satisfy me.
Why start with the People model
It is faster and runs even on machines without WebGPU, and at 25 MB the model is small enough that the first-run wait is barely noticeable. Subjects whose outline breaks into fine strands, like hair or animal fur, are exactly what it is good at.
When to switch to the General model
It is worth trying when the People model breaks the outline. Subjects with a defined edge — cars, buildings, manufactured goods — tend to come out more consistently with it.
The catch is that it will not run without WebGPU support. It processes at 1024×1024 and uses a lot of memory as a result, so on machines without it only "People" is available. In my measurement it also took nearly three times as long as the People model.
How to cut out the background
Free tool that removes the background from a photo with AI and saves it as a transparent PNG or over a colour of your choice. Switch between a model for people (MODNet) and a general one (BiRefNet). Inference runs in your browser, so the photo is never sent anywhere.
media.tools.ryusei.io- Open the background removal page in your browser.
- Drag and drop the image.
- Pick "People" or "General" under "Subject".
- Decide the "Background": leave it transparent, or fill it with white or another colour.
- Run it, and the cut-out image is written out.
The AI model downloads only on the first run — 25 MB for the People model, 100 MB for General. From the second time onward the copy stored in your browser is used, so there is no wait.
The figures in this article come from a machine with an integrated GPU (Radeon 780M) running Chromium, on a 900×1350 photo. They include fetching the model, so later runs are quicker, and a machine with a discrete GPU finishes faster still.
Fixing what the AI misses
Since the AI decides on its own, there will be times when a touch-up is needed. When the subject and background are close in colour, or the subject has a lot of fine detail, the outline may not land where you want it, and I built in a way to correct it by hand from the start on that assumption.
What I built is an editor that allows the following on top of the automatic result.
- Select a region and limit the work to inside it
- Cut out by hand, freehand
- Bring back what was erased too far, or erase what was left behind
As long as the automatic pass gets most of it, all that is left is tidying the remainder with a brush, which takes nothing like the time of cutting the whole thing out by hand.
Output formats and background colour
Only PNG and WebP preserve transparency. JPG has no mechanism for holding a transparent pixel at all, so choosing JPG writes out an image filled with the background colour instead.
Use | Format | Background |
|---|---|---|
Layering onto a document or slide | PNG | Transparent |
Putting it on a website | WebP | Transparent |
Submitting it as an ID photo | JPG | Filled with white |
For an ID photo on a white background, pick white under "Background" and save as JPG, and the result is ready to submit as is.
Frequently asked questions
Is my photo sent to a server?
It is not. After the AI model is fetched into your browser on the first run, inference happens on your own machine. The image itself never travels over the network.
Is it free?
It is. No account is required, and no cap is placed on how many images you process.
Can I put a white background on an ID photo?
You can. Pick white under "Background" and the transparent area is filled with white; JPG or PNG both work.
Does it work on products and animals?
It does. Try "People" first, and switch "Subject" to "General" if the outline breaks.
Does it work on a phone?
The People model does. The General model needs WebGPU and may be unavailable depending on the device, in which case choose "People".
Can I use the transparent image anywhere?
Saving as PNG or WebP preserves the transparency, but anywhere that does not support it will fill the transparent area with white or black. Older software and some services still do this, so check how it looks where you paste it.
Summary
Making a background transparent does not require uploading the image anywhere. Run the AI inside your browser and the photo stays on your machine while it is cut out.
- Try "People" first, and switch to "General" if it does not come out cleanly
- The General model needs WebGPU and takes nearly three times as long
- Whatever the automatic pass misses can be fixed with region selection and a brush
- Only PNG and WebP keep transparency; JPG gets filled with the background colour
Convert, compress and edit images, audio and video entirely in your browser, with AI upscaling and background removal. Free, no upload, no size limit, no sign-up.
media.tools.ryusei.io
No install, no upload. Transcribe a video, fix the SRT, and burn the subtitles into the picture — all inside your browser, free and without an account.
Programing
Python
Hosting