ᾨ4 Background Remover
A background remover for portraits, running in your browser at full resolution. No credits, no account, no upload. Transparent, white or blue backdrop.
Choose a photo of a person, or drop it here
The photo stays in this tab. Full resolution, no credits, no account.
This is a portrait cutout. The model separates people from backgrounds and is good at it, hair included. It is not a general object cutout — a product, a car or a pet is not what it was trained on and the result will show that.
What Background Remover Does
Cutting a person out of a photograph is a matting problem, not a selection problem. A hard outline is easy and looks wrong, because the edge of a person is not hard: hair, the fuzz along a jaw, a soft shoulder against a bright window are all pixels that are genuinely part background and part subject. What a matting model produces is an alpha value per pixel — how much of that pixel belongs to the subject — and getting those in-between values right is the whole difference between a cutout that passes and one that looks like it was done with scissors.
This runs the model in your browser. That is unusual: every well-known background remover uploads the photograph to a server first. Here the file is read by the page, the model runs on your machine, and the result is composited locally — so the photo is never transmitted, and the image you get back is the resolution you put in, with no account, no credits and no watermark.
The model is MODNet, chosen for two reasons that are worth stating because they constrain what the tool can do. It is Apache 2.0, which matters: the model most browser tutorials use is licensed for non-commercial use only and cannot legally ship on a page that carries advertising. And it is small — 24.7 MB, downloaded once and then cached by your browser — where general-purpose segmentation models run into the hundreds of megabytes.
The constraint that comes with it: MODNet is a portrait matting model. It was trained to separate people from backgrounds and it is good at that. It is not a general object cutout. A product on a table, a car, a pet — it has no concept of those, and the result will show it. That is a real limit and it is better said here than discovered on a product photo.
How to Use Background Remover
- Add a photo of a person, or drop it onto the box
- Wait once while the 24.7 MB model downloads — your browser caches it for next time
- Choose a transparent background, or white or light blue for an ID photo
- Nudge the edge slider if a halo survived or fine hair was cut, then save
Where the work happens
Checked against the first page of results on 8 September 2026.
| This page | Every first-page result | |
|---|---|---|
| Photo leaves your device | No — there is no server to send it to | Yes, uploaded and processed remotely |
| Output resolution | The resolution you supplied | Varies by service and plan |
| Account required | No | Usually, beyond a few images |
| Works offline once loaded | Yes — the model is cached by the browser | No |
| Cost of the model | 24.7 MB, once | None to you; borne by the service |
Choosing the background
The output format follows from the choice — a solid color leaves no transparency to store.
| Background | Saved as | Use it for |
|---|---|---|
| Transparent | PNG | Logos, overlays, putting the subject on something else later |
| White | JPG | Passport, visa and most exam-form photos; e-commerce listings |
| Light blue | JPG | ID photos where a plain blue ground is specified instead of white |
| Red / custom | JPG | Some national ID formats, and design work |
How to Read Your Result
The edge slider fixes the two ways it goes wrong
A pale halo of the old background surviving around the subject means the alpha is slightly too generous — move the slider right. Fine hair disappearing means it is slightly too strict — move it left. It shifts the transparency threshold rather than blurring, so the edge stays crisp in both directions. On an even, uncluttered background you will not need it.
Give it a fighting chance
Matting models do their worst work where the subject and the background are the same brightness — dark hair against a dark room, a white shirt against a white wall. If the result is disappointing, the fastest fix is usually a different source photo rather than a different tool: more separation between subject and background, and even light, do more than any slider.
It chains into an ID photo
A plain white or light blue background is what passport, visa and exam-form photos require, which is the one case where this tool and the geometry tools are really one workflow: remove the background here, set the shape and pixels in the cropper, then meet the KB limit in the photo resizer.
Limitations & Accuracy Notes
- Portraits only. MODNet is a human matting model — products, vehicles, food and animals are outside what it was trained on and the cutout will be poor.
- One face at a time is what it handles best. Group photos and crowds produce a less reliable matte.
- The model is 24.7 MB and is downloaded the first time you use the tool. It is then cached by your browser, but the first run needs a connection and a few seconds.
- It runs on your own hardware, so an old phone will be noticeably slower than a laptop, and a very large photo costs more memory.
- Hair against a background of similar tone is the standard hard case for every matting model, this one included. The edge slider helps; it does not work miracles.
- No manual brush or erase. What the model produces is what you get, adjusted only by the edge threshold.
Frequently Asked Questions
Is my photo uploaded to a server?
Does the free version give me a lower resolution?
Why does it only work well on people?
How big is the download and does it happen every time?
The edges look wrong — can I fix them?
Can I use this for a passport or visa photo?
Does it work on my phone?
References & Further Reading
- MODNet — official repository — States the code and models are released under Apache 2.0, which is why this model rather than a more popular one is used here
- MODNet for transformers.js — The ONNX conversion this page loads, listed as Apache 2.0
- BRIA RMBG-1.4 model card — The model most browser tutorials use, released for non-commercial use only — the reason it is not used here
- Transformers.js — The library that runs the ONNX model in the browser