Background Remover
Get a transparent PNG of a person in a few seconds, ready for a profile picture, a poster or a new backdrop.
Loading tool…
How to use it
- Open a photo with a person in it.
- The model downloads once, then runs on your device.
- Download the transparent PNG.
Good to know
The model is trained for portraits. It handles people well, but objects, products and animals may come out poorly.
Fine hair against a busy background is hard for any automatic tool, so results are best with a clear gap between subject and background.
What works well and what does not
| Good results | Weak results |
|---|---|
| One or two people, in focus | Products on a table |
| Clear contrast with the background | Pets and other animals |
| Head and shoulders, or full body | Cars, furniture, logos |
| Even lighting | Hair lost against a similar colour |
How it runs on your device
The cut-out comes from MODNet, a neural network designed for separating people from backgrounds. The model is downloaded from Hugging Face the first time you use the tool, and is then kept by your browser. Your photo is never part of that download.
To stay fast, the outline is calculated on a copy no larger than 1024 pixels. It is then scaled up and applied to your original, so the output keeps full resolution, although very fine strands of hair may be softer than the rest of the picture.
Using the transparent PNG
- Place it on a coloured background in a slide or design tool for a clean profile picture.
- Zoom in on the hair and shoulders before you publish, and retouch any leftover patches if needed.
- A phone may run short of memory on very large photos. Resize to about 2000 pixels first if the page struggles.
- Do not use it for passport or official ID photos unless the rules explicitly allow an edited background.
Questions
Why is the first run slow?
The model has to download once. After that your browser keeps it in its cache.
Is the result full resolution?
The cut-out is applied to your full-size photo, though the mask is calculated at a smaller size.