Remove image background
Cut out people, products, animals or cars in one click – the AI (BiRefNet) finds the subject and removes the background, even around hair and fine edges. Everything happens in your browser: your photos are not uploaded. Transparent PNG, a colour of your choice or a blurred background – for many images at once, too.
No image yet. Drop a photo here – the subject is cut out automatically.
How it works
Choose images
Drop one or more photos into the box, pick them or paste them with Ctrl+V. They are only read locally.
Automatic cut-out
The first time, the browser downloads the AI model once (about 115–190 MB) and keeps it. After that, one image takes one to two seconds on a graphics card.
Adjust & save
Choose a transparent, white, coloured or blurred background or your own image, plus shadow, soft edge, cropping to the subject or an ID-photo template. “Touch up” lets you paint missing parts back or remove leftovers. Then save as PNG, WebP or JPG – several images at once as a ZIP.
Frequently asked questions
Are my images uploaded?+
No. Your photos are only read in your browser and cut out there by the AI – they never leave your device. The only downloads are the AI model (from Hugging Face) and the runtime (from jsDelivr), once; the browser keeps both for later visits.
How well does the cut-out work?+
Very well for clearly visible subjects: people including hair, products, animals and vehicles. Hard cases are subjects that blend into the background in colour, transparent objects such as glass and images with several equally important subjects. The before/after slider lets you check the result closely, and “Touch up” lets you fix it with a brush. If the automatic cut-out picks the wrong subject, choose it yourself with “Pick subject by click” – one click on the object you want is usually enough.
Why is the first start slower?+
The first time, the browser downloads the AI model – about 115 MB with graphics-card support (WebGPU), otherwise about 190 MB. After that it stays in your browser’s storage and starts in seconds. “Delete stored AI model” frees the space at any time.
Which formats and sizes work?+
JPG, PNG, WebP, GIF, BMP and AVIF, plus HEIC photos in browsers that can display HEIC (e.g. Safari). Very large images are scaled down to 6,000 pixels on the long side (4,096 on phones). Save the result as PNG or WebP with transparency, or as JPG.
Why is it slow on some devices?+
With WebGPU (current Chrome, Edge and Safari, many Android phones) the graphics card does the work – one to two seconds per image. Without WebGPU the processor does it on a single core: about 5 to 15 seconds per image and a lot of memory, which older phones may not have.
May I use the results?+
Yes. The AI model BiRefNet is under the free MIT licence, and the cut-out images are yours. As always, mind the rights to your source photos, for example pictures of other people.
Cut-out with the AI model BiRefNet-lite (MIT licence) via transformers.js, entirely in your browser. Results may contain mistakes at difficult edges. Your images stay on your device; only the model (Hugging Face) and the runtime (jsDelivr) are downloaded, once.