remove background from image
Cut out the subject of any photo and keep it transparent, put it on white or a colour, blur the original or drop in your own picture. The AI runs right here in your browser, so your photos stay yours.
Clean edges removes faint leftover background around the outline; soften edges blurs the cut so it blends in.
How to remove a background
- Pick a model: people & portraits (small and fast) or any object (best for products, pets and things).
- Drop, choose or paste one or more images — JPG, PNG, WebP or iPhone HEIC.
- Press remove background. The first time, the model downloads once; after that it's cached.
- Choose the new background: transparent, a solid colour, a blur of the original, or your own picture.
- Fine-tune with clean edges and soften edges, compare with the slider, and download PNG (or JPG for solid backgrounds).
How it works
Your photo is shrunk to about 1,024 pixels on its longest side and given to an AI segmentation model running in your browser (through Transformers.js and ONNX Runtime, the people model on the CPU, the any-object model on your graphics chip via WebGPU). The model returns a mask — how much each pixel belongs to the subject — which is scaled back up and applied to your full-resolution photo, then combined with the background you choose. Nothing is uploaded: the model files come from Hugging Face once, your browser caches them, and every image stays on your device.
people & portraits (fast)
MODNet — made for people — fast on phones, not for products or objects.
Download: ~25 MB, once. Works on any device.
MODNet by Zhanghan Ke et al. (City University of Hong Kong / SenseTime). Licence: Apache-2.0. Model card.
any object (best quality)
BiRefNet lite — products, pets, logos, people — bigger download, slower on phones.
Download: ~110 MB, once. Needs a computer whose graphics chip supports WebGPU with half-precision (fp16).
BiRefNet by Peng Zheng et al. (Nankai University). Licence: MIT. Model card.
Honest limits
- Hair and fur come out soft but not strand-perfect; very fine flyaways against a busy background may be lost.
- Glass, water, smoke and shiny metal show the old background through or in them — no mask can fix that.
- Low contrast (a white mug on a white table, dark hair on a dark wall) confuses every model; a photo with more contrast works best.
- Very large photos are processed at up to 16 megapixels so they fit in browser memory on phones.
- On phones the any-object model is slow and memory-hungry; the people model is quick everywhere.
No AI needed for signatures and logos
Flat artwork on paper or a single colour doesn't need a model at all. The signature background remover and the logo background remover work instantly with simple colour maths — and can recolour the ink to black or pen blue for forms and e-signing.
Frequently asked questions
Is it really free?
Yes. There are no credits, watermarks or size-limited previews, because the AI runs on your own device — there's no server bill to pass on.
Are my photos uploaded?
No. The AI model is downloaded to your browser and runs there. Your images never leave your device, which you can confirm by going offline after the model has loaded — it keeps working.
How big is the download?
The people model is about 25 MB. The any-object model is about 110 MB and is offered on computers whose graphics chip supports WebGPU with half-precision (fp16). Each downloads once and is cached by your browser.
Which model should I use?
People & portraits for selfies, profile pictures and anything with a person as the subject — it's fast even on phones. Any object for products, pets, cars, food and logos photographed in the real world — available on computers with a newer graphics chip. For logos and products on a plain background, the instant logo tool works on every device.
Why is it slow on my phone?
The AI runs on your phone's processor. The people model takes a few seconds per photo. The bigger any-object model isn't offered on phones because it needs more memory than a phone browser allows.
Does it keep the original resolution?
Yes, up to 16 megapixels. The AI looks at a smaller copy, and the resulting mask is scaled up and applied to the full-size photo.
Can I remove backgrounds from many images at once?
Yes — drop up to 20 images. They're processed one after another and you can download them all.