Add a photo, and the background is removed automatically: the main subject stays and everything behind it becomes transparent. Download the result as a PNG with a transparent background, or as a JPG on plain white if that is what an upload form or marketplace listing asks for.
The cut-out is done by an AI segmentation model called ISNet, an open-source model from the DIS (Dichotomous Image Segmentation) research project, published under the Apache 2.0 licence. It works out which part of the picture is the subject, whether that is a person, a product on a table, a dog or a coffee mug, and masks out the rest. There is no brush to drag around and no settings to tune.
The model runs in your web browser on your own computer or phone. That is why the first photo takes longer: the browser has to download the model once, about 40 MB, before it can start. After that it is kept by the browser and later visits go straight to work.
How to remove Background
- Add your photo. One JPG, PNG or WebP image at a time, up to 100 MB. A photo with one clear subject gives the cleanest result.
- Choose the background. Transparent (PNG) is the default and keeps only the subject. White (JPG) puts the subject on plain white for forms and listings that do not accept transparency.
- Wait for the model the first time. On a first visit a progress bar shows the 40 MB model downloading, for example "12 of 40 MB". This happens once; later photos skip it.
- Download the cut-out. The file is named after your photo with -no-background.png on the end, or -no-background.jpg if you chose white.
What happens to your photo, and how long it takes
The model does not look at your photo at full size. It works on a 1024 × 1024 version, produces a mask that marks each pixel as subject or background, and that mask is then scaled back up and applied to the original. The result keeps the full resolution of the photo you added, up to 16 megapixels. Larger photos, such as 48 MP shots from recent phones, are scaled down to 16 MP first, because phone browsers refuse to create canvases bigger than that.
Speed depends mostly on whether your browser supports WebGPU, which lets the model use the graphics chip. Recent Chrome and Edge on most computers, and recent Safari, have it, and a photo takes a few seconds. Without WebGPU the model runs on the processor instead, which can take up to a minute on an older laptop or budget phone. The second and later photos in the same visit are quicker because the model is already loaded.
The 40 MB download is stored by the browser, so it is not repeated on your next visit. If you clear the browser’s site data, or use a private window, the model is downloaded again the next time.
Photos that work well, and ones that do not
Good candidates: a product photographed on a table or bench, a head-and-shoulders portrait, a pet on the lawn, a single object against a wall. Anything with one obvious subject that stands out from what is behind it will usually come out cleanly on the first try.
Harder cases: tight close-ups where the subject fills the whole frame, such as a flower head or the body of a guitar shot edge to edge, because the model expects some background around the subject to tell the two apart; fine hair or fur against a busy background of a similar colour, glass, bottles and other see-through objects, a subject that blends into its surroundings, very low light, and group photos where it is not clear who or what the subject is. In these the model may cut into the subject or leave patches of background behind. Where it is unsure, such as wisps of hair, it leaves the edge partly transparent rather than making a hard cut, which usually looks more natural on a new background.
If a phone with little memory fails on a very large photo, make the photo smaller with Resize Image and try again. A photo around 2000 pixels wide is plenty for most listings and profile pictures.
Common uses for a cut-out
Product listings are the most common reason. eBay, Etsy, Facebook Marketplace and Shopify stores all look tidier with items on a consistent white or transparent background, and the White (JPG) option gives you a file most listing forms accept as is. Profile pictures, presentation slides, stickers and design mock-ups are the other usual jobs for a transparent PNG.
A photo on white can also be a starting point for an ID or passport-style picture. Be aware that official Australian passport photos have rules beyond a white background, covering size, head position, lighting and expression, so a cut-out from here does not by itself meet those requirements.
Once you have the cut-out, Convert Image changes it to another format, Compress Image makes the file smaller for email or upload limits, and JPG to PDF puts it into a PDF document.
Questions people actually ask
How do I remove the background from a photo for free?
Add the photo above, leave the background option on Transparent (PNG), and download the result. The subject is detected automatically by an AI model, so there is nothing to trace or select. There is no account, and the only wait is the one-off 40 MB model download on your first visit.
How do I make an image background transparent?
Choose Transparent (PNG), which is the default. Transparency only survives in formats that support it, so the file is saved as a PNG ending in -no-background.png. If you later convert it to JPG, the transparent areas will be filled with a solid colour, because JPG cannot store transparency.
Why is the first background removal so slow?
The AI model, about 40 MB, has to be downloaded into your browser before the first photo can be processed, and a progress bar shows how far along it is. The browser keeps the model afterwards, so the next photos and later visits start straight away unless the browser’s site data has been cleared.
Can I put the subject on a white background instead?
Yes. Set Background to White (JPG) before you start. The subject is cut out the same way and placed on plain white, and the file is saved as a JPG. This suits marketplace listings and upload forms that reject PNG files or show transparent areas as black.
Why are the edges around hair a bit see-through?
Where the model is not certain whether a pixel belongs to the subject, as with loose strands of hair or fur, it makes that pixel partly transparent instead of cutting it hard. On a new background this usually blends better than a sharp outline, but busy backgrounds of a similar colour can leave a faint halo.