Make an image background transparent

Pick the background colour and everything close to it turns transparent: green screens, white backdrops, scans and flat logos, in a blink.

Drop your files here

.jpg, .jpeg, .png, .webp, .avif · up to 50 MB

Choose files

How to Make Background Transparent

1
Add the image
JPG, PNG, WebP or AVIF, up to 100 files at a time. Phone photos are turned upright before the background is measured.
2
Choose the background colour and the tolerance
Pick a colour, or let the tool read it from the corners, or key out plain white. Raise the tolerance until the background is gone and lower it if the subject starts disappearing.
3
Download the PNG
The result is always a PNG with a real alpha channel, named after your file with -transparent on the end.

How this differs from AI background removal

Remove Background runs a neural network that understands what a subject is. It separates a person from a street, a product from a cluttered desk, hair from anything, and in exchange a reduced copy of the photo goes to our server and the job takes several seconds. This tool does arithmetic on colour instead. It measures the distance between every pixel and one chosen colour and turns the close ones transparent. That makes it useless on a photo of a room, and excellent on the cases it was built for: a green or blue screen, a white studio backdrop, a document scanned on white paper, a logo on a flat plate. It never downloads a model, it never touches the network, and it finishes a three-megapixel photo in about a fifth of a second. When the tool detects that the border of your image is not one colour, it says so and offers the AI tool instead of quietly producing a ragged cut-out.

Why the distance is measured in YCbCr

Comparing colours by their red, green and blue values treats a shadow as a completely different colour from the thing casting it, because all three channels drop together. Chroma keying has used the YCbCr space for decades for exactly this reason: it separates brightness from colour, so a shadowed patch of green screen is still recognisably the same green. The distance here weights the two colour components twice as heavily as brightness, which is what keeps shadows inside the subject opaque instead of punching holes through them. Tolerance sets where the cut happens and softness sets how wide the transition band is. A couple of pixels of soft edge is what stops the cut-out looking like it was made with scissors.

Spill, and the green line around the hair

A green screen lights the subject as well as the wall. Light bounces off the backdrop onto shoulders, cheeks and especially hair, leaving a green rim that survives the key because those pixels are not green enough to be removed. Spill removal fixes it: on the semi-transparent edge pixels, the channel that dominates the key colour is pulled towards the average of the other two. At forty percent, which is the default, the rim goes from obvious to invisible without shifting the subject's own colours. If your subject genuinely is green, turn it down. The edge shrink control does the geometric equivalent, eroding or dilating the alpha mask by up to four pixels. One or two pixels of erosion is the usual cure for a pale halo that spill removal alone cannot reach.

No model, no upload, no network

Everything happens in this browser tab: reading the file, measuring the background, computing the alpha channel and writing the PNG. No neural network is downloaded, nothing is uploaded, and closing the tab clears the image. That also means the tool keeps working with the connection off once the page has loaded.

Frequently asked questions

How is this different from AI background removal?

This tool removes one colour; the AI tool understands a subject. Colour keying is instant and downloads nothing, but it only works when the background really is one colour. If the background is a room or a street, use Remove Background. This tool will tell you when that is the case rather than handing back a ragged edge.

Why is the output always a PNG?

Because JPEG has no alpha channel. Saving a cut-out as JPEG would fill every transparent pixel with a solid colour, which throws away the entire point of the operation. WebP could store the transparency, but PNG is what every design tool, office suite and website accepts without question.

What do tolerance and softness do?

Tolerance is how far a pixel's colour can be from the key colour and still count as background. Raise it if patches of backdrop survive, lower it if the subject starts dissolving. Softness is the width of the fade between fully transparent and fully opaque; a small amount gives a natural edge, none at all gives a hard, jagged cut.

Where does the green rim come from and what is despill?

It is light bouncing off the backdrop onto the subject, strongest in hair and on shoulders. Despill pulls the dominant channel of the key colour back towards the other two on the semi-transparent edge pixels, which removes the rim without changing the subject's real colours. Forty percent suits most footage; raise it if the rim is heavy.

Does it work on ordinary photographs?

Usually not, and the tool checks. It measures the colour spread around the border of the image, and if that spread is wide it stops and points you at AI background removal. A photo of a person outdoors has no single background colour, so no tolerance setting will produce a clean edge.

Can I process a batch of files?

Yes, up to 100 files with the same settings, which is exactly the case a product catalogue shot against one backdrop needs. The results come back as pulsarkit-transparent.zip, and a file that cannot be read does not stop the rest.

Does the file go to a server?

No. Every step runs in your browser. There is no upload, no model download and no stored copy. This is arithmetic over the pixels of a file that never leaves your device.