Guide
Remove Green Screen From a Photo
Short answer: there are only three steps between a photo and a usable transparent PNG, and each one has exactly one thing worth checking first when it goes wrong. Step one is choosing the frame and nominating the key colour; if the result is patchy, the first thing to check is where you sampled, not tolerance. Step two is setting how much gets deleted; if the subject starts dissolving, you have already gone one notch too far. Step three is the boundary and the export; if the edge looks wrong after everything else is right, it is an edge problem and pushing tolerance again will make it worse, not better. Everything below is that path, in the order you actually walk it, with the wrong turn named at each stage.
Step one: pick the frame, and name the key properly
The removal is a colour decision, so the whole job depends on which colour you nominate. Sample it on a genuinely flat, evenly lit patch of the screen. Sampling inside a shadow, on a fold, or close to the subject produces a key that is subtly wrong in a way that only shows up elsewhere in the frame — usually as a dull patch of screen surviving in one corner while everything else is clean.
If you have several shots of the same setup, this is the moment to choose between them. Pick the one with the flattest screen, not the one with the best expression: a frame with a bright patch and a shadowed patch is two key colours wearing one label, and no single setting handles both.
First thing to check if it goes wrong: the sample point. Not the tolerance slider. Most surviving-screen problems are a sampling problem, and moving the slider to compensate will cost you subject pixels somewhere else.
Step two: decide how much is deleted, then leave it alone
Tolerance sets how far from the key colour a pixel may sit and still count as background. Raise it until the screen is gone — and then stop working on this step entirely, which is the part people find hardest. Every further increase deletes subject pixels, and it deletes them first where the subject is closest to the screen colour: dark clothing, olive tones, green accessories.
Worth knowing for this step: brightness is deliberately excluded from the measurement, which is why a screen that looks uneven to your eye can still be cleared by one setting. It is also why a genuinely green object can be eaten along with the screen — the mechanism measures colour distance and has no idea what a subject is.
First thing to check if it goes wrong: whether you are past the point where the screen was already gone. If it was, everything you are seeing now is damage you caused, and the fix is to come back, not to continue.
Step three: rebuild the edge, cap the spill, then save
This is a different job from step two, even though both are done with sliders in the same panel. Deleting the screen decides which pixels vanish; the edge control decides how the transition between vanished and present is shaped. A hard, stair-stepped outline is an edge problem. A pale coloured rim on hair and shoulders is spill — screen light that physically landed on the subject, which is why it must be de-coloured rather than deleted.
Before you save, put the cut-out on a black background and then a white one. Black reveals holes and deleted detail; white reveals surviving screen colour on the edge. A mid-tone background reveals neither, which is exactly why so many cut-outs are approved on grey and rejected on white.
First thing to check if it goes wrong: whether the file actually kept its transparency. A perfect removal saved in a format without an alpha channel is a removal you no longer have. That failure has its own guide, because it is the most common way a good result goes missing.
What to do when the photo has no screen in it
Then there is no key colour to name, and this path does not apply. Nominating one anyway means nominating part of your subject. Removing a background that is not a screen is a recognition problem rather than a colour problem, with a different failure pattern, and expecting colour-key precision from it is the usual reason people conclude the software is broken. The limits are set out here.
Questions
How do I remove a green screen from a photo in three steps?
Sample the key colour on a flat patch of the screen, raise tolerance until the screen is gone and no further, then rebuild the edge and cap the spill before exporting as a PNG. The two most common mistakes sit at the ends: a bad sample at the start, and saving in a format that cannot carry transparency at the finish.
Why is part of the screen still there when everything else is clean?
Almost always because the patch you sampled is not the colour of the remaining area — a shadowed corner, a fold, or a second batch of paint. The precise fix is to sample again on a flat, evenly lit area rather than to raise tolerance, which would delete subject pixels across the whole frame to solve a problem in one corner.
Why does the subject look eaten away after I raise tolerance?
Because you went past the point where the screen was already gone. The pixels now disappearing are the ones closest in colour to the key: dark fabric, olive tones, anything green the subject is wearing. Come back one step rather than continuing forward.
Do I need to keep the original photo?
Yes, for as long as the cut-out is in use. The exported PNG no longer contains the screen colour, so it cannot be keyed again — a second attempt has to start from the original. Storage is cheap; re-shooting is not.
Can I do all of this on a phone?
Yes. The page is built for narrow screens, the file picker opens your photo library, and the processing happens on your device, so nothing is uploaded. Large photos take longer because your phone is doing the work.
Open the green screen remover — free, no account, and the export is a full-resolution PNG with a real transparent background. If you want the depth behind any single step here, the guides page groups every write-up by the problem you are stuck on.