Guide
Green Screen Remover for Photos
Short answer: a still photo is judged harder than a moving picture, and that single fact changes how you set everything. In a video, a wrong pixel flashes for one frame and is gone; in a photo, somebody can open it, zoom to 100% and stare at the same edge for as long as they like. So the bar is not “looks fine at a glance” — it is “holds up at 100%”. The order below is built for that: choose the frame, sample the key on a genuinely flat patch, set tolerance until the screen dies and the subject does not, rebuild the edge, take the spill off hair, verify on two backgrounds, and only then export as a PNG with a real alpha channel. Skipping the order is what produces the two failures in the diagram: holes where the subject should be, and a halo where the edge should end.
What a still photo changes, and what it does not
Nothing about the arithmetic changes. A photo taken against a green screen has the same pixels, the same colour distance between each one and the key colour, and the same need to rebuild the boundary. What changes is the standard you are held to, and three practical consequences follow from it.
| Frame in a video | Still photo | |
|---|---|---|
| How long the edge is on screen | A fraction of a second, usually moving | Indefinitely, and it can be zoomed |
| Does a bad pixel get hidden | Often — motion and the next frame cover it | Never. There is no next frame |
| Can you re-do the take | Yes, re-render from the original footage | Only if you kept the original file, not the exported cut-out |
| What noise does | Averages out over neighbouring frames | Sits still and gets keyed along with the screen |
The third row is the one that decides whether you can afford to experiment. As long as the original frame still exists, every setting is reversible: you can go back, sample a different patch, try a narrower tolerance, and export again. The moment you have only the exported PNG, the screen colour is gone from the file and there is nothing left to key against. So the cheapest thing you can do before touching any slider is to confirm the original is still where you think it is.
The order that avoids both failures
The two failures in the diagram above come from doing these steps out of order, not from the settings themselves.
- Choose the frame or the file first. If you have several shots of the same setup, pick the one with the flattest, most evenly lit screen. A frame with a bright patch and a shadowed patch is two key colours wearing one label, and no single tolerance setting handles both.
- Sample the key on a genuinely flat area. Not near the subject, not near a fold, not in a shadow. If the sample is off, the screen survives in patches while the subject starts dissolving somewhere else, and you will spend the next ten minutes moving the wrong slider.
- Set tolerance until the screen is gone, then stop. Every extra step past that point eats the subject, and it eats it first where the subject is closest to the screen colour.
- Rebuild the edge. This is a separate control from tolerance. A hard, stair-stepped outline is an edge problem. Reaching for tolerance again makes the outline softer in the worst way — by deleting more of the person.
- Take the spill off. Green light bouncing onto hair and shoulders leaves pixels that are genuinely part subject and part screen. Capping how much of the screen’s channel may survive there is the only thing that removes the halo.
- Verify at 100%, on two backgrounds. Black reveals holes; white reveals halos. A mid-tone reveals neither, which is exactly why so many cut-outs are approved on a grey background and rejected on a pale one.
- Export with an alpha channel. A perfect removal saved in a format without transparency is a removal you no longer have.
Why a photo can fail even when the video of the same setup did not
Two mechanisms account for most of it, and neither is a mistake by the person doing the keying.
Compression sits where the edge is. A photo stored as JPEG has been through a lossy step that works in small square blocks, and the boundary between a subject and a screen is exactly where those blocks have the least to work with. The result is a thin band of pixels that are neither pure screen nor pure subject. A tolerance wide enough to clear the whole band can start taking bites out of the subject; a tolerance narrow enough to protect the subject leaves part of the band behind. There is no setting that makes those two goals stop competing — there is only a choice about which one you would rather fix by hand afterwards.
Noise is real pixels. A phone photo shot in a dim room carries sensor noise, and some of that noise lands close enough to the screen colour to be keyed. This is why holes appear inside dark clothing and deep shadows on an otherwise clean removal: those areas are where the noise has the most room to drift towards the key. Tightening tolerance brings the holes back but re-opens the screen. Again, the honest fix is the edge and matte work, not one more push on the same slider.
What this site’s tool does with a photo
It takes the still, lets you sample the key by clicking the screen area, and gives you tolerance, edge softness and spill removal — the same four ideas the list above is built around. The export is a PNG with a real transparent background at the exact pixel size of the file you loaded, and the whole calculation runs in the page, so the photo is not uploaded anywhere. What it does not do is guess your subject: it is a colour tool, so a photo with no screen in it has nothing for it to work with. That limitation is covered in full here, because knowing it in advance saves a lot of confusion.
Questions
Does a green screen remover work differently on a photo than on a video?
The maths is the same, the standard is not. A frame in motion is forgiven quickly; a still is inspected. In practice that means you set the same controls but stop earlier on tolerance, spend more effort on the edge, and always verify at 100% before you save rather than afterwards.
Why does the removal look worse after I save it as a JPEG?
Because JPEG has no alpha channel, so the transparency cannot survive the save. What you get is the cut-out flattened onto a solid colour, and on a light background that reads as a white box. Use PNG, or WebP where you know it is supported, and re-check the file over both a dark and a light background.
There are holes inside the subject after removing the background. What caused that?
Pixels inside the subject were close enough to the key colour to be deleted. That happens most often in dark areas, where sensor noise drifts towards the screen colour, and on anything the subject is wearing that shares a hue with the screen. Lowering tolerance restores them, but re-opens the screen where it is dimmer — which is why the better sequence is to sample the key again on a flatter patch and keep tolerance as low as the screen allows.
Should I shoot in RAW or JPEG if I plan to key it?
RAW gives you more room to correct the screen before the key, because the adjustment happens on the full sensor data rather than on an already-processed file. JPEG is not disqualifying — it just means the boundary band is already fixed in the file, so more of the final fix has to happen on the edge. Either way, keep the original: it is the only thing that lets you try again later.
Do I need to move the subject away from the screen?
Yes, and it is the single cheapest thing you can do before shooting. Distance reduces both the green light bouncing onto the subject and the chance that a shadow from the subject lands on the screen. Both of those are problems that no keying setting fixes afterwards, because the damage is already in the pixels.
Why is the outline still hard even after the screen is gone?
Because the boundary has not been rebuilt. Removing the screen decides which pixels are transparent; it does not decide how the transition between transparent and opaque is shaped. That is a separate control, and on a still photo it is worth more of your attention than tolerance.
Open the green screen remover — free, no account, and the photo never leaves your device. If the edge is the specific problem, the symptom-by-symptom guide takes the controls in order; if you have not chosen the key colour yet, start here.