Short answer: drop your image below, click the green background once to sample it, and raise Tolerance until the green is gone without eating the subject. Then download a transparent PNG. There is no upload, no account and no watermark — the pixels are processed on this device.
Waiting for an image. Everything stays in this browser.
Tip: the preview sits on a checkerboard, so anywhere you see squares is genuinely transparent. Click any solid background colour to key it out — green, blue, or something else.
How the three clicks work
- Load the frame. Drag a PNG, JPG or WebP into the box, or pick one. The image is drawn at up to 1400 px on the long edge — enough to judge edges, small enough to stay instant.
- Sample the background. Click once on a clean patch of green. The colour picker updates to that exact pixel, which matters more than any preset: studio green is rarely
#00b140after compression and white balance. - Tune two numbers. Tolerance decides how far from the sampled colour still counts as background. Edge softness decides how many of the borderline pixels fade out instead of snapping to fully transparent.
The order matters. Sampling first means tolerance is measured from the green you actually have, not from a green a preset assumed.
Starting values that work on this keyer
These are starting points, not laws. Every frame is lit differently, so treat the table as a first guess and then trust the checkerboard, not the number:
| Frame type | Tolerance | Edge softness | Why |
|---|---|---|---|
| Clean studio green, sharp subject | 18–25% | 2–4% | The background is uniform, so a tight key keeps the subject solid |
| Phone photo or compressed JPEG | 30–40% | 4–6% | Compression smears green into neighbouring pixels; you need a wider net |
| Hair, fur or thin edges | 12–18% | 6–8% | A tight key plus a soft fade keeps strands from being chewed off |
| Blue screen | 20–28% | 3–5% | Blue keys behave like green ones once the sample is accurate |
| Uneven lighting across the backdrop | 35–45% | 5–7% | The green drifts in brightness; a wider tolerance covers the darker corner |
If raising tolerance starts deleting the subject, stop and fix the sample instead: click a different patch of the backdrop and re-key. If the backdrop disappears but a green rim remains on the subject, that is spill — the next section explains it, and the despill step is what removes it after the key.
Why green edges survive a key
Three things leave green on a subject that looks clean to the eye:
- Spill. The backdrop bounces green light onto the subject's edge — shoulders, hair, a white shirt. A colour-distance key cannot separate that bounce from real green, because the pixel really is green.
- Compression. JPEG and video codecs blend neighbouring blocks. A two-pixel green fringe can become a soft ten-pixel halo that no tolerance value removes cleanly.
- Motion blur. In video frames, a moving edge mixes subject and backdrop in a single pixel. The key sees a colour that belongs to neither.
The practical fix is upstream: light the subject separately from the backdrop, keep the subject a metre or two away from the green, and avoid white clothing. In the tool, a slightly higher edge-softness value hides a thin fringe better than a higher tolerance, because it fades the borderline instead of deleting it.
Green screen remover for video: what this page does and does not do
This is a still-frame keyer. It removes a green background from one image and exports a PNG with an alpha channel. It does not decode an MP4 and it does not write a transparent video file.
For a video clip, the honest workflow is to export a frame sequence, key the frames, and rebuild the clip — or use an editor that keys per frame. A browser keyer that promises one-click video transparency is usually either uploading your footage or returning an approximate matte. If a page cannot tell you which one it does, assume it uploads.
That is also why this tool is deliberately local: the same colour-distance maths that runs here is what a video keyer runs on every frame, and you can see exactly what it costs in edge quality before you commit a whole clip to it.
Check your result before you use it
- The checkerboard shows through the background everywhere — no faint green sheet left behind.
- Hair and thin edges still have detail instead of a hard cut line.
- No green rim on shoulders, ears or light clothing.
- The subject's solid areas are still fully opaque, not semi-transparent.
- The PNG opens on a dark background without a halo you did not see on the checkerboard.
Green screen remover FAQ
Is this green screen remover really free?
Yes. There is no account, no card and no export limit on this page. The tool is a few hundred lines of JavaScript running in the tab, so there is no server cost to pass on and no reason to meter it.
Are my images uploaded anywhere?
No. The file is read with the browser's local file API, drawn to a canvas, and keyed in memory. There is no upload request in the page. If you want to confirm it, open your browser's network panel and load an image — no request carries the file.
Can it remove a blue screen or a non-green background?
Yes. Use the blue-screen preset, or click any solid area to sample its colour. The key is colour distance, not a hard-coded green, so a grey seamless backdrop works too as long as it is reasonably uniform.
What input and output formats are supported?
Input: anything the browser can decode — PNG, JPG, WebP, and usually GIF's first frame. Output: PNG with an alpha channel, because JPEG cannot store transparency.
Why does the subject get holes when I raise tolerance?
Because part of the subject is close in colour to the backdrop. Green clothing, green eyes, or a strong green cast from spill will fall inside the tolerance radius. Lower tolerance, re-sample a cleaner patch of backdrop, or mask the subject by hand.
Why is one part of the green screen removed but a darker patch stays?
Because the key measures colour distance from the pixel you sampled, and a shadowed corner of the same backdrop is further from that sample than the lit part is. Measured with this page's own keying maths on a test frame, sampled at a lit #00b140: the lit backdrop sits 0% away, but a shadowed corner (#007a2c) sits 13.2% away, so at 10% tolerance the lit area is already gone while the shadowed corner is still fully opaque. Bumping tolerance to 14% clears the shadow too. The trap runs both ways — if you sample the shadowed corner first, the lit area is the one 13.2% out, and 10% leaves it behind. If 14% starts eating the subject, even out the lighting instead of pushing tolerance further.
Does raising tolerance get rid of the green rim on hair and shoulders?
No, and it cannot, because the rim and the subject's own green are both green pixels. In the same measured run, a green spill rim on a shoulder (#3f8f52) sat 16.7% from the sampled green, while a genuinely green piece of the subject — a green shirt or a green eye (#2f7d3f) — sat 15.9% out. So tolerance has to reach 16–17% before the rim goes, and at 16% the subject's own green is already deleted. What helps is Edge softness, which fades borderline pixels instead of removing them: at 14% tolerance with 6% softness the spill rim dropped to roughly half opacity (alpha 115) while skin and white stayed fully opaque. The durable fix is upstream — stand the subject away from the backdrop and light it separately. This page has no spill-suppression slider, and it does not pretend to; the clamp that does the job is on green screen remover spill.
How do I remove a green screen from a JPG if JPEG has no transparency?
Key the JPG here, then save the result as a PNG — a JPEG cannot store the alpha channel, so exporting back to JPG would replace the transparent background with white, black or a solid colour. Load the file, sample the green, tune tolerance, and download the PNG this page produces. A heavily compressed JPG usually needs a wider tolerance (30–40%) and a softer edge, because compression has already mixed green into the subject's outline. The same limit applies to a JPEG frame exported from a video: key it, keep the PNG, then rebuild the clip.
Can I remove a green screen from a video here?
Not directly. This page keys a single still frame and exports a PNG. For video, export frames, key them, and reassemble — or use an editor with a per-frame chroma key. The page is honest about the limit rather than pretending a browser can hand you a transparent MP4 in one click. The full workflow, including the FFmpeg commands and alpha checks, is in green screen remover for video.
Which is better for green screen: chroma key or AI background removal?
Chroma key is better when the backdrop is actually green and reasonably even: it is deterministic, needs no model download, keeps the original edge pixels, and you can see exactly which colour was removed. AI segmentation is better when there is no green backdrop at all, or when the subject is hair-heavy and the background is complex — but it guesses the edge, and a guess can change between runs. For a shot that already has a green screen, keying first and using AI only as a cleanup pass is the more predictable order. A colour key also cannot invent a subject it did not see; it can only remove pixels near the sampled colour, which is why spill still needs its own step.
How do I check a keyed PNG for leftover green before I use it?
Look at the alpha channel, not the checkerboard. Three checks catch almost everything: a matte view should show a solid white subject on solid black, with no grey sheet over the backdrop and no grey holes inside the body; any green that survives as a semi-transparent pixel more than about three pixels inside the silhouette is a leftover, not an edge; and the same PNG on white and on black should not gain a green rim on one of them. A 12/255 veil is only about 7 RGB points away from this site's checkerboard but 16 away from white, which is why dark preview backgrounds pass keys that fail on a bright page. The working checker with counts and pass limits is on green screen remover edge check.
Next step
Key your frame above, then take the transparent PNG into your editor or thumbnail layout. If you are building an AI video job and need the prompt side instead, start at the Seedance 2.5 job builder or read the guides.