Virtual Green Screen Online

How the Virtual Green Screen works

A green screen works because a keyer removes one flat colour, and a physical screen exists only to put that colour behind you evenly. If a model can tell which pixels are you, it can paint the rest of the frame that colour without any fabric, lights or clamps. That is all this page does: MediaPipe's selfie segmenter produces a per-pixel probability that each pixel is a person, the tool fills the frame with the key colour, and then draws you on top with the probability as your alpha.

Why use a virtual green screen at all

Chroma keying is the most reliable way to put a person over a different background, and every video tool understands it. OBS Studio, DaVinci Resolve, Premiere Pro, Final Cut, CapCut and iMovie all ship a chroma key effect, and most of them expect green. What they cannot do well is key a normal room: a bookshelf has no single colour to remove. A real screen fixes that, but a decent collapsible one costs £50–150, needs two soft lights to stay even, and takes a corner of the room. For a thumbnail, a profile picture or a short explainer clip, a painted screen gets you to the same keyable frame in seconds.

Where it gets used

• YouTube thumbnails and channel art. Snap a pose on green, key it in your editor and drop yourself onto any design.
• Reaction and commentary edits, where a still or short clip of the presenter sits over gameplay or slides.
• Profile photos and ID-style shots on a plain White backdrop.
• Testing a keyer. A clean, evenly lit key colour is a quick way to check a chroma key setting in OBS before going live.

Choosing the key colour

Green is the convention because camera sensors have twice as many green photosites as red or blue, so green is the cleanest channel to key, and because almost nobody has green skin or hair. Pick Blue if you are wearing green or holding a plant; Magenta if you are wearing both, since nothing natural is that colour. White and Black are not chroma keys at all but are useful for a quick cut-out preview, a passport-style backdrop, or matching a slide.

Why the softness default is low here

On the background remover a soft edge blends you into a blurred backdrop and hides the seam. A keyer wants the opposite: a crisp boundary between you and pure green, because any half-green pixels in the fringe become half-transparent after keying and read as a glowing outline. Six percent softness is enough to stop the edge from looking stair-stepped without creating spill. If your editor's keyer has its own edge-feather control, use that instead and drop this to zero. Edge threshold sets how sure the model has to be before a pixel counts as you; raise it if a halo of background clings to your outline, lower it if your hair is being clipped.

Tips and pitfalls

• Light your face, not the wall. The model finds people more reliably when they are brighter than the background. A window in front of you beats one behind you.
• Keep your hands in close. Arms stretched towards the lens and objects held up to the camera are where the segmenter guesses worst.
• Use the grid to frame. The rule-of-thirds overlay helps place your eyes on the top line for a thumbnail, but it is baked into the snap, so untick it before saving.
• Key with a similarity setting near the default. Because the colour is perfectly flat, the keyer in your editor needs very little tolerance; pushing it high eats into your edges.

What a snap contains

The PNG is the full camera-resolution frame — typically 1280 × 720 — with the key colour baked in as ordinary opaque pixels. If you want the background removed rather than painted, the background remover with Transparent selected saves a real alpha channel instead and skips the keying step entirely.

How to read the result

The stats show Frame filled by you as a percentage. A head-and-shoulders webcam shot is typically 25–45%; under 10% usually means you are far from the camera or the model has lost you, and a figure that jumps around while you sit still points to poor lighting. Look at the edge of your hair and shoulders in the preview: a thin, clean line against flat green is what a keyer wants.

Limits worth knowing

The segmenter runs at a fixed low resolution and has no idea what is a person beyond what it learned from training photos, so a hand held out towards the camera, a coffee mug or a headset boom can be cut out or kept unpredictably. It also cannot see through a translucent object — a glass of water in front of you becomes solid you. For a live stream, a real screen still wins; for a thumbnail, a profile picture or a quick composite, this is a great deal less effort.

Related tools