Why is there still a green fringe on hair after chroma key?

Because the key removes pixels by colour, and a strand of hair is thinner than one pixel. The camera never recorded that strand as pure green or pure hair — it recorded a blend, so the key leaves it behind as a semi-transparent green edge. Clear it with Spill suppression, not with more Tolerance: extra tolerance removes the fringe and the hair along with it.

What the fringe actually is

A chroma key does not recognise a person. It compares each pixel's colour against one key colour and decides how transparent that pixel should be. Pixels far from the key colour stay solid, pixels close to it become transparent, and pixels in between become partially transparent.

Hair breaks that clean split. A strand may cover only a fraction of a pixel, so the sensor averages the hair with the green behind it and writes down a value that is genuinely part green. On a moving head those half-covered pixels form a soft band around the whole silhouette, and that band is what you are seeing.

This is why the fringe is worst on wavy or backlit hair and almost absent on a hard-edged object like a coffee mug. Curly hair creates far more half-covered pixels per inch than a straight edge does.

Why turning up tolerance makes it worse

Tolerance widens the range of colours treated as background. Push it high enough and those blended edge pixels finally fall inside the range and disappear — along with every other pixel that happened to sit near that colour, which on hair means the fine tips go with them.

The result looks like a bad haircut rather than a fringe: the outline loses its soft detail and starts to read as a solid blob. It is the most common way people try to fix a fringe and the one that does the most damage.

The rule that holds in practice: raise Tolerance only until the background is gone, then stop. Anything left over at that point is a spill problem or an edge problem, not a range problem.

Spill is not the same problem as the key

Spill is green light bouncing off the screen onto the subject — shoulders, forehead, a white shirt, the pale side of a face. It is real green that the camera recorded, so a tighter key will not touch it, because that pixel's colour is close to skin, not close to green.

Spill suppression works differently: it is a green-channel limiter. Wherever green is stronger than both red and blue by more than a set margin, it pulls green down until the pixel reads neutral. Skin rarely trips that condition, which is why you can push it fairly hard before flesh starts to look grey.

If the whole subject starts to look desaturated, you have gone too far. Take the slider back down rather than trying to compensate elsewhere; adding saturation on top of an over-suppressed image never restores what was thrown away.

The order that actually works

Three controls, in this order, and only as far as each one needs to go:

1. Tolerance — up only until the background disappears. Stop there; do not chase the fringe with it.

2. Edge softness — a small amount restores the natural transition that keying removes. Real footage shot on a real lens is never razor-sharp at the boundary, and an edge keyed too hard looks pasted on.

3. Spill suppression — up until the remaining green tint turns neutral. This is the control that actually removes a fringe.

Doing them in this order matters, because each one changes what the next one has to work with. Spill suppression applied before the background is gone will flatten colour across a still-green frame and you will end up fighting both problems at once.

Why 4:2:0 footage gives you the worst edges

Most consumer cameras and nearly all delivery codecs record 4:2:0, meaning colour is stored at a quarter of the resolution of brightness. Brightness detail is full; the colour information that a chroma key depends on has already been thrown away before the file reaches you.

That loss lands exactly where it hurts most — on edges, where colour changes fastest. A 4:2:0 clip gives the key a coarse colour map to work from, so the boundary between hair and screen is decided from pixels that were averaged with their neighbours.

If the shoot is still ahead of you, recording 4:2:2 or 4:4:4 where the camera allows it is the single biggest improvement available, worth more than any amount of post-processing. If the footage already exists, expect to lean on spill suppression and edge work instead.

When the screen itself is the problem

A key assumes the background is one colour. A real screen rarely is: wrinkles create shadowed folds, a hot spot near a light creates a washed-out patch, and the whole thing darkens towards the corners. Each variation is a slightly different green, and one key colour has to stretch to cover all of them.

The fix is physical, not digital — light the screen evenly, keep the subject far enough in front of it to avoid shadows and bounce, and flatten the material. Tuning sliders can compensate for mild unevenness; it cannot rescue a screen with a bright hotspot and a dark fold in the same shot.

If the background is not flat at all — an office, a street, a messy room — colour keying is the wrong tool. Segmenting the subject instead ignores colour entirely and will hold up where no amount of tolerance adjustment will.

When to stop keying and segment instead

Subject segmentation finds the person rather than the colour. It is the better choice when the green screen is unevenly lit, when the shot was never meant to be keyed, or when the subject is wearing something close to the key colour.

It costs more time per frame than a colour key and it is not magic — a raised hand or a wisp of hair can still be missed. Check the result before exporting rather than trusting the mode.

Both approaches live in the same tool: colour keying for a real green screen, segmentation for everything else.