Why the colour matters
A green screen works on one idea: if a region of the frame is a colour that appears nowhere on your subject, a computer can identify it by value alone and delete it. Green wins because it is far from every skin tone, because Bayer sensors sample green twice as often as red or blue and therefore give the cleanest signal there, and because it is easy to buy in bright, saturated, matte form. Blue screen predates it and is still the right answer when the wardrobe is green or the subject is very blonde.
What you get out the far end is an alpha channel: a per-pixel map of how opaque the subject is. That is the actual deliverable. A hard binary cutout looks like a sticker, and the reason good compositing looks good is that the alpha channel carries fractional values along hair, motion blur and glass.
What keying actually does
A keyer takes a target colour and a tolerance, and marks every pixel inside that range as transparent. Everything else about the process is damage control around that blunt operation.
- Tolerance decides how much colour variation counts as background. Too tight and the screen survives in patches. Too loose and it starts removing shadow on the subject.
- Matte choke shrinks or grows the edge by a fraction of a pixel to hide fringing, at the cost of thin detail.
- Despill shifts pixels that are partly green back toward skin or hair colour. Without it, every edge reads faintly radioactive.
- Garbage matte is a rough hand-drawn shape that throws away the lighting stands and screen edges before the keyer ever sees them.
Order matters. Despill before choking, because choking removes the very pixels despill needs to work on.
Lighting it so it keys cleanly
Almost every bad key is a lighting problem rather than a software problem. Three rules cover most of it.
Light the screen and the subject as two separate jobs, with their own instruments. Get the screen as even as you can, ideally within a stop across the whole surface, because unevenness forces the tolerance up and a high tolerance is what eats hair.
Then buy distance. Three metres between subject and screen kills most spill, lets the screen fall out of focus, and stops the green from wrapping around shoulders. If the room is small, spill control matters more than screen size.
Finally, put a hard light behind the subject that is not green. A rim from a tungsten or daylight source rebuilds the edge that despill will otherwise flatten, and it gives the keyer an unambiguous boundary to find.
Green screen versus AI matting
Segmentation and matting models now pull a usable alpha channel from footage shot against a wall, a street, or a bookshelf. For a single talking subject in reasonable light this is faster and cheaper than any keying setup, and it is what background removal tools actually run.
The tradeoff is what each method knows. Chroma key measures real colour values, so it handles fractional opacity honestly: motion blur on a swinging arm, individual flyaway hairs, cigarette smoke, a glass of water. AI matting infers an outline, so those same cases are where it guesses, and guessing shows up as chattering edges frame to frame.
Practical split. Shoot a green screen when the plate has transparency, fast motion, fine hair, or has to hold up at full resolution in a wide shot. Skip it and use matting for talking heads, product shots on simple backgrounds, and anything where a soft edge is acceptable. When you do use matting on video, judge it in motion rather than on a single frame, because temporal instability is its characteristic failure and a still will not show it.