Can you key against a white wall lit green instead of a real green screen?

Two gates, and the invisible one comes first

The idea is appealing: skip the screen, hang a green bulb, throw some light at the wall behind your subject. It fails for two independent reasons, and the one that kills it is the one nobody looks at.

The visible gate is evenness. A lamp does not light a wall uniformly, and a keyer's tolerance for variation is smaller than people expect — our piece on the conditions for automatic removal puts it at 21.8% of brightness variation before the default settings stop cleaning up.

The invisible gate is chroma. The wall has to come out the right green, not merely a green. That is a much tighter requirement than it sounds, and it is the one we will take first, because if it fails, the evenness does not matter.

Gate one: is your green the right green?

Complete removal requires a pixel to sit within 0.161 of the key colour in normalised RGB distance. Between 0.161 and 0.259 it is semi-transparent, and beyond 0.259 the keyer keeps it entirely.

Now shift the key green #00B140 in hue and see where it lands:

• +5 degrees — 0.058, still removed
• +10 degrees — 0.116, still removed
• +15 degrees — 0.174, semi-transparent
• +20 degrees — 0.231, mostly kept
• +30 degrees — 0.347, kept entirely

So the tolerance is roughly thirteen degrees of hue. That is not much, and it is far less than the spread between greens that all look 'green' to the eye. Saturation is tighter still in the direction that matters: dropping it 10% gives 0.082 and survives, while dropping it 20% gives 0.165 and the wall starts coming back.

Here is the part that settles it. Representative sRGB values for common green light — chosen by us, not measured from a lamp:

• pure green LED #00FF00 — 0.396, kept entirely
• cyan-leaning green #00FF80 — 0.396, kept entirely
• yellow-leaning green #7CFF00 — 0.627, kept entirely
• deep green #00A030 — 0.092, removed

Key green is a blue-leaning green, and ordinary green light is not. A green bulb, a green gel or the green channel of an RGB fixture will generally land somewhere near 0.396, which is two and a half times the distance at which anything gets removed at all. You can light that wall as evenly as you like and it will not key.

This is also why the fix is not 'use a better green'. You would need a source whose chromaticity is actually close to #00B140, and that is a specification most green lighting does not publish.

Gate two: how unevenly does a lamp light a wall?

Assume the first gate is passed somehow, and look at evenness. Modelled on a 3.0 x 2.5 m wall, a single point source at wall-centre height, and the most favourable exposure available — the brightest point lands exactly on key green, so every other point can only be darker. Every figure below is therefore optimistic:

• 1.0 m — 81.9% of the wall survives the key, 74.1% of it heavily
• 1.5 m — 59.2% survives, 41.8% heavily
• 2.0 m — 28.4% survives, 8.9% heavily
• 3.0 m — 0% survives
• 4.0 m — 0%

So it can be done, at three metres. And here is the bill for those three metres. Centre illuminance falls as one over distance squared:

• 1.0 m — 1.000 relative, 1x power
• 2.0 m — 0.250, 4x power
• 3.0 m — 0.111, 9x power
• 4.0 m — 0.062, 16x power

Nine times the power to light the same wall, or the same power and a much higher ISO — and noise is its own problem for keying, as covered in our piece on compression artifacts. Either way you have spent more than a screen costs.

Softboxes and second lights: how much do they help?

Both help, and neither closes the gap at a usable distance.

A large softbox, modelled as an array of sources, at 1.5 m:

• 0.3 m softbox — 58.6% survives, 41.1% heavily
• 0.6 m softbox — 56.9%, 38.9%
• 1.2 m softbox — 50.4%, 30.3%
• 1.8 m softbox — 39.9%, 17.7%

A 1.8 m source is not a small purchase, and it still leaves two fifths of the wall behind. Two lamps spread apart do better at the same distance:

• 0.6 m apart — 54.5% survives, 35.7% heavily
• 1.2 m apart — 38.5%, 16.8%
• 1.8 m apart — 18.4%, 3.6%
• 2.4 m apart — 17.8%, 0.4%

Two widely spaced lamps at 1.5 m beat one lamp and beat a large softbox, but 17.8% of the wall still survives. And two lamps mean two shadows, which is the geometric problem from our piece on how far to stand — you have traded an evenness problem for a shadow problem. At 3.0 m every configuration we modelled reached zero, because distance is what actually flattens the distribution.

Why a real screen is not the same object

It is worth being clear that this is not a materials snobbery argument. A screen solves three things at once that a wall does not.

Its colour is manufactured to a specification. You are not hoping the light and the paint happen to produce the right chromaticity — the surface already is that colour, and you are lighting it with white light, which is the thing lamps are good at.

It is matte and uniform, so the reflectance does not vary across it. A painted wall has texture, patching, roller marks and usually a slightly different colour near the ceiling.

And it can be lit independently of your subject. That is the requirement that keeps coming back in every measurement on this site: separate the two, light them separately, and each one gets the exposure it needs.

If you have to shoot against a wall anyway

Sometimes there is no alternative. In that case:

Do not rely on chroma keying. Given that the colour is unlikely to be right, this is the case for a subject-separation approach rather than a colour key — the boundary is covered in our piece on where AI keying wins and where it fails.

If you must light it, light from far away and light it separately. Three metres is where the geometry starts working, and you will need the power for it.

Get the subject away from the wall. Distance is the cheapest spill control there is, and a wall you are standing half a metre from will paint you green regardless of how evenly it is lit.

Check before you shoot, not after. Point the camera at the empty lit wall, run it through your keyer, and look at what survives. If a fifth of it is still there, no setting will fix it.

Where these numbers stop

This is a photometric model integrated over a surface, not a measurement from a real room.

The wall is a perfect Lambertian neutral white. Real walls are off-white, often warm, and their reflectance varies. Every one of those makes the chroma problem worse, not better, so treat the chroma figures as a best case.

The green light sources are representative sRGB values we chose, not measured spectra from real fixtures. Real green LEDs, gels and filters vary, and some will land closer to key green than 0.396. The point stands regardless: you have to check the actual distance, and it has to be under 0.161.

Sources are modelled as points and point arrays. Real softboxes have their own falloff and hot-spot behaviour that an array only approximates, and bounce from the rest of the room is not modelled at all — which in practice usually makes the distribution worse.

And the exposure policy is the most favourable one available, with the brightest point landing exactly on key green. Any other choice puts more of the wall outside the window, not less.

Related questions

Open the green screen remover