How to key out a background when your subject is wearing the same colour

Last updated 2026-09-10 · 8 min read

What is the short version?

A key removes a colour, and it cannot tell your subject's green shirt from the green wall behind them. Tightening the key everywhere fixes the shirt and eats the hair. Keying only the area that has the problem fixes one without costing the other.

1. Go to Vizard Agent with the footage.
2. Say which part of the subject is disappearing.
3. Say the hair and the edges have to survive.

What do you need before you start?

The footage and a description of where it goes wrong. Naming the body part — the elbow, the lower arm, the side of the torso — is far more useful than "the edges look bad", because a localised fault needs a localised treatment.

• The footage with its background.
• What is leaking. A garment, an arm, a shoulder.
• What must be preserved. Hair, first of all.
• What goes behind. A plate, a banner, a scene.
• Whether the subject can be scaled. It sometimes helps.

What do you type into Vizard Agent?

Say where, not just what. "The key is not clean" invites a global adjustment, which is exactly the move that trades the hair for the shirt — naming the region tells Vizard Agent that the treatment should be local.

Prompt

Variants worth knowing:

• "Just that area." Local rather than global.
• "The hair has to stay." What tightening costs.
• "No visible straight edge." The usual sign of a box mask.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real composite where a child's shirt matched the background. Notice how much of it is looking at the matte on its own rather than at the picture it produces.

1. Removes the original background so a new one can go behind.
2. Checks the cutout's format and its hair and body edges.
3. Confirms the file actually carries a transparency channel.
4. Rebuilds the cutout with object segmentation to keep hair and shoulders.
5. Verifies the alpha channel and the edge quality before compositing.
6. Takes the matte from the segmentation and applies it as the alpha directly.
7. Builds the composite with the new background behind the subject.
8. Reviews the result for position and safe space.
9. Checks the matte filter's parameters for shrinking the edge safely.
10. Looks at the matte itself to see the leak's pattern and how far to shrink.
11. Shrinks the edge in two stages rather than one, and rechecks.
12. Isolates the area the review flagged — the arm and the side.
13. Chooses a key safe for that region given the clothing and the skin.
14. Composites with a localised mask rather than a hard box.
15. Checks there is no horizontal cut line where the local fix ends.

Step ten is the step that turns guesswork into a decision. The composite tells you something is wrong; the matte on its own tells you whether the leak is a thin halo, a soft area or a hole, and each of those needs a different amount of tightening.

Step fifteen is the one people forget. A local fix applied inside a rectangle leaves a straight edge across the subject where the treatment stops, which is more obvious than the leak it was fixing.

What does the result look like?

A clean cutout over the new background: the arm and the shirt solid all the way through, the hair still fine at the edges, no green halo anywhere, and no visible boundary where the local treatment Vizard Agent applied begins or ends.

Both the fine detail and the difficult region survive, which is the thing a global setting cannot do.

When does this not work well?

Some overlaps cannot be separated. If the subject's clothing is the exact background colour across a large area, or the lighting spills that colour onto their skin, then no key distinguishes them and the honest answer is to reshoot or to replace the garment.

• Large areas of matching colour. Nothing to separate.
• Colour spill onto skin. The key takes the skin too.
• Soft or blurred edges. Shrinking removes them entirely.
• Fine hair against a busy plate. The edge shows either way.
• Low-resolution footage. The matte has too few pixels.

How do you fix a result that came back wrong?

Name the region that is still wrong and describe what you can see there. Vizard Agent keeps the matte it built and each of the shrink stages it applied, so a fix adjusts that one region rather than re-keying the whole subject from the beginning.

• "His arm is still see-through." The local key strengthened there.
• "The hair looks chewed." The shrink reduced, applied in stages.
• "There is a straight line across him." The local mask feathered.
• "There is a green rim." Spill suppression on the edge only.

How does Vizard Agent compare to doing it yourself?

By hand this is one key with a tolerance slider, adjusted until the worst problem goes away. Tightening removes the shirt problem and the hair together, and the compromise is usually settled by whichever fault the editor noticed last.

Common questions

Why does my shirt disappear? Because it is the colour the key removes. Vizard Agent treats that area separately.

Will the hair survive? That is what the staged shrink protects. Vizard Agent checks it.

Can it key just one part of the frame? Yes, and Vizard Agent feathers the boundary so it does not show.

What if the colour is on the skin? Spill suppression helps, though Vizard Agent cannot recover large amounts.

Can I see the matte? Yes, and it is the fastest way to understand the fault.

Does it work without a proper screen? Often. Vizard Agent can segment the subject rather than key a colour.

Will the edges be soft or hard? Whatever suits the plate. Vizard Agent shows you both.

Can it scale the subject? Yes, and Vizard Agent finds slightly smaller sometimes hides a difficult edge.

What about motion blur on the edges? Vizard Agent keeps it rather than shrinking it away.

Can it fix one shot only? Yes. Name the shot and the region.

Will the background stay sharp? Yes. Vizard Agent composites the subject over it untouched.

What if the file has no alpha? Vizard Agent checks that first and rebuilds it if needed.

Can it handle several subjects? Yes. Vizard Agent builds a matte for each of them.

Why look at the matte rather than the picture? Because the picture tells you something is wrong and the matte tells you what, and those are different questions with different fixes.